How to track brand mentions in ChatGPT

The SEO objection

“Surely this is just SEO” is a reasonable objection. If your site has useful pages, clear language and sound technical access, you have already done work that can help an AI assistant understand your business.

But SEO reports do not tell you whether ChatGPT named your brand in an answer. They do not show which buyer question triggered the mention, which sources appeared beside it, or whether the mention held across repeated checks. To track brand mentions in ChatGPT, you need to inspect the answers themselves.

That leaves you with three practical routes: check answers by hand, collect them with an API script, or use a monitoring tool. The choice is mainly about the cost of repeatable evidence.

Cost of each method

The hand method costs money only if you count staff time. You write a fixed set of buyer questions, submit them to ChatGPT, save the answers and record whether your brand appeared. It is simple, visible and useful for a first baseline.

The API method replaces the browser with code. Your script sends the same questions, stores the returned text and marks mentions using rules you control. You still need to decide what counts as a mention, how to handle a brand variation and how to review doubtful cases.

A monitoring tool packages the repeated work. Depending on the product, it may organise prompts, answers, competitors and citations in one place. We don't have current API or monitoring-tool prices to publish here, so get quotes directly before you decide.

The cheapest route depends on how often you need an answer and how much review you require. A spreadsheet can be cheap and still become expensive when someone has to keep it alive.

Question set

All three methods need the same sampling rule. Pick a stable set of buyer questions, run the same set across the period you are comparing, and record the engine, date, answer, brand status and cited sources. Do not change the questions halfway through and then treat the result as a trend.

The question set should reflect how people shop, compare and reject products. Include category questions, problem questions, comparison questions and questions that name alternatives. The exact mix depends on your market, so there is no single prompt list that fits every brand.

Keep the distinction between a mention and a citation. A response can name your product without linking to your site. It can also cite a page without making your brand prominent in the prose. Record both outcomes.

CiteGraph's measurement method describes the same basic discipline: count answers and sources, then keep the engine results separate. In each scan, CiteGraph tracks whether one specific brand gets named. In its ChatGPT scan data, 2,434 answers were sampled, 3.5 sources were cited per answer on average and that brand was named in 6.1% of answers.

The figure shows ChatGPT naming brands less often than the other three engines, a gap worth planning around rather than explaining away.

The sample has limits. We measure sampled answers, not every answer a buyer could receive, and we do not know what we have not measured. Randomness, question wording, account context and answer changes can all affect an individual check.

Manual checks

Manual tracking is the right first move when you need to understand the shape of the problem. Start with a short, stable question set, run each question in ChatGPT, and save the complete answer rather than only the line containing your brand.

For every result, record five things: the question, whether the brand appeared, the wording used, the sources shown and the page or domain cited. Add a note when the answer makes a factual mistake. A mention that misstates your pricing or audience is not the same outcome as a useful recommendation.

Use the free AI visibility checker for a quick starting point, then keep your own record if the result will guide content or budget. The checker can help you find an initial signal; a fixed sample gives you something you can compare later.

Manual work has a useful advantage. You see context that a simple match may miss.

ChatGPT could mention a brand in a list of products to avoid, or refer to a parent company when you are searching for a product name. A person can flag that.

A script needs an explicit rule.

The weakness is drift. Different people ask slightly different questions, omit a source or record a result differently. If several people will check answers, write the recording rules before they start.

API tracking

An API workflow is useful when you already have engineering support and want the same check to run without a browser. Store the question set in a file or database, send each question through the chosen API, save the returned answer and apply a repeatable mention test.

Start with literal matching, then add review. Literal matching catches the exact brand name, but it can miss abbreviations and false variants. A better system keeps a list of approved names and sends uncertain matches to a person instead of pretending every result is clean.

Track citations separately from mentions. Save the cited URLs where the response exposes them, then normalise the domain and page path for comparison. The guide to measuring AI citations is useful when you need to separate a source that appears often from a brand that appears often.

API tracking does not remove the sampling rule. Your script can run thousands of requests and still produce a poor report if the questions change, the engines are mixed together or the result is reduced to a single percentage.

API pricing and rate limits change often enough that you should check the provider's current terms rather than plan around a number here.

Monitoring tools

A monitoring tool earns its place when the work has become a reporting process rather than a one-off check. You want a prompt set that stays stable, results that remain searchable and a way to see changes in mentions, citations and competitors without rebuilding the file each time.

Before buying, check the unit of work: does the product charge by question, answer, engine or user? Confirm you can export raw answers and tell a mention from a citation, since that decides whether the report can support a page change.

A tool should also make the sample visible. You need to know which questions ran, which engine produced each answer and how many answers sit behind a percentage.

A polished score with an opaque denominator is still opaque.

There is a fair argument against using a tool: with a small question set, a spreadsheet gives you more control, and buying software before you know which decisions the data needs to support only creates another reporting task. That argument holds while the check stays occasional, but it weakens when several people need the same evidence, the prompt set keeps growing or manual collection takes attention away from fixing the pages buyers need. Use a tool for repeatability, not because a dashboard looks more official.

CiteGraph's research page shows the scale a structured system can hold: 9,791 AI answers, 64,667 citations, 77 scans of 44 sites across 48 categories and 507 distinct buyer questions. That is scan data, not a promise that your own sample needs the same breadth.

The decision

Choose manual checks when you are still learning what buyers ask and what a useful mention looks like. Choose an API when you need control, have technical capacity and can maintain the rules. Choose a monitoring tool when repeated collection and shared reporting cost more than the subscription would.

Whichever route you choose, freeze the question set before you compare results. Keep ChatGPT separate from other engines, report the number of answers behind each result and preserve the answer text so someone can inspect the context later.

Then connect the result to an action. If ChatGPT names you but cites another site, study the cited page and the question it answers. If it cites your page but does not name you, improve the page's plain statement of what you sell. If neither happens, check access, clarity and the surrounding pages before declaring the brand invisible.

The right measurement is not a permanent score. It is a repeatable sample that tells you what to fix next.

Track brand mentions in ChatGPT by running the same questions on a schedule and writing down what changes.