Track brand mentions in ChatGPT: AI visibility tracking guide

To track brand mentions in ChatGPT, build a fixed set of real buyer prompts, run each one several times on a regular schedule, and log whether your brand is named, where it appears in the answer, which competitors are named instead, and which sources are cited. Over a few weeks that gives you a mention rate and a share of voice you can actually improve. Classic rank trackers cannot do this, because AI answers have no fixed positions to track.

• AI visibility tracking measures answers, not rankings: are you named, and who is named instead?
• Start with 10 to 50 buyer prompts across "best X for Y", "X vs Y", alternatives, and how-to questions.
• Answers vary between runs, so run each prompt multiple times and look at rates, not single screenshots.
• The cited domains are your to-do list: publish the missing page and get listed where the engine already looks.

Why doesn't rank tracking cover AI answers?

Rank trackers assume a results page with stable positions: you are #3 for a keyword today, #5 tomorrow. ChatGPT does not work that way. It writes a new answer each time, the list of tools it names can change between two identical prompts, and many answers include no links at all. There is also no official equivalent of Google Search Console for ChatGPT that tells you how often your brand appeared, at least at the time of writing.

Three gaps follow from that:

• Visibility without clicks. A buyer can read "the best options are A, B, and C", never click anything, and shortlist you or skip you. Analytics never sees it.
• Ranking does not equal mention. You can rank #1 on Google for a query and still be absent from the ChatGPT answer to the same question.
• Competitors are invisible too. Without tracking, you do not know who is being recommended in your place.

AI visibility tracking (sometimes called LLM brand monitoring) fills that gap by sampling answers directly.

What exactly should you track?

For every prompt and every run, capture:

• Named: does the answer mention your brand, yes or no?
• Position: if named, where? First recommendation, somewhere in a list, or a passing mention at the end.
• Competitors named: which other brands appear, in order.
• Cited sources: which URLs or domains are linked, and whether any are yours.
• Accuracy: is what it says about you correct (pricing, features, audience)?

How do I build a buyer-prompt set?

Your prompt set is the foundation, and it should stay stable so results are comparable week to week. Aim for 10 to 50 prompts, spread across the question types that lead to a purchase decision.

Rules for good prompts

• Mostly unbranded. "Best X for Y" tells you whether ChatGPT thinks of you unprompted. Keep branded prompts to a small slice.
• Phrase them like buyers. Take wording from sales calls, support tickets, and Search Console queries, not from your positioning doc.
• Include the audience. "For Webflow agencies" or "for a 10-person SaaS" changes answers a lot and matches how people really ask.
• Do not lead the model. "Is Acme the best tool for..." inflates your numbers and teaches you nothing.
• Freeze the wording. Changing a prompt resets its history. Add new prompts rather than editing old ones.

How do I track ChatGPT mentions manually?

A spreadsheet is enough to start. One row per prompt per run, with these columns:

date,engine,prompt,run,named,position,competitors_named,cited_domains,notes
2026-11-05,ChatGPT,best invoicing tool for freelance designers,1,no,,"Competitor A; Competitor B",reviewsite.com; forum.com,
2026-11-05,ChatGPT,best invoicing tool for freelance designers,2,yes,3,"Competitor A; Competitor B",reviewsite.com,
2026-11-05,ChatGPT,best invoicing tool for freelance designers,3,no,,"Competitor A",blog.example.com,

1. Use a clean session. Turn off memory and custom instructions, or use a temporary chat, so your own history does not shape the answer.
2. Run each prompt 3 times in separate new chats. One run is an anecdote; three gives you a rough rate.
3. Log the answer as it is, including competitors and every cited domain. Paste the full answer into a notes column or a linked doc if you can.
4. Repeat weekly on the same day, with the same prompts.
5. Note context: which model was selected, whether search was used, and any major product update you shipped that week.

Twenty prompts times three runs is sixty answers a week. That is manageable for a month or two, and quickly becomes tedious beyond that, which is the point where most teams automate. Run the same set in Perplexity and Gemini too, since each engine retrieves differently (see Perplexity SEO).

Which metrics matter for AI visibility?

Break each metric down by prompt category. It is common to have a decent mention rate on how-to prompts and almost none on "best X for Y", which points to a very different fix than the reverse.

Which competitors should you track?

Share of voice only means something if the competitor list is right. Pick three to six brands, and let the data correct you:

• The ones you lose deals to. Ask sales which names come up most often.
• The ones AI names most. After the first baseline run, count which brands appear most often across all prompts. They are your real AI competitors, even if you never considered them rivals.
• One aspirational leader. The category default gives you a ceiling to compare against.

Revisit the list every quarter. New entrants often show up in AI answers before they show up in your pipeline, which makes tracking a useful early-warning signal.

How often should you track, and what should the report show?

Weekly is the right cadence for most teams. Daily data is mostly noise unless you are running an experiment, and monthly is too slow to connect changes to results. A useful weekly report fits on one screen:

1. Overall mention rate this week vs the four-week average, per engine.
2. Share of voice against each tracked competitor.
3. Prompts that changed status: newly named, newly dropped, or a competitor newly appearing.
4. The top ten cited domains, with any new ones flagged.
5. Inaccurate statements about your brand that need correcting.
6. What you shipped this week, so later readers can connect actions to movement.

That last item is easy to skip and is the one that makes the report useful months later. When a prompt flips from "not named" to "named", you want to know whether it followed a new comparison page, a listicle update, or just a model change.

What makes AI visibility data unreliable?

• Personalization. Memory, custom instructions, and previous messages in the chat all influence answers. Test from clean sessions.
• Model and version changes. OpenAI updates models and the default model changes over time. A sudden jump or drop across every prompt is often a model change, not something you did.
• Location and language. Search-backed answers can differ by country. Test from the market you sell into, and note it.
• Search on or off. Answers that use web search cite sources and react quickly to new content; answers from training data do not. Record which one you got.
• Small samples. With one run per prompt, normal variance looks like a trend. Look at multi-week averages before reacting.
• Wording drift. Editing prompts mid-stream breaks comparisons. Version your prompt list.

How do I act on gaps once I find them?

Tracking is only useful if it changes what you publish. For every prompt where you are not named, work through this loop:

1. Look at the cited domains. They show you what kind of page the engine trusts for this question: a listicle, a comparison, a forum thread, a docs page.
2. Write a citation brief. Define the page you need: target prompt, title, a two or three sentence direct answer, FAQs, a comparison outline if relevant, and the schema to add.
3. Publish or fix the page on your own site, with the answer near the top in plain HTML. Make sure AI crawlers can actually read it with the free AI crawler checker.
4. Get listed on the cited domains. Contact the authors of the listicles and comparison posts that keep appearing. Offer accurate, current information and a genuine reason you belong. Correct wrong facts about you.
5. Re-check the prompt over the following weeks. Search-backed answers can shift within days of indexing; training-based answers take longer.

The on-page side of this is covered in how to get cited by ChatGPT, and you can score any page's citability with the free SEO and AEO score. Google's AI answers need their own check too; see Google AI Overviews optimization.

What is a realistic AI visibility goal?

Do not aim for 100%. Even category leaders are not named in every run of every prompt, and some prompts will always favor a specific type of product that is not you. A more useful goal is relative: raise your mention rate on the prompts that matter most, close the gap with one or two direct competitors, and get at least one of your own pages into the cited sources for each high-value prompt. Set the target after four weeks of baseline data, not before, so it reflects normal variance rather than one lucky or unlucky week.

How does SEOSync automate AI visibility tracking?

SEOSync's AI visibility tracking does the spreadsheet work for you. It asks ChatGPT, Gemini, and Perplexity your real buyer prompts and reports whether your brand is named, which competitors are named instead, and which sources are cited. Pro tracks 25 prompts; Agency tracks 100 prompts per client site.

When engines skip your brand for a prompt, SEOSync drafts a citation brief: the title, TL;DR, FAQs, comparison outline, schema hint, and outreach targets for the page you should publish. Pair that with the SEO and AEO audit on the page itself, apply fixes to WordPress, Webflow, or Shopify, and re-score the URL to confirm the change. Plans and the 14-day trial are on the pricing page.

Want to see what this looks like in practice? Our public experiment tracks whether AI engines recommend SEOSync itself for the questions our buyers ask, and logs every change we ship to try to move that number, starting from an unedited baseline.

What are the most common tracking mistakes?

• Testing from your own logged-in account with months of chat history about your product.
• Tracking only branded prompts, which almost always look good.
• Reporting one run per prompt as if it were a stable result.
• Ignoring cited domains, which are the most actionable part of the data.
• Tracking but never publishing anything in response.

FAQ

How do I know if ChatGPT mentions my brand?

Ask it the questions your buyers ask, not branded ones. Build a list of prompts like "best X for Y" and "X alternatives", run each one several times in clean sessions with memory off, and log whether your brand is named, which competitors appear, and which sources are cited. Repeat weekly so you see a rate rather than a one-off result.

What is AI visibility tracking?

AI visibility tracking, also called LLM brand monitoring, measures how often AI answer engines like ChatGPT, Gemini, and Perplexity name your brand for relevant questions. Instead of keyword rankings, it reports mention rate, share of voice against competitors, citation rate, and which domains the engines cite, so you know where you are missing and why.

How many prompts should I track?

Between 10 and 50 is a practical range for most companies. Cover each stage of the buying journey: best-for prompts, head-to-head comparisons, alternatives, how-to questions, and a few branded fact checks. Fewer than 10 gives a noisy picture; more than 50 is hard to review manually, which is when automated tracking becomes worth it.

Why does ChatGPT give different answers to the same prompt?

AI models generate each answer fresh and include some randomness, and answers can change with web search results, the selected model, location, and personalization such as memory or custom instructions. That is why you should run each prompt multiple times from clean sessions and judge visibility by rates over several weeks, not by a single response.

Can Google Analytics track ChatGPT brand mentions?

Only partly. Analytics can show visits referred from chatgpt.com or perplexity.ai when someone clicks a cited link, but it cannot see answers where your brand was named without a click, or answers where a competitor was recommended instead. You need to sample the answers directly, manually or with a tracking tool, to measure visibility.

See if ChatGPT names your brand

Track buyer prompts across ChatGPT, Gemini, and Perplexity, and get a brief for every gap.