How to track mentions of your company in ChatGPT and AI responses?
You open ChatGPT, type "what are the best companies for…", and your brand appears. Good news!
You ask the same question again two hours later. She's disappeared 😵
So, are you visible in AI or not?
The reality is that a single test no longer allows us to seriously answer this question. It provides a snapshot at a specific moment. But to truly drive your visibility in ChatGPT, Perplexity, Gemini, or Google AI, you need to base your strategy on a trend.
That's the whole difference between testing your AI visibility and monitoring it.
In this guide, I want to show you what to track, with which prompts, how often, what the new Google reports change, and why monitoring tools should remain instruments of measurement and not just machines for creating a false image.
You only have 30 seconds? Here's the recap
- A ChatGPT response is not a stable ranking: it can vary from test to test, from account to account, from moment to moment.
- Manual control is useful for diagnosing a problem, but insufficient for monitoring progress.
- The right system combines a stable set of questions, multiple AI engines, repeated observations and time-based analysis.
- Not all useful metrics are equivalent: Mention, citation, AI share of voice (Share of Model) and Google printing they do not measure the same thing.
- Since June 2026, Search Console has been able to provide certain sites with a dedicated view of impressions in Google's generative features (AI Overview, AI Mode). However, it does not measure ChatGPT or Perplexity.
Why a manual ChatGPT test is no longer sufficient
Let's start with the basics. Yes, asking ChatGPT once, "Recommend three agencies for X," can be very useful. It allows you to immediately see if your company is recognized, how it's described, which competitors are mentioned, and which sources seem to influence the answer.
But it remains an isolated test, not a figure, not a metric (famous KPI), not a trend line.
Generative engines are non-deterministic.
What does that mean? WhatThe same wording can produce a different answer depending on the model, the timing, the location, the accessible sources, and the "reflection" itself. Ahrefs reiterates this in its Brand Radar methodology:
“ AI viewability metrics should be read as directional indicators, not as an exact measure of the audience actually exposed .” Ahrefs
This is also the central point of our guide on Share of Model : following a "position #2 in ChatGPT" like you would follow a Google position is a bad idea.
Manual testing therefore remains excellent for investigation, but it is a poor basis for proving a trend and thus justifying an AI visibility (GEO ) strategy.
What is true AI visibility monitoring?
Reliable monitoring always answers the same questions, with a method stable enough to compare two periods.
It must freeze at a minimum:
- the topics your prospects are actually searching for about AI
- a set of representative prompts that attack from different angles
- the engines observed
- the market or location when the tool allows it
- measurement frequency
- the selected metrics
- marketing changes that occurred between two measures
The goal is not to obtain "the true figure of your AI visibility," let's be clear, it doesn't exist today!
The goal is to build a tool that is coherent enough to answer useful questions for guiding your AI visibility strategy, and overall marketing strategy: Are we appearing more often than three months ago? On which topics? In which search engines? With which sources? Against which competitors?
Step 1: Build a set of prompts that resembles the questions your prospects ask.
The classic pitfall is to only test your brand name.
"What do you know about [Company]?" primarily checks if the search engine is familiar with you. It doesn't indicate whether it will recommend you to a prospect who doesn't yet know your name (which is often the case at the top of the funnel when they're just starting to search for a solution to their problem).
Your set of follow-up prompts must therefore cover several levels of intent.
The discovery prompts
They are reproducing a search without a brand name:
- "Which companies can help me to…"
- "how to choose a service provider for…";
- "What solutions exist for…".
These are often the most strategic, because they test your presence before the user has chosen you.
The comparison prompts
They are testing the shortlist:
- "What are the best options for…"
- "X or Y: what are the differences?"
- "Compare three providers for…".
Brand prompts
They verify their understanding of your entity: activity, positioning, geographical area, management, products, clientele.
If the search engine mixes up your company with another or assigns it incorrect information, the problem is no longer just visibility. It's an entity accuracy issue , which we'll detail in our guide to ChatGPT errors on a company.
To delve deeper into this logic, also read about why entities are taking over from simple keywords.
Step 2: Don't just measure ChatGPT
Saying "we are visible in AI" after testing a single AI engine is like measuring all your organic visibility with a single Google keyword.
ChatGPT, Perplexity, Gemini and Google's generative features (AI Overviews) do not work with exactly the same systems, sources or search mechanisms.
At a minimum, your tracking table must distinguish:
- ChatGPT
- Google AI Overviews / AI Mode
- Perplexity
- Claude
- Gemini:
Depending on your market, Copilot or other engines can then be added.
This search engine-by-search engine analysis is essential for gaining a clearer understanding. A brand might have a strong presence on one platform and be almost absent on another. Your optimization strategy should then be prioritized based on the platform(s) most used by your target audience.
Step 3: Separate mention, citation, and Google visibility
These metrics are often mixed together in the same dashboard even though they answer different questions.
| Metric | What she tells you | What she does not prove |
|---|---|---|
| Brand mention | Your company appears in the response | Your site is the source used; you are simply mentioned. |
| Quote | A URL or domain is cited as the source | The fact that the brand is favorably recommended is simply being used to gather information. |
| AI Share of Voice / Share of Model | Your relative frequency of presence compared to competitors on a given corpus | The actual number of people exposed remains a simulation, just the most comprehensive one possible. |
| Google AI Printing | A URL from your site was displayed in a Google generative feature | Even if a prospect has read or remembered your brand, they may not have even seen it 😵 |
Since June 3, 2026, Google has been rolling out performance reports for its generative experiments in Search Console . These reports can show impressions, pages, countries, devices, and changes over time for the websites in question.
This is a major breakthrough, because it's the first AI response platform that gives us access to a metric at the source!
But keep the distinction clear: Search Console only measures Google, not ChatGPT, not Perplexity.
And an impression doesn't automatically become a lead. To avoid this shortcut, our article on reading KPIs in the age of AI is a very good source of additional information.
Step 4: Choose a frequency before you begin
For strategic prompts, you can choose a weekly or monthly frequency depending on how quickly your market and content evolve. Tools like Ahrefs now allow you to track custom prompts daily, weekly, or monthly across multiple assistants (if you're willing to pay the premium price for the tool).
The important thing is not to choose the highest frequency, but to remain consistent in order to gather a sufficient volume of information over time. And always at regular intervals to avoid skewing the measurement.
If you measure ten times this week, once the following month and fifty times after a redesign, you create a series that is impossible to interpret.
Step 5: Keep a manual baseline, but automate the repetition
Here are 5 situations where manual tracking makes sense for monitoring / analyzing / investigating:
- understanding a surprising answer
- read carefully how your brand is described
- identify a problematic source
- test a new research intent
- check for an entity error
On the other hand, manually copying and pasting the same prompts every month is no longer the best use of your time when you are looking to compare dozens of questions, multiple engines, and multiple time periods.
Tools like Ahrefs Brand Radar, Profound, Otterly or specialized modules from SEO suites can automate part of this work.
The selection criterion is not "which dashboard is the prettiest?" (even though I absolutely understand the argument! That's why at Ellevate we developed our proprietary tool to meet exactly our business requirements and criteria).
Instead, ask:
- Can I define my own prompts?
- Are the location and model documented?
- How often are the responses updated?
- Can I see the answers and quotes behind the score?
- Is the metric measured or modeled?
- Can I compare periods using the same method?
Ahrefs, for example, specifies that its "estimated impressions" model potential visibility based on Google search queries: these are not impressions actually observed in ChatGPT . You really need to keep this distinction in mind; it's essential!
The minimum dashboard to build
You don't need 15 indicators battling it out. A simple tracking dashboard that you understand, and that allows you to make concrete business and strategic decisions, is better than a pretty, colorful, but unusable chart.
For each strategic theme, keep in mind:
- frequency of mentions of your brand
- frequency of mention of 3 to 5 competitors
- quotes from your field
- main third-party sources cited
- engine
- data
- change compared to the previous period
- marketing change that has occurred since the last measurement
For Google, add impressions from the generative report in Search Console when available.
Then link this visibility layer to what your business is already measuring: brand search, direct traffic, forms, CRM opportunities, lead origin.
AI monitoring is useful for your business and strategy when it helps explain the decision-making process of the human buyer on the other end of the keyboard.
What mistakes should be avoided?
"I've tested ChatGPT, so I know where I stand."
No. You obtained a one-off diagnosis. Very useful for investigating, but insufficient for measuring a trend and therefore for building and managing your strategy.
"Our brand is third in ChatGPT"
A generated response is not a stable SERP (Search Engine Results Page). Track a frequency of appearance, not a ranking presented as permanent.
"The tool displays 80,000 AI impressions, so 80,000 people have seen us."
Check the metric definition. Some impressions are modeled, while others are actually measured by a platform like Google. These are not interchangeable. This means that someone might have seen you in 150 different ways, and that would count as 150 impressions. Carefully analyze what is being measured before drawing any conclusions.
"We need to monitor 500 prompts from day one."
Not necessarily. Start with the intentions that truly change a decision: discovery, comparison, choosing a provider or product. Then broaden your scope once you have a stable system and a good understanding of the initial elements.
"An increase in Share of Model proves that our content has generated revenue."
No. It's a visibility signal, not a revenue signal. It needs to be considered in conjunction with business metrics before drawing any real conclusions.
To remember
Manually testing in a ChatGPT tells you what the AI is responding to now. Monitoring tells you if your presence is changing, if you're on the right track.
Your goal today is no longer to spend hours stubbornly trying to ask the fateful question of your existence in an AI chat. Nor is it to search for a miracle tool (spoiler alert: there isn't really one on the market as of September 2026). What you need to start doing is building a system to manage your AI visibility. And that begins with creating test prompts that you will test regularly, at regular intervals, to identify a visibility trend.
You can then use the Share of Model for what it really is: a presence indicator, and not a new Google position in disguise.
And if your problem is not the absence of your brand but what search engines understand about it, first go back to the entities and signals that allow them to be disambiguated.
FAQ
How can I find out if my company appears in ChatGPT?
You can start by asking several questions representative of your market in ChatGPT and see if your company is mentioned. However, for reliable tracking, repeat the measurement over time and across multiple search engines. A single prompt provides a diagnosis, not a stable measure of visibility.
Is it possible to track your ChatGPT visibility for free?
Yes, on a small scale, using a spreadsheet and a fixed set of prompts. This method remains useful for understanding responses. However, it quickly becomes time-consuming as soon as you increase the number of search engines, locations, repetitions, and competitors. A monitoring tool can automate the data collection.
What is the best KPI to measure visibility in AI?
There is no single KPI or metric. Mention frequency or Share of Model measures brand presence; citations measure sources; Search Console now measures certain impressions within Google's generative features. These indicators should be interpreted together and considered in relation to business results.
Does Search Console allow for the measurement of ChatGPT?
No. Search Console covers the Google ecosystem. For ChatGPT, Perplexity, or Gemini, you need to use direct controls or specialized third-party tools.
How often should you check your mentions in ChatGPT?
The right frequency depends on your market and the number of actions taken. A monthly frequency is generally sufficient for a strategic dashboard. The key is to do it at regular intervals to avoid skewing the results.
Should you follow a position in ChatGPT?
No, there are no rankings in AI-generated content like there are in SEO on Google and other search engines. The order in which companies or sources are mentioned can vary between two results. A frequency of appearance within a stable corpus is much more defensible than an isolated "ChatGPT rank."
Background information used for this guide
- Google Search Central, June 3, 2026 https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
- Google Search Center https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Ahrefs, Brand Radar methodology, February 26, 2026 https://ahrefs.com/blog/brand-radar-methodology/
- Ahrefs Help, AI Visibility Metrics, June 26, 2026 —https://help.ahrefs.com/en/articles/15501968-ai-visibility-metrics
- Ahrefs Help, custom prompts, June 26, 2026 https://help.ahrefs.com/en/articles/13192745-how-to-set-up-custom-prompts-to-track-brand-visibility-in-ai-assistants
