Structured prompts and workflows that make AI do real SEO work

ROLE & STANDARD
You build a 90-day plan to get a brand cited and ranked in AI answers. Sequence it against how AI visibility actually compounds, because spending out of order wastes the quarter:
  - It is a stack: a page must be crawlable and renderable, then answer-first, then tied to a clear entity, then corroborated by third parties, before citations follow. Earning links to a page bots cannot read, or chasing GEO before the page answers the question, buys nothing.
  - Foundations first: crawl access, the key money pages, and entity consistency come before off-site outreach.
  - Every initiative needs a leading metric you can move this month and a lagging metric it should eventually shift, or you cannot tell whether it worked.
INPUTS
Business [business], category [category], current state [paste what you know: rankings, citation checks, site health], goal [goal], resources [team / budget / hours per week]. If current state or resources are thin, ask me up to 3 questions before planning, then proceed.
TASK
Produce three 30-day phases. For each: the objective, 3 to 5 initiatives, the owner type, the one leading metric to watch this phase, and the lagging metric it targets. Order so foundation work precedes off-site work, and make week one address my single biggest constraint. For each initiative, state in one line why it is sequenced where it is.
OUTPUT
A phased table, then the single week-one action and its rationale. Invent no benchmarks or metric values; mark inferences [assumed] and figures I must supply [NEEDS INPUT].

ROLE & STANDARD
You audit whether a set of pages can be cited and ranked by AI engines. Judge against the visibility stack and report the lowest broken layer first, because a fix high in the stack does nothing if a lower layer is broken:
  - Crawlable and renderable (in raw HTML, not JS-only) -> answer-first structure -> entity clarity -> schema -> off-site corroboration -> measurement. A page bots cannot read gains nothing from better headings.
  - You can only assess what you can see. From pasted text you can judge structure, entity clarity and schema; you cannot judge live crawl behaviour or off-site corroboration, so mark those not assessable rather than scoring them.
INPUTS
[Paste each page's visible text; include raw <head> HTML, robots.txt and any JSON-LD if you have them.] Category [category], the buyer questions these pages should win [list]. If you were given only URLs you cannot open here, stop and ask me to paste the content, because a guessed audit is worse than none.
TASK
Score each lens 0 to 100 with a one-line reason: crawlability, answer-first, entity clarity, schema, off-site corroboration, measurement. Mark each score [observed] (judged from what I pasted) or not assessable (needs data you cannot see). Name the lowest broken layer and why it gates the rest. Give the top fix per lens with the acceptance test that marks it done.
OUTPUT
A score table with the observed / not-assessable column, the ranked fix list (impact x effort), the three to do first, and a one-line self-check naming your weakest assumption and how to verify it.

ROLE & STANDARD
You triage a content portfolio into keep / improve / merge / cut. Make the decision rule explicit, because an unstated threshold makes the call arbitrary:
  - State the cut line up front (for example: no traffic, no conversions, no citation potential, and no link equity worth preserving) and apply it consistently.
  - Two pages targeting the same intent split signals and cannibalize; name one canonical winner and a fate for the rest.
  - Improve only where there is a realistic path to traffic, conversions or a citation the page is currently missing.
INPUTS
Content inventory [paste titles, urls, and any metrics]. Goal for this portfolio [traffic / conversions / AI citations / authority]. If the goal or metrics are missing, ask before classifying, since the cut line depends on them.
TASK
Classify each URL with the reason and the specific AEO/GEO upside, against the stated cut line. Flag duplicate-intent clusters and name the canonical page. For each improve, give the one change that unlocks the upside.
OUTPUT
A decision table sorted by priority, then the top 5 consolidation moves. Judge only from the inventory I gave; mark inferences [assumed] and metrics I must supply [NEEDS INPUT].

ROLE & STANDARD
You map the clusters a brand must build to be treated as an authority on a topic. Prioritise by citation utility, because not all coverage gaps are equal:
  - Engines reward demonstrated breadth and depth on a topic; a thinly-covered topic rarely gets cited as the authority.
  - Rank missing clusters by how often an engine would need that subtopic to answer a real buyer question, not by search volume alone.
  - Judge existing coverage only from what I paste; do not assume coverage you cannot see.
INPUTS
Core topic [topic]; existing coverage [paste titles/urls]; competitors [list]; my product and audience [so the map is mine, not generic]. If coverage or competitors are blank, ask up to 3 questions before mapping, so the gaps are real.
TASK
Map the topic into pillars and subtopics. Mark each well-covered / weak / absent from what I pasted. For each missing or weak cluster, give the buyer question it answers, a HIGH/MED/LOW citation-utility rating with the reason, and the asset to build. Mark each rating [assumed] where it is your inference.
OUTPUT
A pillar/subtopic tree with coverage status and a prioritised build order. Invent no metrics; mark inferences [assumed].

ROLE: You allocate effort between classic SEO and AEO/GEO based on where my buyers actually get answers (a direct AI answer vs a page of links) and where my revenue comes from.
CONTEXT: Resources [hours or budget], current rankings [paste], current AI-citation status [paste], how I make money [model].
TASK: Recommend a percentage split for the quarter. Justify each side from my buyer behaviour and revenue, not generic best practice, and name the one early signal that would prove the split wrong.
OUTPUT: The split, three reasons tied to my context, and the signal to watch. If a field is blank, state the assumption you made.

ROLE & STANDARD
You set quarterly objectives for AI search. Make every key result measurable and non-vanity, because an OKR you cannot measure is a wish:
  - Each KR names its data source and a real baseline; if the baseline is unknown, the first KR is to instrument it.
  - Prefer outcome metrics that map to value: citation rate, share of voice vs named rivals, AI-referral revenue, indexed-but-not-cited count. Avoid impressions and raw rankings as headline KRs.
INPUTS
Company stage [stage], baseline metrics [paste]. If the baseline is thin, ask up to 3 questions first so the KRs anchor to real numbers, then proceed.
TASK
Draft 3 objectives, each with 3 measurable key results. For each KR give the data source and the target, and mark whether the baseline is known or must be measured first.
OUTPUT
Objectives and key results in a table. Invent no baseline figures; mark inferences [assumed] and numbers I must supply [NEEDS INPUT].

ROLE & STANDARD
You protect AI visibility through a site change. Work from the specific things a migration breaks, because "test everything" is not a plan:
  - The usual failure points: lost or chained redirects, dropped schema, removed answer capsules, broken internal links, changed crawl access (robots.txt / meta robots), and JS-rendering regressions that hide content from bots.
  - Each check needs a before-launch baseline, an after-launch verification, and a rollback trigger that says when to revert.
INPUTS
Planned change [redesign / replatform / url change], affected pages [list]. If the change or page list is vague, ask up to 3 questions before building the checklist, then proceed.
TASK
For each failure point: what to capture before launch, what to verify after, and the rollback trigger. Call out anything irreversible (for example url changes with no redirects) to lock before launch.
OUTPUT
A checklist grouped before-launch / after-launch with a rollback trigger per item. Invent no specifics about my stack; mark inferences [assumed] and details I must supply [NEEDS INPUT].

ROLE & STANDARD
You map the keywords and questions a brand must cover to win search and AI answers. Build around how engines actually pick content, not a flat keyword dump:
  - Start from jobs-to-be-done; engines answer the buyer's underlying task, then fan it into sub-questions (compare, alternatives, is-it-worth-it, how-much, for-use-case). Those sub-questions are the real targets.
  - Tag each with the format currently winning it (definition / list / comparison / how-to); a page in the wrong format will not rank however good it is.
  - You cannot see real search volumes here; never invent them. Rate priority by buyer intent and citation utility instead, as a hypothesis to validate in a tool.
INPUTS
Product [product], audience [audience], competitors [list], goal [goal]. If product or audience is thin, ask up to 3 questions first so queries are specific to my buyer, then proceed.
TASK
Identify 6 to 8 jobs-to-be-done. For each, list head terms, long-tails, and the conversational questions a buyer asks an AI (recommendation, comparison, alternatives, pricing, how-to). Tag funnel stage and the likely winning format. Flag the 10 highest-leverage rows with the reason each is high-leverage.
OUTPUT
A table (job, query, type, stage, likely format, priority, angle). Invent no volumes; mark inferences [assumed].

ROLE: You find the real questions people ask an AI assistant about a topic, and you are honest about what is a guess.
CONTEXT: Topic [topic], audience [audience], my product or page [optional, so questions are not generic].
TASK: List 40 genuine questions grouped beginner/intermediate/advanced, labelled by intent. For each, mark whether it likely triggers a direct AI answer (definitional, how-to) or a cited shortlist (best-X-for-Y, comparison), naming the signal, and treat that as a hypothesis to validate by running it. Then cluster the 12 strongest into pages.
OUTPUT: The labelled table, then the 12 clustered with page titles. Invent no search volumes; if a field is blank, state the assumption you made.

ROLE: You expand a seed into long-tail variations, grouped so buyer intent is visible.
CONTEXT: Seed [seed], audience [audience], region [region].
TASK: Generate 100 variations grouped by modifier type (use-case, audience, comparison, problem, location, price), since the modifier signals intent. Mark the 15 with the clearest commercial or buyer intent and say what signals it.
OUTPUT: Grouped lists, then the 15 priorities. Invent no volumes; if a field is blank, state the assumption you made.

ROLE & STANDARD
You find the questions keyword tools miss but buyers ask an AI directly. Justify why each matters despite low reported volume:
  - Conversational, specific, multi-part questions often show no volume in tools yet are exactly what a buyer types into an assistant near a decision.
  - Each must map to a page that could realistically win the answer; a question with no page to own it is noise.
INPUTS
Topic [topic], buyer [persona], my product [so questions are mine]. If the persona is thin, ask up to 3 questions first so the questions match a real buyer, then proceed.
TASK
List 25 specific, conversational questions a buyer asks an AI that likely show low or no search volume. For each, give why it matters near a decision and the page that should own it.
OUTPUT
A table (question, why it matters, owning page). These are candidates to validate, not measured demand; mark inferences [assumed].

ROLE: You classify keyword intent and catch page-type mismatches, since a page in the wrong format for the intent will not rank.
CONTEXT: Keyword list [paste].
TASK: For each, assign dominant intent (informational, commercial, transactional, navigational), the format likely to rank, and whether an answer-first page fits. Flag every case where my planned page type fights the intent.
OUTPUT: A labelled table plus the flagged mismatches. If the list is empty, ask me to paste it rather than inventing keywords.

ROLE & STANDARD
You find the buyer questions where an AI engine names a competitor and not me. Be rigorous about what is observed vs guessed, because you have not run these queries:
  - Treat every claim about who is currently cited as a hypothesis marked [assumed]; the only way to confirm is to actually ask the engine.
  - For each gap, the useful output is the specific asset or edit that would earn the slot, and its effort, not just the observation.
INPUTS
My site [url], competitors [list], category [category]. If competitors or category are blank, ask before analysing.
TASK
Infer 15 buyer questions where an engine likely names a brand. For each, hypothesize the cited competitor and the source type behind it (review site, roundup, community, docs), the gap on my side, the asset to create or earn, and the effort. Mark every who-is-cited claim [assumed].
OUTPUT
A table (question, likely-cited rival [assumed], source type, my gap, asset, effort), then the 3 to verify first by running the query.

ROLE: You map timely content angles to a calendar, with enough lead time to publish and get indexed before each peak.
CONTEXT: Niche [niche], region [region].
TASK: List the recurring seasonal moments and likely trend windows in my niche. For each, give a content angle and a publish-by date that accounts for indexing lead time. Treat windows as estimates, not forecasts.
OUTPUT: A calendar table (period, angle, publish-by). Mark inferences [assumed]; if a field is blank, state the assumption you made.

ROLE & STANDARD
You design a topic cluster that signals topical authority. Build it so the structure itself is the signal:
  - A pillar plus tightly-linked subtopics tells an engine you cover the topic in depth; orphaned or unlinked pages do not contribute to that signal.
  - Each subtopic should own a distinct buyer question, with no two pages competing for the same intent.
INPUTS
Pillar topic [topic], my audience and product [so the cluster is mine]. If the topic is broad or the audience is blank, ask up to 3 questions first to scope it, then proceed.
TASK
Break the pillar into 8 to 12 subtopics. Define the internal-link relationships (pillar to subtopic and useful subtopic to subtopic), and name the single buyer question each page wins. Flag any two subtopics at risk of cannibalizing.
OUTPUT
A cluster map with link directions and the buyer question per page.

ROLE & STANDARD
You produce a content brief for one page whose job is to be cited by AI answer engines (ChatGPT, Perplexity, Google's AI answers) and to rank. Hold the brief to this standard, because it is what separates a citable page from a generic one:
  - Liftability: an engine answers by lifting a self-contained passage into its reply, so every key claim must be a complete sentence that still makes sense quoted alone: no "as noted above", no opening pronoun, no dependence on the previous paragraph.
  - Answer-first: the direct answer to the page's core question sits in the first 80 words, immediately under a question-shaped heading, so the heading-plus-answer pair is what gets extracted.
  - Corroborability: engines trust claims other sources echo, so name the specific entities, standards, products and figures an engine can match against other pages. Vague claims have nothing to corroborate.
  - A reason to be cited over the incumbent: the page needs one thing the currently cited sources lack (original data, a sharper definition, a fuller comparison, a more recent number). Find it.
INPUTS
Core question / keyword [keyword], audience [audience], the action the page should drive [action]; if you have them, the titles and H2s of the pages currently ranking or cited [paste], and any first-party data or product facts I can contribute [paste]. If audience, action, or the competing pages are missing, ask me up to 3 questions first, since the angle depends on them. If I cannot give competing pages, say you are working without sight of the live results and reason from the query shape instead.
TASK
1. Intent and format: state the dominant intent and the format currently winning this query (definition / listicle / comparison / how-to); if inferred without the SERP, mark it [assumed].
2. Differentiator: name the one angle, data point or depth this page can offer that the cited sources do not. This is the reason an engine would switch to us.
3. Answer capsule: write the 40 to 80 word capsule for the top of the page; it must answer the core question in its first sentence, read well quoted verbatim, and name at least one specific entity or figure.
4. Outline: a question-shaped H2/H3 structure mirroring how people ask this of an AI; under each heading, write the single self-contained takeaway sentence that section must land.
5. Entities and evidence: list the named entities and the specific stats or claims the page needs to be corroborable; flag which I must supply [NEEDS INPUT] vs which are common knowledge, so you invent nothing.
6. FAQs: 6 questions matching the follow-ups an engine fans out to from this query (not restatements of the body), each with a 2 to 3 sentence standalone answer.
7. Internal links: 5, each with the anchor and why it strengthens the topical cluster.
OUTPUT
The seven labelled sections, no padding. Invent no statistics, sources or product facts: mark inferences [assumed] and anything I must provide [NEEDS INPUT].

ROLE & STANDARD
You reverse-engineer a winning outline from what currently ranks and gets cited, then beat it. Work only from evidence:
  - Match the format the top results share, because an engine has already decided what shape wins this query; then win on the gaps they all miss.
  - You cannot see the live SERP or AI answer here, so reason only from what I paste and say so if I give nothing.
INPUTS
Keyword [keyword]; top results [paste titles and H2s]; current AI answer [paste if any]. If I did not paste the top results, ask for them rather than guessing what ranks.
TASK
Note what the top results share (format, depth, common sections). Find 3 coverage gaps none of them fill. Produce an outline that matches the winning format plus those gaps, marking under each section where a stat, table, example or FAQ belongs and the self-contained takeaway it must land.
OUTPUT
An annotated outline. Base it only on what I pasted; invent no stats, mark inferences [assumed].

ROLE: You write FAQs an engine can lift as snippets, each answering a real follow-up question rather than restating the body.
CONTEXT: Topic [topic], audience [audience].
TASK: Generate 8 questions phrased the way people ask them, each with a 2 to 3 sentence answer that stands alone with no surrounding context. Skip anything the main body already answers verbatim.
OUTPUT: 8 Q&A pairs. If a field is blank, state the assumption you made.

ROLE & STANDARD
You outline a comparison page AI engines will cite. The thing that earns the citation is credibility:
  - Engines and buyers trust comparisons that are fair, and admitting where your own product is weaker is what makes the rest believable, so build that in deliberately.
  - Compare on the criteria buyers actually weigh, not a feature dump, and end on a clear best-for-whom verdict.
  - You can only compare verifiable facts; do not assert a difference you cannot confirm.
INPUTS
[brand] vs [competitor], audience [audience], verifiable facts for each [paste]. If you lack real facts for either product, ask for them before outlining rather than inventing differences.
TASK
Outline the page: the decision criteria, an honest pros/cons per product (including where mine loses), a decision table with the columns to fill, and the best-for-whom verdict logic. Mark where transparency about a weakness should appear.
OUTPUT
A full outline with the table columns and verdict logic. Do not assert an unverifiable weakness; mark inferences [assumed] and facts I must supply [NEEDS INPUT].

ROLE: You define a repeatable roundup-entry template so every entry is consistent, scannable and individually liftable.
CONTEXT: Roundup topic [topic], example tool to demo [example tool].
TASK: Define the per-entry fields (one-line positioning, where it shines, watch-outs, best for, price), then write one worked entry for [example tool] so a writer can match the pattern.
OUTPUT: The template, then the worked example. Do not invent pricing or features for the example; if a field is blank, state the assumption you made.

ROLE & STANDARD
You plan a surgical content refresh, not a rewrite. Two things matter:
  - Regain rankings and citations by fixing what is outdated and what is missing versus the pages currently winning; a refresh that only changes the date changes nothing.
  - Preserve existing equity: name what to keep so the update does not throw away rankings the page already holds.
INPUTS
Page [url or paste], published [date], target [keyword], current top results if you have them [paste]. If I gave only a URL you cannot fetch, ask me to paste the page, since a plan from memory will miss what is actually there.
TASK
List what is outdated, what is missing versus current winners, and the minimum edits to regain position and citation. Separately list what to keep and why.
OUTPUT
Keep / update / add lists, then the publish-ready change summary. Judge only from what I pasted; mark inferences [assumed].

ROLE & STANDARD
You specify the experience and trust signals a trust-sensitive page must show. Make them visible and real:
  - E-E-A-T is shown, not claimed: first-hand evidence, a credentialed named author, citations, dates and methodology, each placed where a reader and a crawler can see it.
  - Signals must be genuine; fabricated credentials are a trust and legal risk.
INPUTS
Page topic [topic], author [name and real credentials], first-hand evidence I can show [describe]. If the author or credentials are blank, ask before building the checklist, since the trust signals depend on who stands behind the page.
TASK
List the experience, expertise and trust signals this page should show and exactly where each goes (author box, byline, inline evidence, citations, methodology note, dates). Mark any signal I must supply.
OUTPUT
A checklist mapped to page sections. Invent no credentials; mark items I must provide [NEEDS INPUT].

ROLE: You write a section an engine can lift, by leading with a self-contained 40 to 60 word answer that reads correctly quoted alone.
CONTEXT: Section question [question], facts [facts], tone [tone].
TASK: Open with the direct answer (no opening pronoun, no "as above"), then support it with detail, a short list, and one concrete example. Use only my facts.
OUTPUT: One question-shaped H2 and the section, no preamble. If the facts are blank, ask me for them rather than inventing any.

ROLE: You are a ruthless editor who cuts filler while preserving every fact, since engines lift specific sentences and skip padding.
CONTEXT: Draft [paste].
TASK: Remove hedging, throat-clearing and empty transitions, and tighten sentences so each carries information and stands on its own. Keep every fact and key takeaway. Flag any sentence that makes a claim with no support.
OUTPUT: The tightened draft, then the list of unsupported claims to source. If the draft is blank, ask me to paste it rather than writing one.

ROLE: You write title tags and meta descriptions that include the keyword naturally and earn the click without overpromising.
CONTEXT: Page [topic], target keyword [keyword], brand [brand].
TASK: Write 5 title tags (about 55 to 60 chars) and matching descriptions (about 150 to 160 chars). Keep the keyword natural, avoid clickbait the page cannot deliver, and recommend the best pair.
OUTPUT: 5 options in a table, then your pick and why. If a field is blank, state the assumption you made.

ROLE: You make robotic copy read as human without losing accuracy, mainly by varying sentence length and replacing generic phrasing with specifics.
CONTEXT: Draft [paste], brand voice [describe].
TASK: Rewrite to mix short punchy sentences with longer ones, cut stock AI transitions and filler, and add concrete detail, while keeping every fact. Do not add claims to sound impressive, and do not just reword the first line and leave the rest.
OUTPUT: The rewrite, then 3 phrases you removed and why. If the draft is blank, ask me to paste it rather than writing one.

ROLE: You write intros that put the payoff in the first sentence, since both readers and engines reward an answer-first opening.
CONTEXT: Article target [keyword], reader goal [goal].
TASK: Write a 3 to 4 sentence intro that states the payoff first, names who it is for, and avoids any "in today's world" opener.
OUTPUT: The intro only. If a field is blank, state the assumption you made.

ROLE & STANDARD
You enforce sourcing discipline on a draft. The rule is simple and strict:
  - Every factual or numeric claim needs a named, dated, credible source; a claim that cannot be sourced is softened or cut, not left to stand.
  - You do not invent sources or supply specific ones you are not certain exist.
INPUTS
Draft [paste]. If the draft is blank, ask me to paste it rather than inventing claims to source.
TASK
Extract every factual or numeric claim. For each, say whether it needs a source, the type of source that would be credible, and the action (source it, soften it, or cut it).
OUTPUT
A claims table (claim, needs source?, credible source type, action). Supply no specific sources you cannot verify; mark anything uncertain [assumed].

ROLE: You write CTAs matched to the reader's awareness stage, since a hard ask to a cold reader and a soft ask to a ready one both convert worse.
CONTEXT: Page intent [intent], offer [offer].
TASK: Write 3 CTA options (soft, mid, hard) with microcopy, and say which stage each fits and when to use it.
OUTPUT: 3 CTAs with usage notes. If a field is blank, state the assumption you made.

ROLE & STANDARD
You audit a page for extractability: how easily an AI engine can lift a passage from it as an answer. Judge against the tells that actually decide this:
  - The biggest single loss is opening with background or history instead of answering the page's core question in the first 80 words; check that first.
  - The extraction unit is a question-shaped heading followed by a self-contained 40 to 80 word answer; passages that depend on earlier text cannot be lifted.
  - Entities and figures must be explicit, because an engine cannot corroborate "studies show"; schema and a visible date round it out.
INPUTS
[Paste the page content], target question [question]. If I gave only a URL you cannot fetch, stop and ask me to paste the content, since a guessed audit is worse than none.
TASK
Score 0 to 100 on answer-first structure, question headings, extractable chunks, entity clarity, schema readiness and freshness, each with a one-line reason. For each weak area give the specific change with an example rewrite. Identify the passage most likely quoted and the one most likely skipped, and why. Rank fixes by impact and effort.
OUTPUT
Score table, prioritized fixes, quotable/skippable passages. Score only what is in what I pasted; mark inferences [assumed]. Finish with a one-line self-check naming your weakest assumption and how to verify it.

ROLE: You write the short, liftable answers AI quotes, each answering the question in its first sentence and standing alone with no surrounding text.
CONTEXT: Question [question], facts [facts], brand [brand].
TASK: Write three 40 to 80 word capsules that lead with the answer, name a specific fact, and vary the angle. Recommend one based on how the question is likely asked.
OUTPUT: Three labelled options and your pick. Use only my facts; if the facts are blank, ask me for them rather than inventing any.

ROLE: You convert headings into the questions people ask an AI, since a question-shaped heading is what an engine matches and extracts under.
CONTEXT: Headings [paste], primary keyword [keyword].
TASK: Rewrite each heading as a natural-language question without losing the keyword, and flag any section with no clear question to answer, since that signals it is not extractable.
OUTPUT: Old vs new headings, then the flagged sections. If the headings are blank, ask me to paste them rather than inventing a structure.

ROLE & STANDARD
You restructure prose into self-contained chunks an engine can lift. The standard:
  - Each key point becomes a stand-alone block (claim, support, example) that makes sense with no surrounding text; nothing should depend on "as above" or a previous paragraph.
  - You restructure only; you add no new facts.
INPUTS
Content [paste]. If the content is blank, ask me to paste it rather than inventing it.
TASK
Reshape it so each point is liftable on its own. Add short subheads and lists where they help extraction. For each change, note in one line why it improves liftability.
OUTPUT
The restructured content with the per-change notes. Add no new facts; mark anything you inferred [assumed].

ROLE: You write definitions engines cite: one tight sentence that stands alone, then a short expansion.
CONTEXT: Term [term], context [context].
TASK: Define the term in one sentence under 30 words that reads correctly quoted alone, then a 3-bullet expansion, then a one-line common misconception to correct.
OUTPUT: Definition, bullets, misconception. Keep it accurate; if a field is blank, state the assumption you made.

ROLE & STANDARD
You model how an engine fans a query into sub-questions, then check coverage. The point:
  - Engines decompose a query into sub-questions (compare, is-it-worth-it, how-much, alternatives, for-use-case, pros-cons) and assemble an answer from whoever covers each; a page that misses a sub-question cannot be cited for it.
  - Judge coverage only from the page text I give you.
INPUTS
Main query [query], page [paste or url]. If I gave only a URL you cannot fetch, ask me to paste the page before judging coverage.
TASK
List the sub-questions an engine would decompose this into. Mark which my page answers based on what I pasted, and for each gap recommend where to add it.
OUTPUT
A coverage table (sub-question, covered?, where to add). Mark coverage you inferred without the page text as [assumed].

ROLE: You target People-Also-Ask and AI follow-up questions, ordering by how commonly each is asked.
CONTEXT: Topic [topic].
TASK: Produce 10 questions phrased as users ask them, ordered most to least common, each with a concise standalone answer. Note the two most likely to appear as an AI follow-up question.
OUTPUT: 10 Q&A with the two flagged. If the topic is blank, ask me for it rather than guessing.

ROLE: You make a page read as genuinely current, since engines favour maintained content, without faking an update.
CONTEXT: Page [paste or url], last updated [date].
TASK: List the specific freshness signals to add or update (visible date, refreshed stats, current examples, removed stale references). Do not just change the date without changing content, and invent no new stats.
OUTPUT: A change list. If I gave only a URL you cannot fetch, ask me to paste the page rather than guessing what is stale.

ROLE: You are a GEO analyst.
CONTEXT: I will paste an AI answer that cites a competitor for [query]; my page is [url].
TASK: Work in four passes and show each. 1) Deconstruct every cited source, classify it, and infer why the engine trusted it (recency, specificity, corroboration, schema, consensus), marking each [observed] or [assumed]. 2) Diagnose why I was not cited, most-likely first. 3) Plan the takeover: per gap, the exact asset or edit, effort S/M/L, and on-site vs off-site. 4) Self-check the two weakest assumptions and how to verify.
EXAMPLE ROW: reddit thread -> [assumed] community consensus + recency -> gap: no authentic presence -> asset: answer 3 real threads usefully (M, off-site).
OUTPUT: A table for passes 1 to 3, then pass 4 notes. If the answer has no citations, say so and ask for them.

ROLE & STANDARD
You plan the off-site sources that will get a brand cited. GEO is about earned trust, so:
  - Engines cite sources other credible sources corroborate (review sites, roundups, communities, docs, reputable publications); the plan targets those, not the brand's own pages.
  - Every play must be a genuine, non-spammy way to earn the mention; anything manipulative is off the table.
INPUTS
Brand [brand], category [category], target questions [list], competitors [list]. If target questions or competitors are blank, ask before planning, so the source list is real and not generic.
TASK
For each question, list the source types likely cited, name the specific properties worth targeting and why, give an honest way to earn each, and sequence over 90 days.
OUTPUT
A table (question, source type, property, how to earn, effort) and a 90-day sequence. Nothing manipulative; mark source guesses [assumed].

ROLE & STANDARD
You design genuine community presence that can get a brand cited, without breaking platform rules:
  - Engines cite community threads (Reddit, forums) where there is real consensus, so presence has to be value-first and rule-respecting or it backfires.
  - A mention is only useful if it is accurate and in context; never astroturf.
INPUTS
Brand [brand], topic [topic], audience [audience]. If audience or topic is thin, ask up to 3 questions first so the communities are the right ones, then proceed.
TASK
Identify where my buyers actually discuss this. For each community, give a value-first way to participate, the rule to respect, and what an accurate, in-context mention of my brand looks like.
OUTPUT
A table (community, value-first angle, rule, accurate-mention example). Nothing that violates platform rules; mark communities you are unsure exist [assumed].

ROLE & STANDARD
You pitch a brand into the comparison roundups engines cite. The pitch lands on substance:
  - Editors add products that genuinely fit a slot and add reader value; a vague or inflated claim gets ignored, so lead with one verifiable differentiator.
  - Offer facts the author can check, not adjectives.
INPUTS
Brand [brand], the roundup [url or title], my genuine differentiator [differentiator]. If the differentiator is blank, ask for it rather than inventing one, since a fake claim gets the pitch ignored.
TASK
Draft a concise, honest outreach pitch: why my product fits a slot, the unique value for readers, and the verifiable facts offered. Include a subject line and a 90-word body.
OUTPUT
Subject and email body. Use only facts I gave; invent no features or numbers.

ROLE & STANDARD
You map the narrative an engine repeats about a category, so a brand can shape it. Be strict about evidence:
  - Engines settle into recurring claims and framings seeded by a few sources; the value is identifying those and where my brand is framed weakly or wrongly.
  - You have not run the queries, so every source attribution is a hypothesis marked [assumed]; confirming each means actually asking the engine.
INPUTS
Category [category], brand [brand], competitors [list]. If competitors are blank, ask before mapping.
TASK
Identify the recurring claims and framings AI answers use, the sources that likely seed each, and where my brand is positioned weakly or inaccurately, with the counter-evidence to earn.
OUTPUT
A table (claim/framing, likely source [assumed], my position, counter-evidence to earn). Invent no sources.

ROLE & STANDARD
You plan to correct an inaccurate thing AI says about a brand. Correction works by changing the sources, not arguing:
  - Engines repeat what trusted sources say, so the fix is an authoritative on-site page plus updated or newly-earned third-party sources carrying the accurate phrasing.
  - You state only verifiable truths I give you; you do not counter one false claim with another.
INPUTS
The wrong/negative claim [paste], the truth [verifiable facts], where the engine likely got it [guess]. If the truth is blank, ask for the verifiable facts rather than assuming what is correct.
TASK
Plan the correction: the authoritative page to publish, the third-party sources to update or earn, and the exact accurate phrasing to propagate. Order by leverage.
OUTPUT
A prioritised correction plan. State only verifiable truths I gave; mark inferences [assumed].

ROLE & STANDARD
You build the corroboration engines reward: turning a brand claim into cross-source consensus. The mechanism:
  - One claim on your own site is weak; the same claim echoed by original data, a case study, a third-party review and an expert quote becomes consensus an engine will repeat.
  - Plans must be grounded in proof I actually hold; do not invent evidence.
INPUTS
Key claim about my brand [claim], proof I hold [evidence]. If the proof is blank, ask what evidence I actually have rather than assuming I can back the claim.
TASK
List the formats that would turn this claim into consensus and a realistic way to produce or earn each, with effort and the expected trust lift.
OUTPUT
A table (format, how to produce/earn, effort, expected lift). Invent no proof I do not have; mark expected lift [assumed].

ROLE: You plan video that supports AI citations, since engines often cite YouTube; each video needs an on-screen takeaway an engine can lift.
CONTEXT: Topic [topic], brand [brand].
TASK: List 8 short video topics that answer buyer questions, each with the title, the spoken hook, the liftable on-screen takeaway, and the description text to include.
OUTPUT: A table (topic, title, hook, takeaway, description note). If a field is blank, state the assumption you made.

ROLE & STANDARD
You assess a brand's readiness for open knowledge bases honestly. Notability is the gate:
  - Wikipedia and Wikidata require independent, significant coverage; without it an entry will not stand, and self-serving or paid entries breach guidelines.
  - If notability is borderline, say so plainly rather than encouraging a doomed or non-compliant entry.
INPUTS
Brand [brand], notable facts and independent coverage [paste]. If the coverage is blank, ask what independent coverage exists rather than assuming notability.
TASK
Assess whether the brand meets notability, what independent coverage is missing, and the ethical steps to build a defensible entity over time.
OUTPUT
A readiness verdict, the gaps, and ethical next steps. Do not advise self-serving entries that breach guidelines.

ROLE: You help me run a clean AI-citation baseline. Act as a neutral buyer with no prior knowledge of my brand.
PROTOCOL (read first, it is the whole point): run each question in a SEPARATE fresh chat (new conversation, logged out or incognito, memory and personalization off) so earlier answers do not bias later ones, and run each 2 to 3 times, because AI answers vary between runs. If you cannot actually browse the live web here, say so and do not guess a real answer; I will run them and paste the results back.
CONTEXT: Category [category], brand [brand], competitors [list], use cases [list].
QUESTIONS (run each separately): 1) best [category] for [use case]? 2) good [category] for [audience/budget]? 3) [brand] vs [competitor]? 4) alternatives to [competitor]? 5) is [brand] any good?
RECORD per run: is [brand] cited (named with a link), mentioned (named, no link), or absent; which competitors are named; every source cited.
OUTPUT: one row per question per run, plus a consistency note. Flag any question whose result flips between runs, since those are fragile positions. Paste the stable rows into CiteTrack AI to track.

ROLE & STANDARD
You audit how AI describes a brand and catch what is wrong. Accuracy hinges on real facts and a clean read:
  - Run in a fresh, logged-out session; if you cannot browse the live web, say so rather than describing my brand from memory, which may be stale.
  - Only mark a claim wrong if it contradicts the facts I give you; the likely source is a hypothesis.
INPUTS
Brand [brand]; true current facts [facts]. If the facts are blank, ask for them before judging accuracy.
TASK
Have the engine describe my brand, pricing and comparisons, then compare line by line to my facts and flag each accurate / outdated / wrong, with the likely source and the fix.
OUTPUT
A table (AI claim, status, correct fact, likely source [assumed], fix).

ROLE & STANDARD
You build a prompt set to measure share of voice against named rivals. For the measurement to mean anything:
  - The set must span how buyers actually ask (recommendation, comparison, alternatives, use-case), not five variants of one question.
  - Answers vary, so the set is run in fresh sessions and a few times each.
INPUTS
Category [category], competitors [list]. If competitors are blank, ask before building the set, so share of voice is measured against the right rivals.
TASK
Generate 15 buyer prompts across those types, with a one-line note on running each in a fresh session a few times.
OUTPUT
15 prompts grouped by type.

ROLE & STANDARD
You test how prompt wording changes who gets cited. Read the result carefully:
  - Paraphrases of the same intent can surface different sources, but some variation between runs is just noise rather than a phrasing effect, so separate the two.
  - Each paraphrase is run in a fresh session and 2 to 3 times.
INPUTS
Core question [question], brand [brand]. If the core question is vague, ask me to sharpen it first.
TASK
Produce 5 paraphrases (different phrasing, same intent). Tell me to run each fresh and a few times, record whether cited sources change, and interpret what stability vs volatility implies, noting which differences are likely noise.
OUTPUT
The 5 paraphrases and a short how-to-read-the-results note.

ROLE: You compare how different engines answer the same question, since each weights sources differently.
CONTEXT: Question [question], brand [brand].
TASK: Have me run the same question in ChatGPT, Perplexity, Gemini and Claude, each in a fresh session and a couple of times, and give me a table to fill (engine, cited me?, top source, source type, sentiment) plus what differences between engines would mean.
OUTPUT: The fillable table and a short interpretation guide. If a field is blank, state the assumption you made.

ROLE & STANDARD
You diagnose why a brand is absent from specific AI answers. Be clear these are hypotheses:
  - Absence usually traces to one of: not crawlable, not answer-first, weak entity, no off-site corroboration, or the wrong page targeting the query. Name the most likely and the single fix that would flip it.
  - You have not verified the cause, so mark each [assumed] and say what would confirm it.
INPUTS
Queries where I am absent [list]; my relevant pages [list]. If either list is blank, ask before diagnosing.
TASK
For each query, give the most likely reason, the one highest-leverage fix, and a confidence level.
OUTPUT
A table (query, likely reason [assumed], fix, confidence, what would confirm).

ROLE: You read how positively AI frames a brand, by quoting the exact phrase that sets the tone.
CONTEXT: Brand [brand], questions [list].
TASK: For each answer, classify the framing (positive, neutral, negative, caveated) and quote the phrase that drives it.
OUTPUT: A table (question, framing, key phrase). If you cannot browse, ask me to paste the actual AI answers rather than imagining how my brand is framed.

ROLE: You assemble the buyer prompts worth monitoring continuously, balanced so the tracked set reflects real demand, not one angle.
CONTEXT: Category [category], brand [brand], competitors [list], priority use cases [list].
TASK: Produce 20 to 30 prompts balanced across recommendation, comparison, alternatives, problem-solution and pricing, with my domain and competitor domains noted.
OUTPUT: A clean list ready to paste into CiteTrack AI as tracked prompts. If competitors or use cases are blank, ask before assembling, so the list reflects what matters.

ROLE & STANDARD
You read both the Google results and the AI answer for a query to find the opening. Work from evidence:
  - The top results reveal the format an engine already rewards, and the AI answer reveals which sources it currently trusts; the opening is the gap both leave.
  - You cannot see the live SERP or AI answer here, so reason only from what I paste.
INPUTS
Query [keyword]; top results [paste]; current AI answer [paste]. If I did not paste the results, ask for them rather than guessing what ranks.
TASK
Describe what the top results share and the intent assigned; describe what the AI answer includes and cites; find 3 gaps to exploit to rank and get cited; recommend the exact format and the one differentiator.
OUTPUT
Google pattern, AI pattern, my opening, format recommendation. Use only what I pasted; mark inferences [assumed].

ROLE & STANDARD
You teardown a competitor page to build a clearly better, more citable one. Judge on what makes a page cited:
  - Structure, depth, freshness, schema and quotability decide whether an engine lifts from a page; assess each.
  - You can only judge the page I give you; do not describe it from memory.
INPUTS
Competitor page [url or paste], my target [keyword]. If I gave only a URL you cannot fetch, ask me to paste the page.
TASK
Summarize its strengths and gaps across those five dimensions, then give 5 specific ways to make a clearly better, more citable version.
OUTPUT
Strengths, gaps, then the 5 moves ranked by impact. Judge only from what I pasted; mark inferences [assumed].

ROLE & STANDARD
You plan citation win-backs where a competitor is cited and I am not. These are hypotheses:
  - A rival usually wins on one of: a stronger source, better structure, more authority, or recency. Name the likely reason per prompt and the specific fix.
  - Mark each reason [assumed] and note how to confirm it.
INPUTS
Prompts where a competitor is cited and I am not [list]; my pages [list]. If either list is blank, ask before planning.
TASK
For each, give the likely reason they win, the specific fix, and the effort.
OUTPUT
A table (prompt, why they win [assumed], my fix, effort, how to confirm).

ROLE: You infer a competitor's content strategy from their recent posts, marking inferences as guesses.
CONTEXT: Competitor [url], recent posts [paste titles/dates].
TASK: Infer their topic priorities, formats and cadence, note where they are strong, and recommend where I should compete vs differentiate.
OUTPUT: A short read plus a compete/differentiate table, with inferences marked [assumed]. If I did not paste their posts, ask for them rather than guessing their cadence.

ROLE & STANDARD
You build an honest feature/claim matrix for a comparison page. Honesty is the whole point:
  - Mark each cell verified, unknown or not-applicable; an engine and a buyer both lose trust if you assert a competitor lacks something you cannot confirm.
  - Compare on the criteria buyers actually weigh.
INPUTS
My product [facts]; competitor [facts I can verify]. If you lack verifiable facts for the competitor, ask for them rather than guessing what they offer.
TASK
Build the matrix across buyer criteria with a verification column; do not assert an unverifiable gap.
OUTPUT
A matrix with the verification column. Mark inferences [assumed] and facts I must supply [NEEDS INPUT].

ROLE & STANDARD
You find earned-media gaps where sites cite rivals but not me. Be explicit that this is hypothesis, not data:
  - You cannot see real backlink data, so everything here is a plausible guess to confirm in a backlink tool, never a stated fact.
  - The useful output is a realistic way to earn a place on each property.
INPUTS
My brand [brand]; competitors [list]; category [category].
TASK
List the kinds of sites and specific properties that plausibly cite competitors but not me. For each, say why you think so, how I would confirm it in a tool, and how to earn the mention.
OUTPUT
A table (property type, example, why I think so [assumed], how to verify, how to earn, effort). State no link as fact.

ROLE: You pressure-test positioning for whether it is specific and defensible, since a claim a rival can copy is not a position.
CONTEXT: My one-line positioning [paste]; competitors [list].
TASK: Judge whether it is specific and defensible, name where a competitor could claim the same, and rewrite it sharper and harder to copy.
OUTPUT: Critique, then 3 rewritten options. If positioning or competitors are blank, ask before judging.

ROLE & STANDARD
You find a category angle a brand can credibly own and lead. It only works if it is true:
  - The angle must be defensible from my real strengths and distinct from what every rival already says; an invented angle collapses on contact.
  - Name why it is defensible and the first assets that would establish it.
INPUTS
Category [category]; my strengths [list]; what rivals all say [paste]. If my strengths or rivals' messaging are blank, ask up to 3 questions first, so the angle is true and not invented.
TASK
Identify a true, specific sub-angle or term I could own, why it is defensible, and the first three assets to establish it.
OUTPUT
The angle, the defensibility case, and the 3 assets. Build only on strengths I gave; mark inferences [assumed].

ROLE & STANDARD
You audit whether a brand presents as one consistent entity across the web. The mechanism:
  - Engines cite brands they can resolve as a single, consistent entity; inconsistent names, categories, descriptions or facts across properties fragment that entity and lower confidence.
  - You can only check the sources I give you; do not assume what other pages say.
INPUTS
Brand [brand]; paste how it appears on [website, LinkedIn, Crunchbase, G2, social, directories]. If I pasted only one source, ask for the others before judging consistency.
TASK
Compare name, category, one-line description, founding facts and key claims across the sources. Flag every inconsistency and give the single canonical version each should use.
OUTPUT
A consistency table (attribute, per-source values, canonical), then the fixes ordered by visibility. Judge only from what I pasted; mark gaps [NEEDS INPUT].

ROLE & STANDARD
You assess and plan for a Google knowledge panel and strong entity understanding. Be realistic:
  - A panel is earned when independent authoritative sources corroborate a consistent set of facts; Organization schema and sameAs support this but markup alone does not summon a panel.
  - Recommend only genuine, earnable signals; nothing fabricated.
INPUTS
Brand [brand]; current entity signals [website, schema, Wikipedia/Wikidata, notable coverage]. If coverage and schema are blank, ask what exists before planning.
TASK
List what is present vs missing for entity clarity (structured facts, sameAs, independent corroboration) and the realistic steps to build it, ordered by leverage. Note where notability is the real blocker.
OUTPUT
A present/missing table and a prioritised plan. Mark inferences [assumed].

ROLE: You build an author/founder entity that lends a brand credibility, using only real, verifiable credentials.
CONTEXT: Person [name], real role and credentials [paste], topics they are expert in [list].
TASK: Draft a consistent author bio (long and short), the topics to associate them with, and the profiles to align (site author page, LinkedIn, bylines). Invent no titles, awards or credentials.
OUTPUT: Long bio, short bio, topic list, profile checklist. If credentials are blank, ask for them rather than inventing any.

ROLE: You assemble the verified profile links that tie a brand's identity together for sameAs markup, since corroborating profiles strengthen entity resolution.
CONTEXT: Brand [brand]; profiles I actually control or that exist [paste URLs].
TASK: Organize the real profile URLs into a sameAs list for Organization schema, flag obvious gaps worth creating, and note which are highest value for entity clarity. Include no URL I did not provide or confirm.
OUTPUT: A sameAs URL list plus suggested gaps. If I pasted no profiles, ask for them rather than guessing URLs.

ROLE & STANDARD
You plan to disambiguate a brand whose name collides with other entities. The mechanism:
  - When a name is shared, engines blend the entities; consistent disambiguating context (category, location, distinctive descriptor) plus structured signals is what separates them.
  - Work from the real collisions; do not assume which other entities exist.
INPUTS
Brand [brand]; the other things it gets confused with [list]; my category and location [paste]. If the collisions are blank, ask what it gets confused with before planning.
TASK
Give the consistent disambiguating descriptors to use everywhere, the structured signals to add, and the on-page context that separates my entity, ordered by impact.
OUTPUT
A disambiguation plan. Mark inferences [assumed].

ROLE & STANDARD
You design an original-data study built to be cited as the source. Why it works:
  - A genuine original statistic is the strongest citation magnet, because the engine and every downstream writer must credit the source of the number; nobody else has it.
  - You design the study only; you never invent the findings, sample or results.
INPUTS
My topic and audience [paste]; data I can realistically gather [describe]. If the available data is blank, ask what I can actually collect before designing.
TASK
Propose 3 study angles that would produce a citable statistic, and for the best one specify the question, method, sample, the headline-stat format, and how to package it for citation and outreach.
OUTPUT
The 3 angles, then the full spec for the chosen one. Invent no results or numbers; mark anything I must supply [NEEDS INPUT].

ROLE & STANDARD
You recommend the right schema types for a page and how to implement them. The rules that keep schema working:
  - Markup must describe what is actually visible on the page; schema that contradicts or invents on-page content gets ignored or penalised.
  - Pick the types that match the content (Article, FAQPage, Product, HowTo, Organization, Breadcrumb); more types is not better.
  - Always validate in Google's Rich Results Test before deploying.
INPUTS
Page type and purpose [describe]; key on-page content [paste]. If the on-page content is blank, ask for it, since schema must match what is really there.
TASK
Recommend the schema types with why each fits, the required and recommended properties to fill from real content, and the validation step. Flag any type that would not match the visible content.
OUTPUT
A type-by-type plan with properties and the validation note. Invent no property values; mark anything I must supply [NEEDS INPUT].

ROLE: You generate valid FAQPage JSON-LD from real Q&A, and you keep it honest, since the questions and answers must also appear on the page.
CONTEXT: Page [topic]; my real FAQs [paste Q&A].
TASK: Output valid FAQPage JSON-LD using only the Q&A I provide, and remind me the same content must be visible on the page and validated in the Rich Results Test before deploying.
OUTPUT: The JSON-LD block plus the two reminders. If I pasted no Q&A, ask for it rather than inventing questions.

ROLE & STANDARD
You generate Product schema that helps a product win rich results without risking a penalty. The hard rules:
  - Every property must reflect real product data; price, availability and attributes must match the page.
  - Include aggregateRating and review only if real, genuine reviews exist on the page; never fabricate ratings or counts.
  - Validate in the Rich Results Test before deploying.
INPUTS
Product details [name, price, availability, attributes]; do real on-page reviews exist? [yes/no and paste]. If details are missing, ask before generating.
TASK
Produce valid Product JSON-LD from the real data. Add aggregateRating only if real reviews were provided; otherwise omit it and say why. Include the validation step.
OUTPUT
The JSON-LD plus a note on what was omitted and why. Invent no values; mark anything I must supply [NEEDS INPUT].

ROLE: You generate Article (or BlogPosting) JSON-LD from real metadata, keeping author and dates accurate since these feed entity and freshness signals.
CONTEXT: Article [title]; author [real name]; published/updated [dates]; publisher [brand].
TASK: Output valid Article JSON-LD from the real values, and remind me to validate in the Rich Results Test before deploying.
OUTPUT: The JSON-LD block plus the reminder. If author or dates are blank, ask for them rather than inventing them.

ROLE & STANDARD
You review existing schema for errors and conflicts. The usual culprits:
  - Duplicate or conflicting blocks (the theme and an SEO plugin both emitting schema), invalid or missing required properties, and markup that contradicts the visible page.
  - You judge only the markup and content I paste; recommend validating in the Rich Results Test before deploying any fix.
INPUTS
Current JSON-LD [paste]; the page content it describes [paste]. If I pasted only the markup, ask for the page content so you can catch content mismatches.
TASK
Find invalid, missing, duplicated or conflicting properties, and any claim that does not match the content. Give the corrected markup and the validation step.
OUTPUT
An issue list, the corrected JSON-LD, and the validation note. Change no values to ones I did not provide; mark inferences [assumed].

ROLE: You generate the right supporting schema (HowTo, BreadcrumbList or Organization) for a page from real data, matched to what the page actually is.
CONTEXT: Which one I need [type]; the real data [paste steps, breadcrumb path, or org details].
TASK: Output valid JSON-LD for the requested type using only my data, note the must-match-the-page rule, and remind me to validate before deploying.
OUTPUT: The JSON-LD block plus the reminders. If the data is blank, ask for it rather than inventing steps or details.

ROLE & STANDARD
You advise on AI-crawler access: robots.txt directives and whether to publish llms.txt. Be accurate, because this area is full of myths:
  - Access is governed by robots.txt and meta robots. To be cited you must allow the AI crawlers you want (for example GPTBot, PerplexityBot, Google-Extended, ClaudeBot); blocking them removes you from those answers.
  - llms.txt is a proposed convention, not an established citation signal: as of now major engines do not commit to reading it and studies show most never fetch it, so treat it as low-cost-optional, not a ranking lever, and do not imply it drives citations.
  - Crawler names change and new ones appear, so flag that the bot list may be outdated and should be verified against current docs.
INPUTS
My goal [allow/block which AI crawlers]; current robots.txt [paste]. If robots.txt is blank, ask for it before advising.
TASK
Recommend the robots.txt directives for the crawlers I want to allow or block, explain the trade-off of each, and give an honest take on whether publishing llms.txt is worth it for my case (and draft a minimal one if I want it). Flag any bot name to re-verify.
OUTPUT
Suggested robots.txt lines with rationale, plus the llms.txt recommendation with its caveat and optional draft. Mark anything version-sensitive [verify against current docs].

ROLE & STANDARD
You diagnose whether content is visible to bots that do not run JavaScript. The core fact:
  - Most AI crawlers (and some search crawlers) do not execute JavaScript, so anything injected client-side is invisible to them; content that matters must be in the raw HTML.
  - Judge only from what I paste; the definitive test is comparing the rendered page with view-source.
INPUTS
How key content loads [describe: SSR, CSR, hydration]; a view-source snippet if you have it [paste]. If you are unsure how it loads, tell me to compare view-source with the rendered page before concluding.
TASK
Identify what likely is or is not in the raw HTML, the risk to citations, and the fix (SSR, prerender, or move critical content server-side), ordered by impact.
OUTPUT
A visibility assessment and prioritised fixes, with the exact test to confirm. Mark inferences [assumed].

ROLE & STANDARD
You improve how efficiently bots crawl a site. The principle:
  - Crawlers spend limited budget; wasting it on low-value, duplicate or parameter URLs means important pages get crawled less and cited less.
  - Recommend from real signals (paste data); do not assume crawl stats you cannot see.
INPUTS
Site size [pages]; known issues [paste: thin pages, parameters, duplicates, GSC crawl stats]. If I gave no data, ask for crawl stats or a URL inventory before advising.
TASK
Identify likely crawl-budget wasters and give the directive or structure change per case (keep, noindex, redirect, canonical, prune), so crawl focuses on what matters.
OUTPUT
A per-URL/pattern decision table. Mark inferences [assumed]; flag what to confirm in GSC.

ROLE & STANDARD
You diagnose why pages are not indexed. Be a careful diagnostician:
  - Common causes: noindex, blocked in robots.txt, canonical pointing elsewhere, thin or duplicate content, discovery gaps, or crawl issues. Name the most likely per case.
  - These are hypotheses to confirm in Search Console, so mark each [assumed] and give the GSC check that verifies it.
INPUTS
Affected URLs [list]; any GSC status or messages [paste]. If you have no GSC data, ask for the index-coverage status before diagnosing.
TASK
For each URL or pattern, give the most likely cause, how to confirm it in GSC, and the fix, ranked by likelihood.
OUTPUT
A table (URL/pattern, likely cause [assumed], GSC check, fix). Confirm nothing you cannot see.

ROLE: You turn Core Web Vitals issues into a prioritized fix list, working only from the real numbers I give you.
CONTEXT: My CWV data [paste LCP, INP, CLS, by template]; platform [platform].
TASK: For each failing metric, give the likely causes and the highest-impact fixes in order. Invent no scores; if I pasted no data, tell me to pull it from PageSpeed Insights or Search Console first.
OUTPUT: A fix list grouped by metric, ranked by impact. If data is blank, ask for it rather than guessing my scores.

ROLE & STANDARD
You plan redirects and canonicals to preserve equity and avoid duplication. The rules:
  - Redirect chains and loops bleed equity and slow crawlers; each old URL should reach its final destination in one hop.
  - One canonical per intent; conflicting canonicals or canonicalising to the wrong URL splits or misroutes signals.
INPUTS
The change [migration / consolidation / duplicates]; URL mapping or duplicate set [paste]. If the mapping is blank, ask for the old-to-new list before planning.
TASK
Produce the redirect map (one hop each), the canonical decision per intent, and the conflicts to fix. Flag anything irreversible to verify before launch.
OUTPUT
A redirect/canonical table plus the pre-launch checks. Mark inferences [assumed] and mappings I must supply [NEEDS INPUT].

ROLE: You improve how fast new and updated pages get discovered, since a page that is not discovered cannot be indexed or cited.
CONTEXT: Platform [platform]; current sitemap setup [paste or describe]; publishing cadence [cadence].
TASK: Recommend sitemap structure, what to include or exclude, lastmod hygiene, and submission/ping steps to speed discovery. Note what to verify in Search Console.
OUTPUT: A discovery checklist. If setup details are blank, ask for them rather than assuming my configuration.

ROLE & STANDARD
You interpret server logs or crawl stats to see how bots treat a site. Read them carefully:
  - Useful signals: which bots crawl, how often, which sections they favour or ignore, and where they hit errors or waste time.
  - Bot user-agent names change and can be spoofed, so flag any name to verify against current docs; and a short log sample may not represent normal behaviour, so note the sample-size caveat.
INPUTS
Log sample or crawl-stats export [paste]; what I want to learn [question]. If I pasted no logs, ask for a sample before analysing.
TASK
Summarise bot activity, flag under-crawled important sections and error patterns, and recommend actions. Mark any bot name to re-verify and note if the sample looks too small to generalise.
OUTPUT
A findings summary and action list, with the caveats. Confirm nothing beyond the sample; mark inferences [assumed].

ROLE & STANDARD
You design the internal-link structure for a topic cluster. The structure is the signal:
  - A pillar linked tightly to its subtopics (and useful subtopic-to-subtopic links) tells an engine you cover the topic in depth; pages with no internal links contribute nothing to that authority.
  - Links should follow real topical relationships, not be forced.
INPUTS
Pillar and subtopic pages [paste titles/urls]. If the page list is blank, ask for it before planning.
TASK
Specify the links: pillar to each subtopic, subtopic back to pillar, and the genuinely related subtopic-to-subtopic links, each with suggested anchor text and the reason it strengthens the cluster.
OUTPUT
A link map (from, to, anchor, why). Use only pages I listed.

ROLE & STANDARD
You audit and fix internal anchor text. The principle:
  - Natural, varied anchors that describe the destination help users and engines; repetitive exact-match anchors look manipulative and add little.
  - Anchors should still be descriptive enough to signal the target topic.
INPUTS
Internal links and current anchors [paste]; target pages [list]. If anchors are blank, ask for the current link/anchor list before auditing.
TASK
Flag over-optimised or repetitive anchors, and suggest a varied, descriptive set per target that keeps topical relevance.
OUTPUT
A table (link, current anchor, issue, suggested anchors). Judge only from what I pasted.

ROLE: You find pages with too few internal links and fix their isolation, since an orphaned page gets little authority and rarely ranks or gets cited.
CONTEXT: My pages and their current internal links [paste]; priority pages [list].
TASK: Identify orphaned or weakly-linked pages and, for each, suggest 2 to 3 specific source pages to link from with anchor text. Use only pages I listed.
OUTPUT: A table (orphan page, suggested links from, anchors). If I pasted no link data, ask for it rather than guessing my structure.

ROLE: You wire a new post into the existing site both ways, since internal links in and out are what fold it into the topical cluster.
CONTEXT: New post [title/topic]; related existing pages [paste titles/urls].
TASK: Suggest which existing pages should link to the new post (with anchors) and which pages the new post should link out to (with anchors), based on real topical fit. Use only pages I listed.
OUTPUT: Two lists, inbound and outbound, with anchors. If related pages are blank, ask for them rather than inventing URLs.

ROLE & STANDARD
You resolve keyword cannibalization. The mechanism:
  - When two pages target the same intent they split signals and compete, so one canonical page should own the intent and the others are merged, redirected, re-targeted or de-optimised.
  - Decide from the real overlap; do not assume pages you cannot see.
INPUTS
Competing pages [paste titles/urls and the shared query]. If the set is blank, ask which pages and query overlap before deciding.
TASK
Identify the overlap, name the page that should own the intent, and give each other page a fate (merge, redirect, re-target, de-optimise) with the reason.
OUTPUT
A decision table (page, fate, reason) plus the consolidation steps. Judge only from what I pasted.

ROLE: You design a resource-hub link module that routes authority to the priority pages it lists.
CONTEXT: Hub topic [topic]; pages to feature [paste titles/urls].
TASK: Organize the pages into a logical hub structure (groupings and order) with descriptive anchors, so the hub passes authority to the most important pages. Use only pages I listed.
OUTPUT: The hub structure with groupings and anchors. If the page list is blank, ask for it rather than inventing entries.

ROLE & STANDARD
You turn a brand's data or expertise into story angles journalists cover. Why it earns citations:
  - Earned mentions on trusted publications build the third-party consensus engines repeat; original data and a genuine newsworthy hook are what make a journalist bite.
  - Angles must be grounded in data or expertise I actually have; do not invent findings.
INPUTS
My data or unique expertise [paste]; audience and beat [describe]. If the data is blank, ask what I actually have before pitching angles.
TASK
Propose 5 angles, each with the hook, the supporting data point, the type of outlet that would run it, and why it is newsworthy now.
OUTPUT
A table (angle, hook, data point, outlet type, why now). Invent no data; mark anything I must supply [NEEDS INPUT].

ROLE: You write a concise, genuinely newsworthy pitch built around one verifiable hook, since journalists ignore vague or inflated claims.
CONTEXT: Story angle [angle]; my verifiable proof [paste]; outlet or journalist [name or type].
TASK: Write a tight pitch (subject line plus about 120-word body) leading with the hook and the verifiable proof, with a clear reason it fits their beat. Use only facts I gave.
OUTPUT: Subject and body. If the proof is blank, ask for it rather than inventing a claim.

ROLE: You write a sharp response to a journalist source request (HARO, Qwoted, Featured, Source of Sources and similar).
CONTEXT: The query [paste]; my relevant credentials and a specific first-hand point [paste].
TASK: Write a concise response that answers the exact question, leads with one specific, first-hand, verifiable insight, and includes a one-line credential. Note: these platforms are now flooded with low-value AI-written answers, so a specific concrete point only I could make is what gets picked, not generic advice. Use only what I gave.
OUTPUT: The ready-to-send response. If credentials or the query are blank, ask for them rather than inventing expertise.

ROLE & STANDARD
You design assets that naturally earn links and citations. The mechanism:
  - Assets that become reference points (original research, a useful free tool, a definitive guide, a benchmark) get cited repeatedly because others need to point to them.
  - Concepts must be things I can realistically build and back with real data; do not propose fabricated research.
INPUTS
My topic and audience [paste]; resources I can commit [describe]. If resources are blank, ask what I can realistically build before proposing.
TASK
Propose 5 linkable asset concepts, each with why it earns links, the data or work required, and who would cite it.
OUTPUT
A table (asset, why it earns links, work required, who cites it). Mark anything I must supply [NEEDS INPUT].

ROLE: You build a podcast outreach shortlist matched to a real area of expertise, since relevance is what earns the booking and the mention.
CONTEXT: My expertise and talking points [paste]; audience I want [describe].
TASK: List the types of podcasts and the kinds of episodes that fit, a tailored angle for each, and a short pitch template. Note which would best reach my audience.
OUTPUT: A target-type list with angles and a pitch template. If expertise is blank, ask for it rather than guessing.

ROLE & STANDARD
You plan presence on the review platforms engines cite (G2, Capterra, Trustpilot and similar). The mechanism and the rules:
  - Engines lean on review-site consensus for recommendations, so genuine reviews and a complete, accurate profile are what move it.
  - Review generation must follow each platform's rules; never incentivise or fake reviews.
INPUTS
My category and the platforms that matter [paste]; current profile state [describe]. If unsure which platforms matter, ask before planning.
TASK
Recommend which platforms to prioritise and why, the profile-completeness checklist, and a compliant way to earn genuine reviews.
OUTPUT
A prioritised plan with the checklist and a compliant review-generation approach. Nothing that breaks platform rules.

ROLE: You find the location-based questions buyers ask an AI near a decision, so I can build pages that answer them.
CONTEXT: Business type [type]; city/area [location]; services [list].
TASK: List 25 local-intent questions (near me, in [area], best for, pricing, hours-type) buyers ask an assistant, grouped by intent, and note the page that should answer each.
OUTPUT: A grouped question list with owning pages. If a field is blank, state the assumption you made.

ROLE: You optimize a Google Business Profile for local and AI visibility, using only accurate business details.
CONTEXT: Business [name, category]; current profile state [describe].
TASK: Give a prioritized checklist (categories, services, description, attributes, posts, Q&A, photos) and a compliant way to earn more genuine reviews. Invent no hours, attributes or details; flag what I must confirm.
OUTPUT: A prioritized checklist. If profile details are blank, ask for them rather than assuming them.

ROLE: You write a service-area section that names real locations without keyword-stuffing cities, since genuine local relevance is the signal.
CONTEXT: Service [service]; real areas served [list]; a local detail or two [paste].
TASK: Write a natural section that names the served areas and includes genuine local context, avoiding a thin list of city names. Use only areas I actually serve.
OUTPUT: The section copy. If areas are blank, ask for them rather than inventing locations.

ROLE & STANDARD
You generate LocalBusiness schema that strengthens local entity understanding. The hard rules:
  - Every value (name, address, phone, hours, geo, areaServed) must be accurate and match the site and Google Business Profile; inconsistent NAP fragments the entity.
  - Include aggregateRating only if real reviews exist; never invent hours, ratings or coordinates.
  - Validate in the Rich Results Test before deploying.
INPUTS
Business details [name, address, phone, real hours, areas]; do real reviews exist? [yes/no]. If details are missing, ask before generating.
TASK
Produce valid LocalBusiness JSON-LD from the real data, omit aggregateRating unless real reviews were provided, and include the validation step.
OUTPUT
The JSON-LD plus a note on what was omitted and why. Invent no values; mark anything I must supply [NEEDS INPUT].

ROLE: You write review responses that read as genuine and build local trust, since responsiveness is a visible quality signal.
CONTEXT: Business type [type]; tone [tone].
TASK: Write response templates for positive, neutral and negative reviews, each adaptable and specific rather than canned, plus one worked example of handling a tough negative review professionally.
OUTPUT: The templates and the worked example. If a field is blank, state the assumption you made.

ROLE & STANDARD
You plan content and entity structure for a multi-location business. The mechanism:
  - Each location needs a genuinely distinct, locally-relevant page and consistent NAP, or the pages read as thin duplicates and the locations blur together as entities.
  - Plans use only real location data; invent nothing per location.
INPUTS
Locations [list]; what differs per location [paste]; services [list]. If per-location detail is blank, ask for it before planning.
TASK
Recommend the location-page structure, how to make each genuinely distinct, the NAP-consistency rules, and the internal-link pattern between locations.
OUTPUT
A structure plan with the distinctiveness and consistency rules. Mark anything I must supply [NEEDS INPUT].

ROLE: You find the specific questions shoppers ask an AI before buying, since product-level answers are what get cited in shopping queries.
CONTEXT: Product or category [paste]; audience [audience].
TASK: List 20 pre-purchase questions (fit, comparison, is-it-worth-it, compatibility, sizing, returns) shoppers ask an assistant, grouped by stage, with the page that should answer each.
OUTPUT: A grouped question list with owning pages. If a field is blank, state the assumption you made.

ROLE: You write a product description that answers buyer questions and is liftable, using only real attributes.
CONTEXT: Product [name]; real specs and attributes [paste]; audience [audience].
TASK: Write a description that leads with what it is and who it is for, answers the top buyer questions, and presents key specs scannably. Invent no attributes, materials or claims.
OUTPUT: The description plus a short spec list. If specs are blank, ask for them rather than inventing any.

ROLE & STANDARD
You outline a category buying guide that gets cited for "best X for Y" queries. Why it works:
  - Engines cite guides that genuinely help a buyer choose, structured around real decision criteria and best-for-whom segments, not a thin product list.
  - Recommendations must reflect real products and attributes; do not invent specs.
INPUTS
Category [category]; the products to feature [paste with real attributes]; buyer segments [list]. If product data is blank, ask for it before outlining.
TASK
Outline the guide: the decision criteria, best-for-whom picks with the reason each wins its segment, a comparison table's columns, and the FAQs to answer.
OUTPUT
A full outline with table columns and FAQs. Invent no specs; mark anything I must supply [NEEDS INPUT].

ROLE & STANDARD
You outline an honest product-vs-product comparison built to be cited. The mechanism:
  - A fair comparison on the attributes buyers weigh, ending in a clear best-for-whom verdict, is what an engine trusts and lifts; overclaiming kills it.
  - Compare only verifiable attributes for both products.
INPUTS
[product A] vs [product B]; real attributes for each [paste]; buyer priorities [list]. If attributes for either are missing, ask before outlining.
TASK
Outline the comparison: the decision criteria, an honest pros/cons each (including where mine loses), the table columns, and the verdict logic by buyer type.
OUTPUT
A full outline with columns and verdict logic. Assert no unverifiable difference; mark facts I must supply [NEEDS INPUT].

ROLE & STANDARD
You plan review and UGC markup that earns rich results safely. The hard rules:
  - Review and aggregateRating markup may be used only with real, genuine on-page reviews; fabricated or incentivised ratings risk a manual penalty.
  - The markup must match the reviews visible on the page, and be validated in the Rich Results Test before deploying.
INPUTS
Do real on-page reviews exist? [yes/no and paste a sample]; product/page [describe]. If no real reviews exist, say markup is not appropriate yet and recommend how to collect genuine ones first.
TASK
If real reviews exist, specify the Review/aggregateRating properties to mark up from them and the validation step; if not, give a compliant plan to gather them.
OUTPUT
Either the markup spec with validation, or the collection plan. Fabricate no ratings or counts.

ROLE & STANDARD
You diagnose why a store is missing from AI shopping answers and plan recovery. Be honest about measurement:
  - Visibility usually fails on one of: product data not crawlable, no Product schema, thin product answers, or weak third-party corroboration (reviews, roundups). Name the likely cause per query.
  - AI-referred shoppers often arrive labelled direct or organic, so you cannot cleanly attribute AI revenue; say so and propose proxy signals rather than implying precise tracking.
INPUTS
Queries where I am absent [list]; my product pages [list]. If either is blank, ask before diagnosing.
TASK
For each query, give the likely cause, the fix, and a proxy signal to watch (since direct attribution is limited). Mark causes [assumed].
OUTPUT
A table (query, likely cause [assumed], fix, proxy signal). Be explicit about the attribution limit.

ROLE: You map seasonal shopping moments to content with enough lead time to be indexed before each peak.
CONTEXT: Store category [category]; region [region].
TASK: List the seasonal moments and gift/use-case angles for my category, each with a content idea and a publish-by date accounting for indexing lead time. Treat windows as estimates.
OUTPUT: A calendar table (moment, angle, publish-by). If a field is blank, state the assumption you made.

ROLE & STANDARD
You write a client-ready AI-visibility report that leads with the takeaway. What makes it land:
  - Lead with the conclusion and the business implication; an executive should get the story from the first lines, with the data supporting it underneath.
  - Use only the real numbers I provide, and frame changes against the baseline so movement is meaningful.
INPUTS
Period [dates]; data [paste citation rate, share of voice, tracked-prompt wins/losses, sources]; client goal [goal]. If the data is blank, ask for it before writing.
TASK
Produce: a headline takeaway, what changed vs last period and why, wins and losses, and the recommended next actions tied to the goal.
OUTPUT
A structured report with the headline first. Invent no figures; mark anything I must supply [NEEDS INPUT].

ROLE & STANDARD
You translate AI-visibility results into business value an executive accepts. Be credible:
  - Map metrics to value (citations on buyer queries -> qualified demand -> pipeline) using a transparent, conservative chain the reader can follow.
  - AI referrals are hard to attribute cleanly (often counted as direct or organic), so state assumptions explicitly and avoid precise revenue claims you cannot defend.
INPUTS
Results [paste]; business model and rough economics [paste]; what leadership cares about [goal]. If the economics are blank, ask before modelling value.
TASK
Build the value narrative with a clear, conservative logic chain, the assumptions stated, and a defensible way to talk about impact given attribution limits.
OUTPUT
The value framing with assumptions labelled. Invent no figures; mark estimates [assumed] and the attribution caveat plainly.

ROLE: You build a repeatable monthly AI-visibility reporting checklist, so the same metrics are pulled the same way each period.
CONTEXT: What I track [paste]; tools [list]; audience for the report [internal/client].
TASK: List what to pull, the order to assemble it, and the consistency rules (same prompts, fresh sessions, baseline comparison) so months are comparable.
OUTPUT: A step-by-step checklist. If tracked metrics are blank, ask what I measure rather than assuming.

ROLE & STANDARD
You distill AI-visibility performance into a one-page executive summary. The discipline:
  - One page, conclusion first: the state, the trend, the few numbers that matter, and the decision or ask. No vanity metrics.
  - Use only real figures; frame each against its baseline.
INPUTS
Key results [paste]; the decision or ask for leadership [paste]. If results are blank, ask before writing.
TASK
Produce the one-pager: headline state, trend vs baseline, the 3 to 5 metrics that matter, and the clear ask.
OUTPUT
A one-page summary, conclusion first. Invent no figures; mark anything I must supply [NEEDS INPUT].

ROLE: You turn one citation win or loss into a short narrative a client gets immediately, since a concrete story lands better than a metric.
CONTEXT: The win or loss [paste what happened and any numbers]; client context [paste].
TASK: Write a short before/after narrative: the situation, the action, the result, and the lesson, using only the real details I gave.
OUTPUT: A tight narrative (about 150 words). If details are blank, ask for them rather than inventing the story.

ROLE & STANDARD
You specify an AI-visibility dashboard that shows decision-useful metrics. The principle:
  - Show the metrics that drive decisions (citation rate, share of voice vs rivals, tracked-prompt wins/losses, sentiment, AI-referral proxies), each with its source and a baseline to compare against; drop vanity counts.
  - Every metric needs a real, definable data source; if a metric cannot be sourced, flag it rather than inventing it.
INPUTS
Audience [internal/client]; what matters to them [paste]; available data sources [list]. If sources are blank, ask what data I can actually pull before specifying.
TASK
Specify the dashboard: the metrics, the definition and source of each, the comparison/baseline shown, and the layout by priority.
OUTPUT
A dashboard spec (metric, definition, source, comparison, placement). Mark any metric needing a source I must supply [NEEDS INPUT].

ROLE & STANDARD
You write a data-backed memo recommending next quarter's AI-visibility focus. Make it decision-ready:
  - Recommend a clear focus justified by this quarter's real results, with the expected outcome and the leading metric that would prove it working.
  - Reason only from the data I provide; do not invent results to support a recommendation.
INPUTS
This quarter's results [paste]; constraints [team/budget]; goal [goal]. If results are blank, ask before recommending.
TASK
Produce the memo: what the data shows, the recommended focus and why, the expected outcome, and the leading metric to watch.
OUTPUT
A concise decision memo. Invent no figures; mark inferences [assumed] and data I must supply [NEEDS INPUT].