How to Track AI Citations in 2026
Quick answer: A practical, end-to-end guide to Answer Engine Optimization. How AI engines decide what to cite, how to measure whether they cite you, and how to fix the pages that get crawled but skipped.
⚡ Key takeaways
- → "Tracking AI citations" is actually three different things. Whether engines cite you on your prompts, whether AI crawlers visit your pages, and whether AI-referred humans land on your site. Most teams measure one and miss the other two.
- → Engine citation tracking is what most "AI visibility" tools cover. They query ChatGPT, Perplexity, Gemini, and Claude on your buyer prompts and record what each said. Coverage varies: 4 engines (most tools) to 8 (Trakkr).
- → AI crawler tracking happens server-side because GA4 cannot see bots. You read access logs or use a plugin that hooks WordPress requests with reverse-DNS verification to filter spoofed agents.
- → AI-referred visit tracking is the hardest because AI engines strip the referrer header. You combine UTM signals, timing patterns, and landing page distribution to build a reasonable model.
- → For WordPress, CiteTrack covers all three layers in one plugin. Other platforms cover the citation layer only. The right setup depends on which layers actually matter for your work.
Most teams treat "AI citation tracking" as one thing. It is not. It is three different measurement jobs, and they need different tools.Set up one and skip the other two? You miss most of what is actually happening with AI on your site.Quick context on what you are trying to measure: in our 101-question benchmark, AI engines cited sources 95% to 100% of the time. ChatGPT used 5.6 per answer. Gemini used 14.3. So there are citations happening. The question is whether you can see them. Here is what each layer measures, how to set it up, and which tools cover each.
What's in this guide
The three layers of AI citation tracking
To know how AI affects your site, you need answers to three different questions. Each one needs its own data source.
The three measurement layers
Layer 1, engine citations
Do AI engines actually cite my pages?
Data source
AI engines queried via API or web
Tools
Profound, Peec, Otterly, CiteTrack, Trakkr, LLM Pulse
Layer 2, AI crawlers
Do AI bots visit my pages?
Data source
Server access logs (invisible to GA4)
Tools
Server log grep, WordPress plugins, CDN bot analytics
Layer 3, AI-referred visits
When AI cites me, do humans click through?
Data source
Signal fingerprinting (UTM, timing, landing pages)
Tools
WordPress plugins, custom server-side analytics
Layer 1. Do AI engines actually cite my pages? The question. When someone asks ChatGPT, Perplexity, Gemini, or Claude about your category, does your URL appear in the citations? This is what most "AI visibility" tools measure. The data source is the engine itself, queried through its API or web interface.
Layer 2. Do AI crawlers visit my pages? The question. Are bots like GPTBot, ClaudeBot, and PerplexityBot crawling your site for training or live retrieval? This is invisible to Google Analytics. The data source is your server's request logs.
Layer 3. When AI cites me, do real humans click through? The question. When ChatGPT sends a buyer to your page, does GA4 record it? Usually not, because AI engines strip the referrer header. So those visits become dark traffic, blending into "Direct" with the attribution lost. The data source is signal fingerprinting on incoming visits.
Most teams measure Layer 1 (citation tracking via a SaaS tool) and forget Layer 2 and Layer 3. The result is they know which prompts cite them but not whether the cited pages actually got crawled or whether anyone clicked through. Both Layer 2 and Layer 3 affect actual revenue more directly than Layer 1, so missing them is a real gap.
The complete tracking setup covers all three. The right tooling depends on whether your site runs WordPress and what your budget shape looks like.
The crawled-but-not-cited funnel
Layer 2
Hidden step
Layer 1
The dropoff between crawled and cited is where most visibility is lost. Layer 2 without Layer 1 cannot see it. Layer 1 without Layer 2 cannot explain it.
Layer 1: tracking engine citations
This is the layer everyone talks about. When someone says "AI visibility tracking," they almost always mean this.
The question. Across the AI engines your buyers use, which URLs get cited when someone asks the buyer questions that matter to your business?
Manual tracking method. Pick your 10-15 most important buyer prompts. Ask each one to ChatGPT, Perplexity, Gemini, and Claude. Record the sources cited in each answer. Track this weekly in a spreadsheet.
This works for small prompt sets and is genuinely useful for getting a feel for how AI handles your category. The limit is volume.
How manual tracking scales
At 50+ prompts the manual workflow breaks down. That is the crossover where tooling pays for itself.
Tooled tracking method. A dedicated AI visibility tool queries the engines for you on a schedule. Most tools support 50-500 prompts per plan with daily or weekly cadence.
Engine coverage varies. The major tools cover.
- Most cover 4 engines: ChatGPT, Perplexity, Gemini, Claude. This is the baseline for Profound, Scrunch, Semrush AI Toolkit at the Brands/Core tier.
- Otterly covers 4 base engines (ChatGPT, Perplexity, Copilot, AI Overviews) with Gemini and AI Mode as paid add-ons.
- LLM Pulse covers 5 base engines on every tier.
- Peec AI covers 6 base engines on every paid tier with unlimited users.
- Trakkr covers all 8 AI models on every paid tier (the broadest base coverage).
- CiteTrack covers 5 engines (ChatGPT, Perplexity, Gemini, Claude, Copilot) via your own API keys.
Citation tracking tools, side by side
| Tool | Base engines | Cadence | Entry price |
|---|---|---|---|
| Otterly | 4 + Gemini/AI Mode as add-ons | Scheduled | $25/mo annual |
| LLM Pulse | 5 flat | Weekly | $50/mo annual |
| CiteTrack (WordPress) | 5 via BYO API keys | Configurable | $59/yr flat annual |
| Peec AI | 6 flat, unlimited users | Daily | $80/mo annual |
| Trakkr | 8 flat | Daily | $83/mo annual |
| Profound | 4 base, 10+ at Enterprise | Daily | $82.50/mo annual |
| Semrush AI Toolkit | 4 flat (no Claude) | Daily | $99/mo standalone |
USD. CiteTrack covers Layer 1 + Layer 2 + Layer 3 on WordPress. Other tools cover Layer 1 only; you handle Layer 2 and 3 separately on non-WordPress platforms.
What to record per citation. URL cited, prompt that triggered it, engine that cited it, position in the source list, competitors cited on the same prompt, sentiment of the mention (positive, neutral, negative).
Cadence matters. Tools that run weekly (LLM Pulse) give you 52 data points per prompt per year. Tools that run daily (most others) give you 365+. For iterative fix-and-measure work, daily cadence matters. For monthly reporting, weekly is enough.
Layer 2: tracking AI crawlers
Layer 2 is the one most teams skip. It is also the one that explains most "why am I not getting cited" questions.
The question. Are AI crawlers visiting your pages? Which pages? How often? With what user-agents?
Why GA4 cannot see this. Google Analytics 4 relies on JavaScript execution in the browser. AI crawlers are programs that fetch HTML directly without running JavaScript. So every crawler visit is invisible in GA4 reports. The data exists in your server logs but GA4 cannot read it.
Method 1: Read server logs directly. SSH into your server, locate the access log (typically /var/log/apache2/access.log or /var/log/nginx/access.log), grep for known AI bot user-agents. Free, accurate, requires technical comfort.
The user-agents to look for in 2026.
- OpenAI: GPTBot, ChatGPT-User, OAI-SearchBot
- Anthropic: ClaudeBot, Claude-Web
- Google: Google-Extended (separate from Googlebot), Google-AIO (AI Overviews)
- Perplexity: PerplexityBot, Perplexity-User
- Microsoft: Bingbot for Copilot in some configurations
- Plus 30+ other AI crawlers documented as of mid-2026
Method 2: WordPress plugin that hooks requests. A plugin that runs on every incoming WordPress request can identify AI bots by user-agent, verify them via reverse-DNS, and log them to a database. The reverse-DNS step is critical because bad actors spoof crawler user-agents to bypass paywalls or scrape content. Without verification you log fake bot data.
CiteTrack ships verified AI crawler analytics in its free tier (no license required for crawler logging). Other WordPress plugins (some SEO suites, some security plugins) include basic bot logging but usually without reverse-DNS verification.
Method 3: CDN bot analytics. Cloudflare, Akamai, and similar CDNs offer bot analytics that identify AI crawlers at the edge before they hit your server. Useful if you already use a CDN. Limited if you do not.
What to record per crawl. Bot name (verified), IP address, page crawled, timestamp, user-agent string, response code (did the page load successfully). The combination tells you which AI engines are actively indexing your content and which pages they prioritize.
Layer 3: tracking AI-referred visits
Layer 3 is the hardest. And the most valuable if you sell anything.
The question. When an AI engine cites your page and the buyer clicks through, can you tell that visit came from AI?
Usually not. The HTTP referrer header is supposed to tell you where a visit came from, but AI engines handle referrers inconsistently.
- ChatGPT typically opens citation links with no referrer or in a new tab without one
- Perplexity passes a referrer in some configurations, not in others
- Gemini varies by surface (AI Mode, AI Overviews, Gemini app)
- Claude passes referrer in some browsers, strips it in others
When the referrer is empty or generic, GA4 categorizes the visit as "Direct." So your AI-referred buyers blend into your direct traffic. The revenue attribution is wrong.
What GA4 shows you vs what is actually happening
GA4 traffic sources
What is actually happening
The purple slice is real AI traffic hiding inside "Direct." Without Layer 3, you never see it.
Three signals you can combine to identify AI-referred visits without a clean referrer.
UTM parameters. Some engines (Perplexity in particular) add UTM parameters to outbound citation links. If you see ?utm_source=perplexity or similar in the URL, that is a direct signal of AI origin. Track and tag these.
Timing patterns. Genuine direct visits (users typing your URL) arrive at a steady rate. AI-referred direct visits arrive in bursts. When a popular question gets asked and you happen to be cited, you see a spike in direct traffic on the cited URL, then a decay over hours and days. Pattern-matching this in your traffic data identifies AI-referred bursts.
Landing page distribution. Genuine direct visits land mostly on the homepage. AI-referred direct visits land on the specific page that was cited. If a deep buyer-guide page suddenly gets direct traffic it never had before, that is usually an AI citation kicking in.
How signal fingerprinting identifies AI-referred visits
?utm_source=perplexity
Spike then decay on one URL
Direct visit to /blog/guide, not /
Probably AI-referred
Reclassified from "Direct" to "AI"
Revenue attribution. Once you can identify AI-referred visits, you tie them to actual revenue events. On WooCommerce or EDD, the visit-to-order link is the path. On other platforms, you need server-side analytics that can correlate visit sources to conversion events.
Setting up all three layers on WordPress
For WordPress sites the simplest setup pairs your existing analytics with one plugin that handles the AI layers.
CiteTrack is the WordPress plugin built to cover all three layers in one install.
What it does at each layer.
- Layer 1 (engine citations). Queries 5 AI engines (ChatGPT, Perplexity, Gemini, Claude, Copilot) on your buyer prompts using your own API keys. Records cited, mentioned, or absent per URL per engine.
- Layer 2 (AI crawlers). Logs verified AI bot visits with reverse-DNS verification. Covers 50+ AI bots. Free without a license.
- Layer 3 (AI-referred visits). Detects AI-referred human visits using signal fingerprinting (UTM, timing, landing pages). Ties detected visits to WooCommerce or EDD orders for revenue attribution.
Pricing is flat annual. Solo $59/yr (1 site), Studio $129/yr (5 sites), Agency $249/yr (unlimited sites). Crawler logging is free at any tier. Honest limit: WordPress only, you manage your own API key spend for Layer 1 citation queries.
Before you set up tracking, confirm AI can actually reach you. Run your key pages through the free AI Crawlability Checker to confirm you are not blocking GPTBot, ClaudeBot, or PerplexityBot. If access is broken, fix that first, or you will be tracking a result you have already lost.
For your prompt set, skip your brand name (you will usually win that) and focus on the high-intent questions where buyers choose between options: "best X for Y," "alternatives to Z," "is this tool good for small teams." Ten to twenty of these, chosen well, tell you more than a hundred vague ones. The free AI Prompt-Pack Generator helps draft a buyer-question list from your brand and competitors.
The setup workflow inside WordPress.
- Install the plugin and activate the free tier (Layer 2 crawler logging starts immediately).
- Add your own API keys for OpenAI, Anthropic, Google, Perplexity for Layer 1 citation queries.
- Configure your buyer prompt set (start with 10-20 prompts that matter most).
- Enable Layer 3 visit detection in the settings.
- Connect WooCommerce or EDD if you want revenue attribution.
- Wait a few days for data to accumulate, then read the per-URL per-engine reports.
Once the data comes in, act on it. For prompts where you are absent, study the page that got cited instead and build or rework yours to answer the question more directly. For prompts where you are mentioned but not cited, the gap is usually a missing answer capsule, weak schema, or a page the engine could not attribute cleanly. Fix the closest-to-winning pages first, then re-check and watch whether your status moves. For a wider view of how AI frames your brand across all prompts, the Brand Voice dashboard rolls presence, perception, accuracy, and source provenance into one 360° read.
Setting up tracking on non-WordPress sites
If your site does not run WordPress, the tooling shape is different. You typically need three separate tools.
For Layer 1 (engine citations). Use a hosted SaaS platform. Profound, Peec AI, Otterly, Scrunch, Trakkr, LLM Pulse, and Semrush AI Toolkit all cover this layer at different prices and engine breadths. See our best citation tracking tools roundup for the comparison.
For Layer 2 (AI crawlers). Use server log analysis or CDN bot analytics. Cloudflare's bot management identifies AI crawlers at the edge across any site. Akamai and similar offer comparable features. For self-managed servers, GoAccess or similar log analyzers can grep for AI bot user-agents. You handle the reverse-DNS verification yourself.
For Layer 3 (AI-referred visits). This is harder on non-WordPress sites because most platforms lack the request-hook layer that makes signal fingerprinting easy. Custom JavaScript on landing pages can capture some signals (UTM, document.referrer) and forward them to your analytics platform. Server-side analytics integrations can correlate visit signals with conversion events. Both are custom builds rather than off-the-shelf tools.
The combined cost on non-WordPress is typically higher than CiteTrack's flat annual pricing on WordPress, because you are paying for three tools instead of one. But the trade-off is platform freedom. Shopify, Webflow, custom CMS, headless sites all need this separated-tool approach.
Common tracking mistakes
Six mistakes that show up in almost every setup we review.
Measuring impressions instead of citations. Some tools report "AI visibility" as the number of times your brand name appeared anywhere in any AI answer. That metric is loose because it counts mentions in passing, not actual citations with a clickable link. Track URL citations, not brand mentions.
Skipping reverse-DNS on crawler logs. Bad actors spoof crawler user-agents to bypass paywalls, scrape content, or simulate AI traffic for SEO theater. If you log crawlers without reverse-DNS verification you record fake data. The verification step (checking that the IP actually belongs to OpenAI, Anthropic, Google, Perplexity) filters this out.
Assuming GA4's "direct" traffic is real direct traffic. Without Layer 3 signal fingerprinting, you treat AI-referred visits as brand-recognition direct visits. The revenue attribution is wrong. Marketing strategy decisions based on wrong attribution send budget the wrong direction.
Tracking only your brand, not competitors. Layer 1 citation tracking is most useful when you see who got cited instead of you. Tools that only track your own URL miss the competitive signal. Get a tool that records all sources cited on each prompt and compares them against your brand.
Setting up tracking but never acting on it. The most common pattern. Teams stand up an AI visibility dashboard, watch it for a week, then never look again. The work that moves citation rates (fixing pages, updating content, building authority) requires acting on what the tracking shows.
Conflating cadence with quality. Daily citation tracking is more frequent but not necessarily more useful. The right cadence is the one that matches your workflow. If you ship content fixes weekly, weekly cadence is enough. If you ship daily, daily cadence helps. Paying for daily when you act weekly is wasted spend.
FAQ
Do I need all three layers of tracking?
Depends on your use case. Layer 1 (engine citations) alone is enough if you only want to know whether AI cites you. Layer 2 (crawlers) is essential if you want to know whether AI engines are reading your pages. Layer 3 (referred visits) is essential if you sell products and want to attribute AI-driven revenue. Most serious sites benefit from all three.
What is the cheapest way to track AI citations?
On WordPress, CiteTrack Solo at $59/yr covers all three layers (with BYO API spend for Layer 1). For non-WordPress sites, the cheapest Layer 1 tool is Otterly Lite at $25/mo annual. Layer 2 via server logs is free if you have technical comfort. Layer 3 typically requires custom work on non-WordPress.
How often should I check citation tracking data?
Weekly is enough for most teams. Daily makes sense if you ship content fixes daily and want to see citation outcomes inside 24 hours. Monthly is fine if AI is a secondary channel for you. Match cadence to workflow, not to the tool's maximum capability.
Which AI engines should I prioritize tracking?
Start with the four major ones (ChatGPT, Perplexity, Gemini, Claude). Add Copilot if Microsoft is part of your buyer's tool stack. Add Google AI Overviews if Google search is critical for your category. Less common engines (Pi, You.com, Brave Search AI) matter only if your audience uses them specifically.
Do I need API keys to track AI citations?
It depends on the tool. SaaS platforms (Profound, Peec, Otterly, etc.) include API access as part of the subscription. CiteTrack uses your own API keys, which means lower flat licensing cost but you pay engines directly for query volume. The math favors your own keys at higher volume.
Can I track citations on Shopify or Webflow?
For Layer 1 (engine citations) yes, via any of the hosted SaaS platforms. For Layer 2 (crawlers) yes, via Cloudflare or server log analysis. For Layer 3 (referred visits) it is harder; you typically need custom JavaScript and server-side analytics integration. WordPress has the simplest unified setup for all three.
What user-agents identify AI crawlers in 2026?
The major ones: GPTBot, ChatGPT-User, OAI-SearchBot (OpenAI), ClaudeBot, Claude-Web (Anthropic), Google-Extended, Google-AIO (Google), PerplexityBot, Perplexity-User (Perplexity). The full list runs to 50+ as of mid-2026 and grows monthly. Plugins and CDN bot databases keep updated lists.
How do I tell if a crawler visit is real or spoofed?
Reverse-DNS verification. Take the IP address that made the request, resolve it back to a hostname, and check whether the hostname belongs to the claimed organization (openai.com, anthropic.com, google.com, etc.). If the reverse-DNS does not match the user-agent's claimed source, the visit is spoofed. Tools that do this automatically save you from logging fake data.
What if AI engines change their referrer behavior?
Layer 3 signal fingerprinting will need updating when engines change. The UTM signal is the most stable (engines either send UTMs or they do not). Timing patterns and landing page distribution are inferred signals that survive any specific engine behavior change. So the model degrades gracefully when engines update; it does not break.
How do I tie AI citations to revenue?
On WordPress with CiteTrack, the plugin connects to WooCommerce or EDD and recognizes AI-referred visits in the order flow. On other platforms, you need custom server-side analytics that correlates visit sources to conversion events. Pure SaaS citation trackers do not handle this directly because they live at the engine layer, not the site layer.
Where to go from here
Tracking is the foundation. The next question is what to do with the data. If a page gets crawled but never cited, the gap is usually Citability, the structural signals that make a page quotable. Our 101-question benchmark shows what normal citation rates look like across engines so you can gauge where you stand. For the WordPress-specific deep dive on setup, the citation tracking walkthrough and the crawler tracking setup go step by step. And if you are comparing tools before committing, the citation tracker comparison has current pricing and tradeoffs.
See it in your own WordPress dashboard
Find out if AI recommends you — or a competitor
Install CiteTrack AI and see exactly who crawls you, who cites you, and what it earns. Flat pricing from $59/year, 14-day money-back guarantee.
Want to see if AI cites pages like this?
CiteTrack AI tracks your AI crawlers, citations, and the revenue they drive, inside WordPress.
Explore: Features · Crawl Academy · Glossary · Compare · Pricing