How to Track Brand Mentions in AI Search (Manual to Tools)

By Yahav Fuchs •

Short answer: start with a spreadsheet of 30 buyer prompts, run each one several times, and record mention, position and sources. Add GA4 or server logs to catch visits that come from AI engines. Move to a dedicated tool when the manual run takes more time than the decisions it informs.

There is no single place that reports “your brand was mentioned in ChatGPT N times.” The engines do not publish a dashboard for it. Every method here is an approximation from a different angle, and the useful skill is knowing what each one can and cannot see.

  • Manual prompt tracking tells you about answers. It is noisy, slow and free.
  • Analytics tells you about visits that came from AI engines. It misses most of the influence.
  • Server logs tell you which AI crawlers fetched your pages.
  • Dedicated tools automate the prompt runs at scale, with repeats and competitor comparison.

Decide what you are measuring first

Before picking a method, separate four things that get mixed together:

MeasureWhat it isWhere you can see it
MentionYour brand named in the generated answerPrompt runs
CitationA link to your page or a page about youPrompt runs, source lists
Referral visitA person clicks through and lands on your siteAnalytics
CrawlAn AI crawler fetches your pageServer logs

They are related but not the same. You can be mentioned and never cited. You can be cited and get no clicks, because the answer was enough. And the influence you care about most, the buyer who sees your name in an answer and later searches for you directly, shows up as none of the four. It appears as direct or branded search traffic later.

Method 1: A manual prompt spreadsheet

This is where I would start, always. It is free, it teaches you how the engines talk about your category, and it keeps you honest about noise.

Build the prompt list. Aim for 30 to 50 prompts. Mix category prompts with no brand named, comparison prompts, problem prompts and a few branded ones. Draw the wording from sales calls, support tickets and community threads, not from keyword tools. Real buyers write longer and messier prompts than keywords. More on this in how to choose prompts to track.

Pick the engines. ChatGPT, Perplexity, Gemini and Google’s AI features are the usual four. Add Claude, Copilot, Grok or Meta AI if your buyers use them. Use a clean session, logged out or in a fresh browser profile where possible, so your history does not bias the answer.

Run each prompt several times. Three runs per prompt per engine is a workable minimum. Answers differ, and one run can mislead you in either direction.

Log the same columns every time:

  • Date, engine, prompt, run number
  • Mentioned (yes/no), position (first, in a list, afterthought)
  • Sentiment or framing in one phrase
  • Competitors mentioned
  • Source URLs cited

Limits. At 40 prompts, 4 engines and 3 runs, that is 480 answers per cycle. Nobody keeps that up weekly by hand. Manual works for a baseline and for spot checks after a change. It does not work as ongoing reporting.

Method 2: Look at the sources, not just the answers

The most useful column in that spreadsheet is the sources. If the same five domains keep being cited across your category prompts, that is where you need to appear. Sort by frequency and look for the pattern:

  • Independent roundups and comparison articles
  • Review sites and directories
  • Reddit threads, YouTube videos or LinkedIn posts
  • Vendor documentation and competitor pages

This turns tracking into a to-do list. If a Reddit thread titled “best X for small teams” is cited in a third of your category runs, you know where a useful answer would matter. Reddit monitoring can surface those threads continuously, and our post on AI competitive analysis for brand mentions shows how to compare yourself with named competitors at the source level.

Method 3: Analytics for referral visits

When someone clicks a link in an AI answer, many engines send a referrer such as chatgpt.com or perplexity.ai, and your analytics can record it. In GA4, Google added a native “AI Assistant” channel to the default channel group in May 2026, according to industry reports, and a custom channel group is the common workaround for traffic it misses. The setup is in how to measure AI referral traffic in GA4.

What analytics can tell you: which AI engines send visits, which landing pages they land on, and whether those visitors convert.

What it cannot tell you: how often you were mentioned without a click, what the answer said, or who you lost to. Referrers are also sometimes stripped, so some AI traffic lands in Direct. Treat referral numbers as a floor, never as total AI influence.

Method 4: Server logs and crawler hits

Your web server logs record which bots fetched which pages. The AI-related user agents documented by their operators include OpenAI’s OAI-SearchBot, GPTBot and ChatGPT-User, and Perplexity’s PerplexityBot and Perplexity-User. Google’s AI features use Google’s regular crawling.

Logs answer a narrower question: can the engine read this page at all, and how recently did it try? If a page you want cited has never been fetched by a search crawler, no amount of content polish will matter until you fix access. A fetch does not mean a mention, though. It only tells you the page was available.

Method 5: Dedicated AI visibility tools

Tools in this category run your prompt list on a schedule across engines, repeat each prompt, extract mentions and citations from the answers, and chart the results with competitors alongside. What they add over a spreadsheet:

  • Scale and repeats. Hundreds of prompts, several runs, daily or weekly.
  • Trend lines. The comparison over time is the whole point.
  • Competitor share of voice. Who gets mentioned when you do not.
  • Source-level detail. Which exact pages and threads are being cited.

What to check before buying any of them:

  1. How is the answer collected? Some tools query the engines’ APIs, which can differ from what a person sees in the product. Others capture the consumer interface. Ask which, because the results can differ.
  2. How many repeats per prompt? One run per prompt is not a measurement.
  3. Which engines are covered, and do they report each separately?
  4. Can you bring your own prompts, or are you stuck with generated ones?
  5. Do the metrics have definitions you can read? Our AI visibility score explainer lists the ones to ask about.

We reviewed several in best LLM SEO rank trackers 2026. ReddGrow is one option in this space, with a focus on the Reddit threads that engines cite: it tracks prompts, share of voice and citations, and links them to the Reddit conversations behind them. See AI brand visibility for how that works. It is built for brands that treat Reddit as a channel, and it is less useful if you only want a generic score.

How to combine them

For a small team, this stack is enough:

  1. Monthly: a manual run of 30 prompts across three or four engines, three runs each. About two hours.
  2. Continuous: GA4 with an AI channel so you see visits and conversions.
  3. Quarterly: a look at server logs to confirm search crawlers can reach your key pages.
  4. When it hurts: a dedicated tool once the manual cycle is eating more time than it is worth.

For an agency or a company with several products, skip straight to a tool and use the spreadsheet only to sanity check it.

Mistakes I see repeatedly

Tracking only branded prompts. If you only ask “what is [Brand],” you will see yourself everywhere and learn nothing. The category prompts, where you are not named, are where visibility is won. See branded vs unbranded prompts.

Reading one run as truth. The same prompt can return a different list five minutes later.

Ignoring personalization. Your own logged-in session is biased toward you. Use clean sessions.

Counting mentions as wins without reading them. A mention that says your product is “limited” or “pricey” is not the same as a recommendation. Record the framing.

Reporting a single blended number. Engines behave differently. Report them separately and combine only for the headline.

Where to start today

Open a spreadsheet, write 30 prompts, run them in three engines, and log the sources. In an afternoon you will know whether you are visible, who is, and which pages and threads the engines trust in your category. Everything after that is a way of repeating that exercise faster and more reliably.

Frequently Asked Questions

Can I track brand mentions in ChatGPT for free?
Yes, at small scale. Write 20 to 30 buyer prompts, run each several times in the engines you care about, and log whether you are mentioned and which sources are cited. It is slow and noisy, but it is enough to learn your baseline before paying for anything.
Why do I get different answers every time I ask the same prompt?
These engines generate answers probabilistically, and many also run live searches whose results change. Personalization, location and conversation history can also shift the answer. That is why a single run proves nothing and why serious tracking repeats each prompt.
Does Google Search Console show AI mentions?
Only partly. Google says traffic from AI features such as AI Overviews and AI Mode is included in the Web search type in Search Console's Performance report, but it is not broken out separately for you to filter on. It tells you about clicks and impressions, not about whether your brand was mentioned in an answer.
What is the difference between a mention and a citation?
A mention is your brand name appearing in the generated text. A citation is a link to your page, or to a page about you, in the sources. You can be mentioned without being cited and cited without being named, so track both.
Do I need a paid tool to track AI search mentions?
Not at the start. A paid tool earns its cost when you need more prompts than you can run by hand, daily or weekly repeats, several engines, competitor comparison and source-level detail. If you are tracking fewer than 30 prompts once a month, a spreadsheet is fine.