How to check if ChatGPT mentions your brand, and what AI visibility tracking costs

To check if ChatGPT mentions your brand, open a temporary chat (or use ChatGPT signed out) and ask 5 to 10 questions a buyer would ask, without your company name in them. Note which businesses get named and which websites get cited. Run each question with web search on and off, and repeat every question at least 3 times, because answers can differ from one run to the next. A first look is about 10 chats and costs nothing. Paid AI visibility trackers run the same kind of prompts on a schedule across several AI assistants. Their real cost is the subscription plus the time you spend building the prompt list and reading the results.

Your options at a glance

Most small businesses should start with the manual check and the GA4 setup below. Both are free, and they tell you whether paying for more makes sense.

A first-look manual check

Set up a clean session

Your own ChatGPT account knows you. Memory, custom instructions and earlier chats can push answers toward brands you've talked about, including your own. A clean session removes that bias.

• Use a temporary chat, or open a private browser window and use ChatGPT signed out.
• Start a new chat for every prompt. Follow-up questions carry the context of the earlier ones.
• Write down the date, the model ChatGPT shows, whether you were signed in and which city you were in.
• Test the way your buyers use it. If most of them are on the free version and never sign in, a paid, signed-in account with memory on is the wrong test.

The prompt template

Use unbranded prompts for the main check. If your name is in the question, ChatGPT will talk about you, so it tells you nothing about whether you get mentioned on your own. The example column uses a made-up bookkeeping firm that works with restaurants.

Take the wording from real sources: the queries in your Google Search Console Performance report, questions people ask on sales calls and in support emails, and the People Also Ask box in Google. Buyers often word things differently from your homepage.

Keep 2 or 3 branded prompts in a separate list, such as "What do you know about [your brand]?" and "Is [your brand] a good choice for [use]?". These don't measure visibility. They check whether the facts ChatGPT states about you (prices, locations, services, founding year) are right.

For a first look, pick one prompt from each of the first five types and run each once with search and once without. That's 10 chats. How long it takes depends on how long the answers are and how many sources they cite, so time your first two or three chats and multiply before you plan a bigger run.

Run each prompt with and without web search

ChatGPT can search the web before it answers, and answers that used search show the sources they drew on. OpenAI's help page on ChatGPT search explains the feature; check it for the current controls, because they change.

Without search, the answer comes from what the model learned in training. If you're missing there, the fix is slow: being written about across the web over months. With search, the answer is built from pages ChatGPT finds today, and the cited links show you which pages it trusts. If you're missing there, the fix can be a specific page, a listing or an article on a site that already gets cited.

To ask for a search, turn on the search option in the chat box if you see one, or add "search the web" to the prompt. To keep search out, start the prompt with "Answer from what you know, without searching the web." Either way, check whether the answer shows sources before you log which mode it was.

What to write down

A spreadsheet is enough. Use one row per answer with these columns:

• Date, prompt ID and the exact prompt text
• Search on or off, signed in or out, model, city
• Run number (1, 2, 3)
• Brands named, in the order they appear
• Your position, or "not named"
• Cited URLs, and whether any of them is on your site
• Anything it said about you that's wrong

Give every prompt a fixed ID and never edit its wording later. If you change a prompt, add it as a new one. Otherwise next month's numbers can't be compared with this month's.

Why one check isn't enough

The same question in a new chat can return a different list of businesses. One screenshot of "we're in it" or "we're not" is a single answer, and you don't know yet how typical it is. These are the settings that can change what you see:

• The run itself. Log more than one run per prompt so a single answer doesn't decide your conclusion.
• Search on or off. The two modes pull from different sources, as described above.
• Location. A "near me" or city prompt depends on where ChatGPT thinks you are.
• Account state. Signed in or out, free or paid, memory on or off, custom instructions.
• Time. OpenAI updates the models behind ChatGPT, so don't assume an answer from one month still holds the next. Rerun the same prompts.
• Wording. "Best bookkeeper for restaurants" and "restaurant bookkeeping service" can return different names.

For US local businesses, location is the easiest one to get wrong. If you serve the Phoenix metro area, run local prompts from there, or write down where you ran them and read the results with that in mind. If you use a VPN to test another city, pair it with a signed-out session so your account history doesn't mix in.

How many runs you need

Say your brand shows up in about half of the answers for a set of prompts. The table shows how far off that 50% could be at different sample sizes. These margins are our own calculation, not figures from a study, using the standard 95% margin of error at a 50% rate (the widest case):

Margin of error = 1.96 × √(0.25 ÷ n), where n is the number of answers you logged.

The formula assumes every answer is an independent draw at the same rate. Real answers aren't quite like that: runs of one prompt can resemble each other, and pooling different prompts mixes different rates. Treat the margins as a lower bound. Your true uncertainty is likely a bit wider.

In practice: 3 runs per prompt is fine for a first look. For a baseline you'll compare against later, pool the answers. Ten prompts with 3 runs each gives 30 answers per search mode, or 60 chats with search on and off. Repeat the same set monthly to see a trend.

The numbers to track

• Mention rate: the share of answers that name you. Named in 9 of 30 answers is 30%.
• Top-3 rate: the share of answers where you're among the first three names, the ones a reader sees first.
• Own-site citation rate: the share of search-on answers that link to a page on your site.
• Wrong facts: a running list of errors from the branded prompts, with the date you first saw each one.

Report branded and unbranded prompts separately. Mixing them makes your mention rate look higher than it is.

Tracking AI referral traffic in GA4

A mention is not a visit. When someone clicks a link in an AI answer and the browser passes on where they came from, Google Analytics 4 records the AI site (such as chatgpt.com or perplexity.ai) as the session source.

Start by looking. Open Reports, Acquisition, Traffic acquisition, set the dimension to Session source and search for AI domains. That alone tells you whether AI answers send anyone. Then check the default channel group in your property and see whether any channel already catches those domains. Google's help page on the default channel group is the reference for how the built-in channels are defined. If nothing catches them, build your own channel.

Build a custom channel group

1. In GA4, go to Admin and open Channel groups (under Data display in current versions).
2. Create a new channel group. It starts with the default channels.
3. Add a new channel and call it "AI assistants".
4. Set the condition to Source, matches regex, and paste: .*(chatgpt\.com|chat\.openai\.com|perplexity\.ai|copilot\.microsoft\.com|gemini\.google\.com|claude\.ai).*
5. Place the new channel above Referral.
6. Save, then open Traffic acquisition and switch the table's primary dimension to your new channel group.

Test it before you rely on it. If chatgpt.com sessions still show up under Referral in the new group, move your AI channel higher. Add any AI domains you found in Session source that the regex misses. If your screens don't match these steps, Google's help page on custom channel groups is the place to check.

What GA4 can't show you

• Visits with no referrer. A click that arrives without source information, which can happen from apps, is counted as Direct.
• People who read your name and then search for you. They show up as branded Google searches or as Direct. Watch branded impressions in Search Console as a side signal.
• Which question was asked. A referral tells you the visit came from an AI site, not what the person asked it.

For Google's own AI answers, see our guide to showing up in Google AI Overviews.

Expect small numbers at first. Judge the trend over months, not days.

What paid AI visibility trackers measure

A tracker automates the manual check: it sends a fixed list of prompts to one or more AI assistants on a schedule and stores every answer. Most report some mix of these:

• Mention rate per prompt, per topic and overall
• Position: where in the answer you're named
• Share of voice: your mentions against the competitors you choose to watch. If you're named in 7 of 20 answers and your main rival in 14, the rival has twice your share
• Cited sources: which domains and pages the assistants link to. This is often the most useful part, because it shows which review sites, directories and articles you need to be on
• Sentiment or description: how you're described when you are named
• Segments by AI assistant, country, city or language
• History and alerts when a number moves

Ask how the tool gets its answers

Before you pay, ask the vendor these questions. The answers decide whether the numbers match what your buyers see.

• Do you query the consumer app or the API? API answers can differ from what a person sees in the ChatGPT app.
• Signed in or signed out? With memory or without?
• With web search on, off or both?
• From which country or city, and can I choose?
• How many times do you run each prompt per check? If the answer is once, use the sample-size table above to judge how much each number can move by chance.
• Can I export my full answer history if I leave?

We don't know how each vendor answers these, and many don't publish it. If a vendor can't tell you, treat its numbers as a trend line, not a measure of what buyers see.

How trackers charge, and what's left out

The pricing levers

Tracker prices are typically built from these levers, and an entry-level plan often limits several at once:

• Number of prompts tracked
• Number of AI assistants (ChatGPT, Perplexity, Gemini, Claude, Copilot, Google's AI Overviews), some included and some sold as add-ons
• How often prompts run: daily, weekly or monthly
• Runs per prompt per check
• Seats for team members or clients
• Projects, brands, countries or languages
• Billing period: monthly or yearly

We're not listing vendor prices on this page. They change often and we couldn't confirm current figures from the vendors' own pricing pages for this article. Use the formula below on any pricing page instead.

Work out your real cost per answer

Cost per answer = monthly price ÷ (prompts × AI assistants × runs per prompt × checks per month)

An example with made-up numbers, not any vendor's price: a plan at $100 a month covers 50 prompts on 3 assistants, one run each, checked weekly. That's 50 × 3 × 1 × 4 = 600 answers a month, or about $0.17 per answer. If you need 150 prompts to cover your categories, cities and competitors, you're on a bigger tier. If the assistant your buyers use most is an add-on, add that too. Run the math on the setup you actually need, not the one the cheapest plan allows.

Costs that aren't on the pricing page

• Building the prompt set. Writing 50 to 100 good prompts from Search Console data, sales questions and competitor lists is a job of its own. The tracker won't know your buyers' wording.
• Reading the results. Someone has to look at the dashboard every week or month and decide what it means. If nobody does, the subscription buys nothing.
• Acting on it. A tracker shows gaps. It doesn't write the page, fix the wrong fact or get you listed on the site that keeps getting cited.
• Switching. A new tool may sample differently, so expect to start a new baseline. Check the export question above before you buy.
• Yearly billing. If you pay for a year to get a lower rate, you're committed for that year.
• Extra seats when a colleague or an outside marketer needs access.

Which option fits which company size

• Single-location local business (plumber, dentist, salon): a manual check once a month from your own metro area, plus the GA4 channel group. A handful of local prompts covers most of what buyers ask, so a standalone tracker is rarely worth it.
• Small online store: manual baseline and GA4 first. Pay for tracking only if AI referral sessions grow or your category is crowded. GA4 tells you whether AI answers send buyers at all before you spend on measuring mentions.
• B2B company or service firm with a long sales cycle: tracking is usually worth it, especially for comparison and alternatives prompts. Buyers research for weeks, and being left off a shortlist costs a whole deal.
• Multi-location or multi-country brand: paid tracking with location segments. Answers can vary by city and country, and manual checks from one place can't cover that.

Who shouldn't pay: if you can't name the person who will read the results each month and act on them, don't buy a tracker yet. A monthly manual check and the GA4 report will tell you as much.

What to do with the results

You're never named

Check the cited sources in search-on answers. They're the pages ChatGPT trusts for your category: review sites, directories, comparison articles, local guides. Getting onto those pages, and having clear pages on your own site that answer the same prompts, is where to start. Our guide on how to get your business recommended by ChatGPT walks through it step by step.

You're named sometimes

Compare prompt types in your log. If you're named for "best X for Y" prompts but not for price or problem prompts, check whether you have a page that answers those questions directly. If you don't, write it.

You're named with wrong facts

Find where the wrong fact comes from. Search for it, check old directory listings, your Google Business Profile, and outdated pages on your own site. Fix it at the source, then rerun the branded prompts after a few weeks. Clear structured data on your pages helps too; the free FoundWell plugins for WordPress and Umbraco add schema, generate an llms.txt file and audit how likely each page is to be quoted by an AI assistant.

Your site is cited but you're not recommended

ChatGPT is using your content as a source but naming others as the answer. That usually means your page explains the topic well but doesn't make the case for your business: no clear statement of who you serve, where, at what price and why. Add those facts in plain sentences an AI can quote.

Competitors own the answer

Read what the AI says about them. The reasons it gives ("known for fast turnaround", "serves restaurants specifically") tell you which claims you need to make, with evidence, on your own pages and on the sites it cites.

Where FoundWell fits

FoundWell includes AI answer tracking in every plan, next to the work that responds to it. Starter ($250 a month) tracks AI answers monthly. Growth ($500 a month) and Complete ($1,000 a month) track them weekly. On Complete, tracking covers ChatGPT, Perplexity, Claude and Google's AI answers, and Complete adds pages for the gaps it finds. If you're looking at Starter or Growth and need a specific assistant covered, check the plan details before you sign up.

Every plan also writes articles, fixes pages, adds schema and llms.txt, and syncs Search Console and checks indexing every day, inside your own WordPress, Umbraco, Shopify or Wix site. A person on our team reviews every article before it goes live, and in approval mode nothing is published until you approve it.

We don't promise that ChatGPT or any other assistant will mention you. The 90-day guarantee on Complete is about Google: 25% more clicks from Google in 90 days, or we keep working for free until you get them, under the conditions listed on our site (for example, at least 90 days of Search Console data when you connect). Plans are month to month with no contract.

See how this compares with an agency, a freelancer or doing it yourself on the comparison page, or read what each plan includes on the FoundWell homepage. When you're ready, connect your site and start. It takes about ten minutes.

Frequently asked questions

Can I just ask ChatGPT whether it mentions my brand?

Not usefully. ChatGPT can't see other people's conversations, so it can't tell you how often it names you. And if you put your brand name in the question, it will talk about you, which says nothing about whether it names you unprompted. Ask unbranded buyer questions instead and count how often you appear.

How often should I check what ChatGPT says about my business?

For a small local business, once a month with the same prompt list is enough. If you're actively publishing pages to close gaps, check weekly or use a tracker so you can see whether the changes show up. Always use the same prompts, settings and location so the results are comparable.

Does ChatGPT show everyone the same answer?

No. The same prompt in a new chat can return a different list, and results also depend on whether search is on, where you are, your account and memory settings, and which model is running. Log several runs per prompt before drawing conclusions: with 10 answers, a 50% mention rate could really be anywhere from about 20% to 80%.

How do I see ChatGPT traffic in Google Analytics 4?

Open Reports, Acquisition, Traffic acquisition and set the dimension to Session source, then look for AI domains such as chatgpt.com or perplexity.ai. To group all AI assistants together, create a custom channel group in Admin, add an AI assistants channel that matches their domains, place it above Referral and check that the sessions land in it.

Is a paid AI visibility tracker worth it?

It depends on whether someone will act on the results. Single-location local businesses usually get enough from a monthly manual check and GA4. B2B firms with long sales cycles and brands in many cities or countries get more value, because manual checks can't cover enough prompts and locations.

Does FoundWell guarantee that ChatGPT will mention my business?

No. FoundWell tracks AI answers on every plan and, on Complete, creates pages to close gaps, but it doesn't promise rankings or AI mentions. The 90-day guarantee on the Complete plan is about clicks from Google, measured in Google Search Console, under the conditions published on the site.

About the author

Peter Sinke, Founder

Peter Sinke builds websites at Datapad in the Netherlands and made FoundWell from the daily AI SEO routine he ran for Datapad's own site.