What matters first.
- AI citation tracking records which web pages ChatGPT, Perplexity, Claude and Google AI Overviews cite as sources when they answer your buyers' questions, and how often your own domain is among them.
- The list is only the start. Sort every source into four piles: yours, a rival's, a neutral third party, noise. Work first on the third-party page that AI answers cite when they name two or more rivals and not you.
- A higher citation rate is a win only when it is larger than normal variation. FameScan shows it with a 95% interval and a verdict: Real change, Within noise or Too few answers.
AI citation tracking records which web pages ChatGPT, Perplexity, Claude and Google AI Overviews cite as sources when they answer your buyers’ questions, sorts each source into yours, a competitor’s or a third party’s, and shows the share of answers that cite your domain over time. The list of sources is the easy part. The decision is which source to work on first, and whether a move in your citation rate is real.
Start tracking your cited sources
FameScan is AI citation tracking with the evidence attached: it asks ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews your buyer questions, stores every answer with the sources it cites, labels each source by type, and reports your citation rate with a 95% interval and a “real change versus noise” verdict. Facts about the engines and about Bing come from their official pages, opened on November 13, 2026.
What is AI citation tracking?
AI citation tracking is the practice of recording which web pages AI engines link as sources when they answer the questions your buyers ask, across engines and over weeks. A citation here is the link an engine shows as evidence for its answer. It is not the academic kind of citation, and it is not the same as being named.
A mention is not a citation, and a citation is not a recommendation. They are three different wins, and mixing them into one number hides the real problem. The table shows what each one proves.
| Win | What you see | What it proves |
|---|---|---|
| Mention | Your brand name in the text | The engine counts you in |
| Citation | A link to a page, shown as a source | That page shaped the answer |
| Recommendation | The engine tells the reader to choose you | You won the comparison |
Each win calls for different work: for a mention, how the pages an engine reads describe you; for a citation, the page itself, yours or a third party’s that names you; for a recommendation, your differentiators and the comparison and review sources. The three come apart often: an engine can name you while it cites a review site about you. It can cite your pricing page and still recommend a rival. FameScan tracks all three separately. This guide is about the second: the sources.
Sort every cited source into one of four piles
Most pages on this topic stop at the list of sources a tool collects. The list is where the work starts: every source belongs in one of four piles, and each pile calls for different work.
- Yours: pages on your own domain. They tell you which of your pages engines trust as evidence. Keep them current and make them worth citing, with data, documentation and comparisons.
- A rival’s: pages on a competitor’s domain. You cannot get onto them, but they show you which kind of page earns the link.
- Neutral third parties: review and directory sites such as G2, Capterra or Wikipedia, communities such as Reddit, trade media, blogs and best-of lists. This pile has the most leverage, because here you can earn a place.
- Noise: a third-party source cited once, by one engine, in one answer, and never again. One answer is a snapshot; the pattern is the truth. Watch it, do not act on it yet.
FameScan labels each source by type: Owned, Competitor, UGC, Reference, Editorial, Institutional, Corporate, Other. Owned and Competitor come from the domains you enter for your project and your competitors. The other types come from curated lists of review, community and media sites, from words in the domain name and from the domain ending; Corporate collects every other company site. Before you plan work on a source, open the page and read what it says about you and your rivals.
Each URL also gets a page type: Homepage, Comparison, Alternative, Listicle, How-to guide, Article, Product page, Category page, Pricing page. When the same type keeps winning across your questions, you know which kind of page to build, or to get onto.
The order of work follows from the piles: the third-party page first, then your own page, then a rival’s page type, noise last. The full rule is under which source first.
Can you track AI citations for free?
Partly, and the free route is a good way to learn what a sample needs. Three routes before you pick a tool:
- By hand: ask each engine your top buyer questions without your brand name and log every source in a sheet, with the date, engine, question and who was named. It stops where the variation starts: ten questions in four engines once a month give you a snapshot, not a trend.
- Bing Webmaster Tools: the AI Performance report, a public preview since February 10, 2026, shows for your site how often your content is cited in AI answers from Microsoft Copilot, AI summaries in Bing and select partner integrations: total citations, average cited pages, a sample of the queries behind them, citations per URL and the trend over time. It reports on content from your own site.
- Google Search Console: according to Google, traffic from AI features such as AI Overviews and AI Mode is counted in the Performance report under the Web search type. Here too you see only your own pages, not the sources cited instead of you.
A tool pays off once you need rivals, third-party sources, several engines and a verdict you can defend. If you are comparing citation analysis options for AI search, ask each one four questions: can I open the answer behind a number, does it show the exact page cited and not only the domain, does it separate my domain from rivals and third parties, and does it tell me when a change is real? If Profound is on your shortlist and its price or fixed trial is the problem, see how the Profound alternatives compare.
FameScan Lite costs $12 a month, or $9 a month billed yearly, with 20 Answer Credits a month, 10 prompts, 3 competitors and 2 engines. Starter costs $49 a month, or $39 a month billed yearly, with 200 Answer Credits, 50 prompts, 5 competitors, 3 engines and daily runs, checked November 13, 2026. The first check is free and one time; after that you continue on a plan with a 14-day trial. Every plan is on the pricing page.
Which engines cite sources, and how?
Engines show sources differently, so compare citation rates within one engine first and across engines second. The table shows how FameScan asks ChatGPT, Perplexity, Claude and Google AI Overviews, and what you see as sources.
| Engine | How FameScan asks | What you see as sources |
|---|---|---|
| ChatGPT | With its own web search | The sources ChatGPT returns with the answer |
| Perplexity | With its own web search | Numbered citations linking to the original sources |
| Claude | With its own web search | Citations of the pages the answer draws on |
| Google AI Overviews | From the retrieved search result, by country | The links the Overview shows |
For Google AI Overviews there are no additional requirements, Google says: a page needs to be indexed and eligible to show in Search with a snippet.
How do I track my citations?
Fix a set of buyer questions, ask them in each engine on a schedule, store every answer with its sources, and count how often your domain is among them. In FameScan that looks like this:
- Questions: 20 to 50 buyer questions without your brand name, grouped by topic. Questions that name your brand already contain you; FameScan keeps them out of visibility and citation rate and uses them for sentiment. How to build a question set that mirrors real buyers is in our guide to prompt tracking.
- Domains: enter your domain and each competitor’s. That is what turns a cited link into Owned or Competitor.
- Engines and countries: the engines your buyers use and the markets you sell in.
- Repeat: FameScan spreads your Answer Credits over the month and asks first the combinations of question, engine and country that were observed least recently, so every question gets its turn over the period.
- Read: under Answers you open any stored answer with its cited sources; your own domain is marked “your domain”. The Sources group holds Citation overview, Domains, URLs and Gap analysis. Domains lists every domain with its type, retrievals and the share of answers that cite it; URLs shows the page type and whether your brand was mentioned in those answers.
When is a change in citation rate real?
A change is real when it is larger than the normal variation between answers, and only an interval can show that. Answers vary. A move counts when the statistics say so.
Citation rate in FameScan is the share of answers to non-branded questions that cite your domain. FameScan pools every answer in the selected period, compares it with the previous period of equal length, and shows the rate with its 95% interval and a verdict: Real change, Within noise or Too few answers. Answers to the same question are treated as correlated, so the interval stays honest rather than flattering.
An invented example: your citation rate goes from 11 to 14 percent, from about 180 answers per period. The interval runs from roughly 9 to 21 percent and the verdict reads Within noise. Report it as stable, not as a win. A manual sheet with ten questions in two engines every week gives you 80 answers in four weeks, under half the example’s 180, so only large moves will show. Our guide to measuring AI visibility explains why a sample beats a screenshot.
Compare like with like: the same questions, engines and countries in both periods. Add ten questions or a new engine and your rate moves although no answer changed, so compare only periods that both start after the change instead of reading that move as a win or a loss.
Alerts fire only for what passes this test, and they are corrected for testing many numbers at once, so chance moves rarely raise an alarm. For citations that means “your brand is cited less often” or “is cited more often”, with your brand name in front, when your rate really moves, and “Source no longer cited” when a domain was cited at least three times in the previous period and not once now, although the number of answers makes that unlikely by chance.
A worked example: one question, four engines
The question is “best shift scheduling app for a restaurant with 20 staff”. In one week, the engines answer like this:
- ChatGPT names Rotabase and Crewly and cites a best-of list in a hospitality trade magazine and Rotabase’s comparison page.
- Perplexity names Crewly and cites a Reddit thread from restaurant managers, the same best-of list and a Capterra category page.
- Claude names Shiftwell and Rotabase and cites Shiftwell’s pricing page.
- Google AI Overviews names both rivals and cites the best-of list and the Capterra category page.
Sorted, it looks like this. Yours: the pricing page, cited by one engine. A rival’s: Rotabase’s comparison page. Third parties: the best-of list in three engines, the Capterra page in two. Noise: the Reddit thread, cited once so far.
The decision falls on the best-of list. Three engines cite it, each time in answers that name a rival and never Shiftwell; ChatGPT and Google AI Overviews name both. Over four weeks it tops the Gap analysis with 12 answers without Shiftwell and 2 competitors there, a gap score of 24. The Capterra profile comes next. Rotabase’s comparison page shows what is missing: Shiftwell has no comparison page of its own. The Reddit thread stays on watch.
Shiftwell’s weekly report to the team gets three things anyone can check line by line: the three stored answers that cite the list without naming Shiftwell, the Gap analysis row with its score, and the date the editor was contacted. So the first outreach goes to the trade magazine, not into another round of edits on Shiftwell’s own pages.
Which source should you work on first?
Work through the piles in this order, because it raises the odds that your work changes an answer:
- The third-party page cited in answers that name two or more rivals and not you. It shapes the answer you are missing from, and you can earn a place there: a listing, reviews, a correction, data for the editors. Match the move to the type: on a directory, complete your listing and ask customers for reviews; on a best-of list, send the editor the facts the list compares on; in a community thread, answer openly as the vendor where the rules allow. In FameScan, open Sources, Gap analysis, set “Min. 2 competitors” and sort by Gap score: answers without you times the competitors there. The list works at the level of domains, hosts or URLs.
- Your own page, when you are named often but rarely cited. The Action Center flags this pattern as “Talked about, never cited”. Make the page worth citing, with data, documentation and comparisons, and check that AI crawlers can read it at all, as our guide to AI crawlers explains; the AI readiness audit checks robots.txt for each AI crawler and whether your key pages show their text without JavaScript.
- A rival’s page that keeps being cited. Read its page type and build the counterpart on your domain. Note what it has that yours lacks: data, a comparison table, a recent date. Which on-site steps come first is in our guide to generative engine optimization.
- Noise. Wait until a source comes back.
If the third-party page describes you wrongly, that is an accuracy job first: how to find and correct the responsible source is in our guide to AI brand monitoring.
You do not have to read the Gap analysis yourself. Ask the Copilot in the portal “Which sources cite my competitors but not me in the last 30 days, at URL level?” and it answers with charts from the same data. One click on “Ask the Copilot” on an Action Center item starts a conversation with the question already filled in. If you work in Claude or Cursor, the MCP server returns the same list through its source gaps tool.
When you mark an Action Center item such as “Talked about, never cited” completed, FameScan compares your citation rate in the 14 days before and after and says whether the difference is real, or still too early to judge.
Today: one note to one editor
The smallest step that can change an answer is a short note to whoever keeps your strongest third-party source up to date. Take one buyer question you are losing, ask it without your brand name in two engines, and find the third-party page cited in the answers that name rivals and not you. Then write three lines: the fact the page compares on, where your proof is, and your request.
Log the date you sent it. Whether answers that cite the list start naming you shows over the following weeks in your visibility and in a falling gap score for the list, not in tomorrow’s answer.
Frequently asked
What is AI citation tracking?
Recording which web pages AI engines cite as sources when they answer the questions your buyers ask, across engines and over time. It shows whether your own domain is cited, which rivals' pages and which neutral third-party pages are, and how the share of answers citing you moves. A mention names you, a citation links a page as evidence, a recommendation tells the reader to choose you. Track them separately.
What is the difference between an AI mention and an AI citation?
A mention is your brand name in the answer text. A citation is a page the engine lists as a source. You can be mentioned without being cited, when the engine relies on other pages to describe you, or cited while the engine recommends a rival. Both matter, but they call for different work: the description for a mention, the page itself for a citation.
Does every AI engine show its sources?
Not in the same way. Perplexity adds numbered citations that link to the original sources, Claude with web search includes citations in its responses, and Google AI Overviews show links to pages from Search. When FameScan asks ChatGPT, it answers with web search and returns its sources. Each engine shows sources its own way, so compare citation rates within one engine first.
How often should I check AI citations?
Run the same questions on a fixed schedule and judge a change over periods of four weeks or more, never one run against the next. Answers vary from run to run, so one check is a snapshot. A fixed set of buyer questions, repeated across weeks, shows whether a change in citation rate is larger than normal variation. FameScan compares each period with the one of equal length before it and shows the interval and a verdict.
Can I track AI citations for free?
Partly. You can ask the engines yourself and log every source in a sheet, Bing Webmaster Tools has an AI Performance report on how your own pages are cited on Microsoft's AI surfaces, and Search Console counts traffic from Google's AI features. None of these shows rivals, third-party gaps or whether a change is real. FameScan Lite costs $12 a month, or $9 a month billed yearly, with 20 Answer Credits, checked November 13, 2026.
Which cited source should I work on first?
Start with a third-party page that AI answers cite when they name two or more of your competitors and not you. It already shapes the answer, and unlike a rival's own page, you can earn a place there: a listing, reviews, a correction or data. Then fix your own page if you are named often but rarely cited.
Sources & further reading
Sources checked on 13 November 2026.