How to Track Brand Mentions Across ChatGPT, Perplexity, Gemini and Google AI
Amit Malvi
Founder, SearchAxe. Former Senior SEO Manager, Invideo AI
Published Updated 9 min read
Quick answer
Keep one fixed list of buyer prompts, run it on every engine on a set schedule with several runs per prompt, and log each answer as one row in a shared sheet. That is how to track brand mentions across AI engines so one month compares fairly with the next. Read mention rates as monthly trends per engine, not single answers, and add Search Console's generative AI report and Bing Webmaster Tools' AI Performance report for Google and Microsoft surfaces.
On this page
- Why does tracking brand mentions in AI engines need a system?
- How does each engine surface brand mentions?
- How to track brand mentions across AI engines, step by step
- How many times should you run each prompt, and how often?
- What should the tracking spreadsheet look like?
- Should you use a tracking tool or check manually?
- How do you read AI mention trends without overreacting?
Key takeaways
- AI answers change between runs and between weeks, so a tracking system samples each prompt several times on a schedule instead of taking screenshots.
- Profound measured 40.5% to 59.3% of cited domains changing within one month depending on the engine, so read three-month trends.
- Log one row per run with engine, mode, prompt, mention, recommendation, competitors and cited URLs, then pivot to rates per engine.
- Only Google and Microsoft give site owners first-party AI reports, and both show impressions or citations, not brand mentions or clicks.
- A tool adds volume and daily sampling, but someone still has to read the answers and their sources each month.
To track brand mentions across AI engines you need a small system, not a habit of checking ChatGPT when a prospect mentions it. That system is a fixed prompt list, a sampling rule, a log with one row per answer and a monthly review. If you lead marketing at an AI startup and want one view of where you appear across ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, Google AI Mode and Copilot, this is how to set it up and keep it honest.
Why does tracking brand mentions in AI engines need a system?
Because the answer you saw this morning is one draw from a distribution. Run the same prompt again and the brands, their order and the sources change. A monitoring system turns those draws into rates you can compare month to month, and it separates real movement from noise.
Two studies show the size of the noise. In SparkToro and Gumshoe's repeated-run tests, reported by Search Engine Journal in January 2026, ChatGPT and Google's AI almost never returned the same brand list twice. Sources drift too, at different speeds per engine.
How does each engine surface brand mentions?
Each engine builds answers from a different mix of training data and live search, shows sources differently and offers different reporting to site owners. That is why a mention on one engine predicts little about another, and why the log needs an engine column on every row.
| Engine | Where answers come from | How sources appear | First-party report for site owners |
|---|---|---|---|
| ChatGPT | Training data, plus web search it may run automatically or when asked | Inline citations and a Sources panel when search is used | None; referral clicks carry utm_source=chatgpt.com |
| Perplexity | Real-time web search on every answer | Numbered citations on each answer | None |
| Claude | Training data, plus web search when it helps | Direct citations and source links when it searches | None |
| Gemini | Model knowledge, with web sources on some answers | A Sources button when sources exist; not every answer has one | None |
| Google AI Overviews | Google's index, using query fan-out | Supporting links inside the overview | Search Console generative AI report, impressions |
| Google AI Mode | Google's index, query fan-out, optional personalization from search history | Links throughout the response | Search Console generative AI report, impressions |
| Copilot | Bing's index | Linked citations | Bing Webmaster Tools AI Performance, citations |
The documentation behind that table: OpenAI's ChatGPT search help page and publishers FAQ, Perplexity's How does Perplexity work, Anthropic's web search help page, Google's Gemini sources help page, AI features documentation and AI Mode help page, and Microsoft's AI Performance announcement.
Find out which AI engines recommend you
A free consultation on where you appear today across AI engines and search.
Schedule a Free ConsultationHow to track brand mentions across AI engines, step by step
Set up cross-engine brand tracking
- Freeze one prompt list for every engineUse the same buyer prompts everywhere, grouped by intent, so engine-to-engine differences are real and not a wording effect.
- Fix the competitor listRecord the same four to six rivals on every row so share of voice can be calculated later without rereading answers.
- Standardise the sessionRun logged out or in a fresh session with memory and personalization off, from the same country, because ChatGPT and Google AI Mode both tailor answers to memory, location or search history.
- Set a sampling ruleDecide how many runs per prompt per engine and keep it constant, since changing the sample size changes the rate.
- Log every run as its own rowOne answer, one row, including runs where you are absent, because absences are the denominator of every rate.
- Pull the first-party reportsExport Search Console's generative AI report and Bing Webmaster Tools' AI Performance report on the same day as the prompt run.
- Review monthly, act quarterlyRead rates per engine and intent group each month, and change strategy only on trends that hold for three months.
The session step matters more than it looks. OpenAI's help page says that if memory is on, ChatGPT may use saved memories when rewriting a search query, and that it uses approximate location from your IP address. Google's AI Mode help page says AI Mode can personalize from your Search history. Your own logged-in answers are the least representative sample you can take. Building the prompt list itself is covered in how to measure AI search visibility for a SaaS company.
How many times should you run each prompt, and how often?
Run each prompt several times per engine in one sitting, once a month, and add a weekly spot check on your top shortlist prompts during a launch. More runs per sitting buy accuracy; more sittings buy trend data. For most seed to Series B teams the monthly run is the one that matters, because answers move too much week to week to read single weeks.
The same SISTRIX study gives the most useful frame for reading those numbers. In Google AI Mode, 86% of prompts had a stable core of one to five domains while the rest rotated at 89% a week, and for 43% of brand queries the brand's own domain was cited in all 17 weeks. ChatGPT had no stable core at all for the median prompt. Your goal is to move from the rotating edge of an answer into its stable core, and only repeated runs show which one you are in.
What should the tracking spreadsheet look like?
Use one raw tab with a row per run and one summary tab that pivots it into rates. Keep the raw tab boring and complete; every chart and board slide should be rebuildable from it.
| Column | What to record | Example |
|---|---|---|
| Run date | The day the prompt was run | 2026-10-05 |
| Engine | One of the seven engines | Perplexity |
| Mode | Search on or off, logged in or out, model if shown | Logged out, default |
| Prompt ID and group | Stable ID and intent group | P07, alternatives |
| Run number | Which repeat this is | 3 of 5 |
| Mentioned | Yes or no | Yes |
| Recommended | Yes if suggested for the stated need | No |
| Competitors named | From your fixed list, in order | Rival A, Rival C |
| Your URLs cited | Full URLs | https://yourproduct.com/compare/rival-a-alternative/ |
| Third-party URLs cited | Full URLs of reviews, roundups, threads, videos | A G2 category page, a Reddit thread |
| Accurate | Yes or no against your fact sheet, with a note | No, says no free plan |
| Sentiment | Positive, neutral or negative | Neutral |
The summary tab pivots that into mention rate, recommendation rate and citation rate per engine and per intent group, plus share of voice against your competitor list. Add two extra columns from the first-party reports: Google AI impressions from Search Console and Copilot citations from Bing Webmaster Tools. The formula and benchmarks for share of voice are in what share of voice in AI search is and how to calculate it.
Should you use a tracking tool or check manually?
Use both once you can afford a tool: the tool for volume and daily sampling, a person for the monthly read. A tool will tell you a mention rate dropped on Gemini; only reading the answers tells you it dropped because a new roundup now leads the citations.
Check coverage before you buy. Ahrefs Brand Radar lists ChatGPT, Perplexity, Gemini, Google AI Overviews, Google AI Mode and Copilot. Its Custom Prompts option also covers Claude. Semrush's AI Visibility Toolkit describes prompt tracking on ChatGPT and Google AI Mode. Our neutral comparison is in AI visibility tracking tools compared for startups, and if you want to start with no spend at all, use the one-hour manual check without paid tools.
How do you read AI mention trends without overreacting?
Read rates over three months, per engine and per intent group, and only act on a change that is larger than the month-to-month swing you have already seen for that prompt. A drop on one engine with no change elsewhere usually points at that engine's sources. A drop across every engine usually points at you: a pricing change, a renamed product or a page that broke.
- Annotate the sheet with your launches, site changes and major engine updates, so a jump has a likely cause next to it.
- When a competitor gains, open their cited URLs; the new source that put them there is usually visible in the answer.
- Separate brand prompts from shortlist prompts, because being described correctly when asked by name is a different win from being recommended.
- Check whether your own domain sits in the stable core or the rotating edge of answers before celebrating a single good month.
If the review keeps finding the same gaps, a one-off AI visibility audit traces each lost prompt to its cause. For ChatGPT, which sent 74.78% of AI referral traffic in SE Ranking's June 2026 study, the deeper method is in how to track brand mentions in ChatGPT.
Frequently asked questions
How often should you track brand mentions in AI search?
Run your full prompt list once a month, with several runs per prompt on each engine, and add weekly spot checks on your most important shortlist prompts during launches. Answers change quickly: SISTRIX measured ChatGPT replacing 74% of its sources each week and Google AI Mode 56%. Monthly runs smooth that out, and three-month trends are what you should act on.
Can Google Search Console track brand mentions in Google AI Overviews?
Not mentions, only links. Search Console's generative AI performance report, announced in June 2026 and available to all sites from August 31, 2026, shows impressions of your URLs in Google AI Overviews and Google AI Mode by page, country, device and date. It does not show answers that name your brand without linking to you, and it does not show clicks, so you still need a prompt log.
Do I need to log in to each AI engine to track mentions?
No, and logged-out or clean sessions are better for tracking. OpenAI says ChatGPT web search works for people who are not signed in, and that memory and approximate location can shape the queries it runs. Google says AI Mode can personalize answers from Search history. If you do use an account, turn memory and personalization off and keep the same country for every run.
Is a tracking tool more accurate than checking manually?
It is larger, not more privileged. Tools run more prompts more often, which makes rates steadier, but Google states that no third-party tool has access to its internal ranking or AI systems, and engine coverage differs by vendor. A manual check reads fewer answers but reads them closely, which is how you spot misdescriptions and the sources behind a competitor's gain. Most teams end up using both.
