Short answer
Ask the assistants the questions your buyers ask, in a signed-out or fresh session, and run each question several times. A citation is a link to your page among an answer’s sources; a mention is your brand named without a link. Then confirm it with data: ChatGPT-User and Perplexity-User visits in your server logs, the AI Assistant channel in Google Analytics, and Bing’s AI Performance report for Copilot.
Key takeaways
- A mention and a citation are different signals: one names your brand, the other links to your page as a source.
- Check in a clean session. ChatGPT can use saved memories and your approximate location when it searches, so a signed-in check shows your own version of the answer.
- Run every question several times. In SparkToro’s 2026 test, ChatGPT and Google’s AI returned the same list of brands less than once in 100 runs.
- Your server logs show when an assistant reads your page for a user: OpenAI’s ChatGPT-User and Perplexity’s Perplexity-User fetch pages while they answer.
- Google Analytics’ AI Assistant channel and Bing’s AI Performance report turn the spot check into numbers you can track.
Mentions and citations are different signals
An AI assistant can include your brand in two ways. A mention names the brand in the answer text. A citation links to one of your pages as a source for the answer. Both count, and they answer different questions: a mention shows the model connects your brand with the topic, and a citation shows your page was retrieved and trusted enough to quote.
They often come apart. A model can name a brand it knows from training with no link at all, and it can cite a page from your site for a fact while recommending a competitor. So I record both, for every question and every engine. This article is the hands-on check; the full method, with the metrics and the monthly report, is in how to measure AI search visibility.
How to check by hand
A useful spot check takes an afternoon:
- Write ten questions your buyers ask before they buy, in their words: what the best options are in your category, how two options compare, how to solve the problem your product solves, and what your brand is like.
- Use a clean session: signed out, or a new session with memory off. ChatGPT may use saved memories when it rewrites a search, and an approximate location from your IP address [1]. Signing out strips chat history and memory, so you see the default answer [8].
- Ask each question in ChatGPT, Gemini, Perplexity, Google (AI Overviews and AI Mode) and Microsoft Copilot.
- For each answer, record whether your brand is named, whether one of your pages is linked, which page it is, and which competitors appear.
- Ask each question three times in each engine. In SparkToro’s study of 2,961 runs, ChatGPT and Google’s AI returned the same list of brands less than once in 100 runs [7], so one answer proves little.
- Put the results in one sheet, a row per question, engine and run, so the next check compares like with like.
Ten questions, five engines and three runs give 150 answers, enough to see whether you appear at all and where.
Where each engine shows its sources
| Engine | Where the citations appear | What to know |
|---|---|---|
| ChatGPT | Citations in the answer, and a Sources button, when available, that lists the cited pages and related links [1] | Answers that use web search may include citations [1] |
| Gemini | A Sources button at the bottom of the response or within it, which opens the links in a side panel [2] | Responses without the Sources button have no links [2] |
| Perplexity | Links to the pages it read, in the response [3] | Pages it reads for a user are fetched by Perplexity-User [3] |
| Google AI Overviews and AI Mode | Supporting links shown with the answer [4] | A page has to be indexed and eligible for a snippet [4] |
| Microsoft Copilot | Citations displayed as sources in the answer [5] | Counted in Bing’s AI Performance report [5] |
Open each cited link and note the exact page. A citation to your pricing page and a citation to an old blog post are different results, and the page tells you what the engine trusts you for.
What your server logs show
Your logs record the moment an assistant reads your page for someone. OpenAI says that when users ask ChatGPT a question, it may visit a web page with the ChatGPT-User agent, and that this agent does not crawl the web on its own [6]. Perplexity says Perplexity-User may visit a page to help answer a question and include a link to it in the response [3].
So a request from ChatGPT-User or Perplexity-User is the closest signal to a citation your own server holds: a real question, answered with your page in hand. Filter your access logs or your CDN’s logs for those two agents, and list the pages they fetch by week. Check the requests against the IP ranges both companies publish, because anyone can copy a user agent [3] [6].
The search crawlers tell a related story. OAI-SearchBot and PerplexityBot fetch pages for each engine’s search results [6] [3]; regular visits from them show your pages can be found, the step before any citation.
What analytics and webmaster tools show
- Google Analytics: the AI Assistant channel collects visits from assistants such as ChatGPT, Gemini, Copilot and Grok when the referrer matches Google’s list [9]. Its landing pages show which of your pages the assistants link to and people click.
- Bing Webmaster Tools: the AI Performance report counts citations of your pages in Copilot, Bing’s AI summaries and selected partners, with the grounding queries the AI used to retrieve them and the cited pages [5].
- Google Search Console: the Generative AI performance report shows your impressions in AI Overviews and AI Mode by page, country and device [10], and the Performance report’s Web search type counts their clicks and impressions together with the classic results [4].
Bing’s report also names the grounding queries behind your citations, the quickest way to see which questions your pages answer for an AI engine.
A worked example: one afternoon’s spot check
The firm and its numbers here are illustrative. An accounting software company runs the check above: ten questions, five engines, three runs each, 150 answers.
The brand is named in 21 answers, 14%, and linked in 6, all of them on Perplexity and Copilot, all to its pricing page. A competitor is named in 63. The firm’s logs show ChatGPT-User fetching the pricing page 38 times that month and its comparison pages never, and Google Analytics shows 52 visits in the AI Assistant channel, 40 of them landing on pricing. Bing’s AI Performance report lists “accounting software for contractors” among the grounding queries behind its citations.
That picture is specific enough to act on: the engines trust the pricing page, ignore the comparison pages, and connect the brand with contractors. The next steps are to rebuild the comparison pages around the questions in the check and to cover the contractor angle in depth, then to run the same 150 answers again next month.
What to do when AI names your competitors instead
The answers that leave you out show you what wins. Open the pages the engines cite in your place and note what they hold: a direct answer in the first lines, prices, comparisons by name, recent dates, reviews on other sites. That list becomes your content strategy.
Then work on three fronts. Build the pages the questions call for, organised so your site covers the whole subject, because topical authority is what makes an engine return to the same site search after search. Make the brand an entity the engines connect with the category, with the same name and description on your site, your profiles and the sites that write about you. And earn mentions in the reviews and comparisons the engines already cite. I cover the ChatGPT side step by step in how to get your brand cited by ChatGPT.
When the spot check grows past what one person can run, it becomes a tracked prompt set, run every month in every engine with tools I build. On one program for a FinTech comparison brand across 26 markets, AI Overview and assistant citations rose from 14 answers to more than 160, tracked prompt by prompt against its competitors. That is the work of AI visibility tracking.
Frequently asked questions
Why does ChatGPT give a different answer each time I ask?
AI answers are generated fresh on every run, so the brands and their order change. Ask each question several times and read the share of answers that name you, across all the runs.
Should I check signed in or signed out?
Signed out, or in a new session with memory off. Saved memories and your location can shape what ChatGPT searches for, so a signed-in check shows an answer tailored to you.
What is the ChatGPT-User agent in my logs?
It is the agent ChatGPT uses when it visits a web page to answer a user’s question. A request from it means someone’s question led ChatGPT to read that page.
Can I see which questions led to a citation?
For Copilot and Bing, yes: Bing’s AI Performance report lists the grounding queries behind your citations. For the other engines, tracking tools rebuild the picture by running prompts and recording the answers.
Does a citation bring traffic?
Some citations bring clicks, which Google Analytics files under the AI Assistant channel or, for Google’s AI features, under Organic Search. A mention without a click still shapes which brands a buyer considers.
Sources
- OpenAI Help Center, “Searching the web with ChatGPT”, updated October 2026. help.openai.com
- Gemini Apps Help, “View related sources from Gemini Apps”, read 28 October 2026. support.google.com
- Perplexity, “Perplexity Crawlers”, read 28 October 2026. docs.perplexity.ai
- Google Search Central, “AI features and your website”, last updated 10 December 2025. developers.google.com
- Microsoft Bing Webmaster Blog, “Introducing AI Performance in Bing Webmaster Tools Public Preview”, 10 February 2026. blogs.bing.com
- OpenAI, “Overview of OpenAI Crawlers”, read 28 October 2026. developers.openai.com
- SparkToro, Rand Fishkin, “NEW Research: AIs are highly inconsistent when recommending brands or products; marketers should take care when tracking AI visibility”, 28 January 2026. sparktoro.com
- Ahrefs, Louise Linehan, “How ChatGPT Does Query Fan-Out (15 Million Queries Studied)”, 12 October 2026, corrected October 2026. ahrefs.com
- Google Analytics Help, “[GA4] Default channel group”, read 28 October 2026. support.google.com
- Search Console Help, “Generative AI performance report (Search)”, read 28 October 2026. support.google.com