Guide · 19 July 2026

ChatGPT brand monitoring

Somebody asked ChatGPT what to use in your category this week. It named three or four products and the rest of the market did not come up. You have no record of that happening, no impression, no click, no line in any dashboard. This is how to find out which side of that answer you are on, what can honestly be monitored, what cannot, and what the tools that do it cost.

What it actually is, and what it is not

ChatGPT brand monitoring is checking whether ChatGPT names your product when someone asks it what to use, which competitors it names instead, and which pages the answer was built from. That is the whole definition, and the second half of it is the part most tools skip.

It is worth being blunt about what it is not, because the category borrows language from social listening and the borrowed language is misleading. Social listening works because posts are public: you can search them. ChatGPT conversations are not public. There is no feed to monitor, no API that returns what people asked, and no way to know how many buyers in your category asked anything at all. Any tool whose marketing implies it can see what people are asking about you is describing something nobody can do.

What can be observed is the output side. You can ask the question yourself and record the answer. That is a sample, not a census, and treating it as a sample is what separates an honest measurement from a number invented to fill a dashboard.

Why the blind spot is worse than it sounds

Every other channel you run has a feedback loop. A search ranking drops and your traffic tells you. An ad stops converting and the cost per acquisition tells you. Being left out of an AI answer produces no signal at all. The buyer does not bounce off your site, because they never reached it. They read a paragraph naming three competitors and moved on.

That makes this the only significant channel where the default state is not knowing. And the not-knowing is asymmetric: founders who check tend to find they are absent, because the shortlist is short. In our own measurements the average answer named around seven or eight products out of categories that contain dozens.

What we found when we measured it properly

Rather than assert any of this, we ran it. We put 40 real software buying questions to ChatGPT, Gemini and Perplexity and recorded every product named and every source cited. The full method, including its limits, is in what sources AI actually cites and how much the assistants disagree. Four findings change how you should monitor.

Checking once tells you almost nothing

Of 266 distinct products named across those questions, 223 appeared in exactly one question. Five sixths of everything named showed up once and never again. So your product appearing in one answer is not evidence that it will appear in the next, and a founder who asks ChatGPT once, sees their name and relaxes has measured a coin flip.

Checking only ChatGPT tells you about 60% of the story

Across the three assistants, only 30% of products were named by all of them and 47% were named by exactly one. Checking ChatGPT alone would have shown you 58% of the products named across the three. If your buyer happens to use Gemini, which is answering inside Google searches they never thought of as AI, your ChatGPT result told you nothing about them.

The sources decide, and they are not where companies spend

Reddit appeared in the sources of 38 of the 40 answers. The five classic software review directories together accounted for 0.7% of all citations, and four of the five were never cited once. If you are paying a directory specifically because you believe it is how AI finds you, that belief is not supported by what we measured.

Your own pages are in the answer more often than you would think

In 29 of the 40 answers, at least one named product had its own domain sitting in the source list. Just under three quarters. This is the finding that makes monitoring worth doing rather than depressing: the answer is not decided entirely by strangers writing about you. A page you control, written to actually answer the question, gets retrieved.

How to monitor it by hand, this afternoon

You do not need a tool for the first pass and we would rather you did it manually once, so that you know what any tool is claiming to automate.

Write down ten questions your buyers ask before they know you exist. Phrase them the way somebody types, not the way a marketer writes. The one rule that matters: never put your own product name in the question.A question containing your name guarantees an answer containing your name and teaches you nothing at all. “Is Glotier any good” is not monitoring, it is a mirror.

Put each question to ChatGPT with web search enabled, and to at least one other assistant. For every answer, record four things: whether you were named, which competitors were named, whether anything said about you was wrong, and which sources the answer cited. Then count.

Your coverage is the fraction of questions where you appeared. Your accuracy problems are the wrong statements, and those can cost more than absence, because a buyer confidently told the wrong thing about your pricing has decided against you for a reason you could have fixed. Your roadmap is sitting in the source lists.

Do that once and you will know more about your position than any article can tell you, including this one. The reason people buy a tool is not that the first pass is hard. It is that doing it every week, on the same frozen question set, so that a change is visible, is the part nobody sustains in a spreadsheet.

You ran it and you are absent. Now what?

This is where most writing on the subject stops, which is unhelpful, because the diagnosis is the easy half. Four moves, in the order the measurement justifies rather than the order that sounds impressive.

Read the sources before you change anything

The pages that fed the answers you lost are the answer to why you lost them. Open them. In most categories you will find a handful of comparison posts, a couple of community threads and one or two vendor blogs doing most of the work. If a competitor is discussed in all of them and you are in none, that single gap explains more of your absence than anything on your own site, and no amount of rewriting your homepage fixes it.

Fix the sentence, because it is nearly free

A model can only name you if it can describe you. If your homepage says one thing, your meta description says another and your structured data says a third, an assistant that retrieves your page has nothing confident to repeat, and a hedged product gets dropped from a shortlist that only has room for seven names. One plain sentence of the form “X for Y who want Z”, identical in all three places, costs an afternoon.

Write the page that answers the question you lost

Not a landing page. The page that resolves the question completely, including the part where you say who you are not for. Our source study found the non-community half of citations was overwhelmingly comparison shaped: “best X for Y”, “X vs Y”, “alternatives to Z”. Those get retrieved because they answer. A page that withholds the answer to drive a demo request is the opposite of retrievable, and it is the default shape of most software marketing.

Be genuinely useful where your category is discussed

Given Reddit fed 38 of our 40 answers, the temptation is obvious and the trap is equally obvious. Posting promotional replies at volume gets the account removed and produces a thread that argues against you, which is worse than absence because it is retrievable. The version that works is slower: find the threads where your category is actually being discussed, answer the question that was asked, including when the honest answer is a competitor, and mention yourself only where it genuinely fits. A comment that helps gets upvoted into the part of the thread that gets read.

The failure worse than absence

Presence is the only thing most monitoring measures, and it is not the only thing that can go wrong. An assistant that names you and describes you incorrectly is doing damage at scale, in a conversation you will never see, and it will not show up in any presence metric because on paper you were named.

The versions of this we have seen are mundane and expensive. A price from a pricing page you changed a year ago, still sitting in a roundup that gets retrieved. A feature you removed, described as current. A positioning aimed at the wrong buyer, because the strongest page about you is a review written when your product did something else.

The fix is unglamorous: when you monitor, read what was said, not just whether your name appeared. Then go and correct the source. A directory listing or a roundup post is usually editable by asking, and the request is easy to make because you are offering a correction rather than a favour. This is the highest value-per-hour work in the whole category and almost nobody does it, because it does not feel like marketing.

How often to check

Weekly, on a frozen question set, is the honest answer for most products, and the two halves of that sentence matter equally.

Weekly because retrieval changes continuously but your position does not move daily, and a daily check on a small product mostly measures noise. If a tool is showing you a number that moves every morning, ask what it is actually sampling.

Frozen because a measurement is only a comparison if the questions stayed the same. If the question set drifts, an improving score can mean nothing more than an easier question, and that is the most common way a visibility dashboard lies without anyone intending it to. Write the ten questions down, keep them, and only change them deliberately, knowing it starts a new series.

Set expectations on the timescale too. Publishing a page and re-checking the next day tells you nothing; the page has to be crawled, indexed and then retrieved for a question. Weeks, not days, is the honest window, and any tool implying otherwise is selling a dopamine loop rather than a measurement.

What to look for in a monitoring tool

Four questions, in the order they matter. Most of the category answers the first one and stops.

What the tools cost

Entry pricing as we recorded it on 2026-07-17. These are the published starting prices, not what a given team ends up paying, and they change. We are quoting them rather than characterising them.

The honest reading of that table is not that the cheapest tool wins. Free graders exist, and a free grader that gives you a number and nothing else is worth exactly what it costs. The question is what you get at each price.

Where Glotier sits, and why we think it is the best value

We built this, so read the next three paragraphs with that in mind. Everything in them is checkable, which is the only kind of self-praise worth writing.

The free check does the thing the four questions above ask for. No account, no card. It works out what your product is, generates the questions your buyers would actually type, puts each one to ChatGPT, Gemini and Perplexity as a live web query rather than from training memory, and shows you which of the three named you, which competitors were named instead, and every source page each answer was built from. The tools in the table that show source-level detail across multiple assistants start at $95 and $99 a month. That gap is the argument, and it is the whole argument.

Whether you were named is decided by looking, not by asking a model. We check for your name in the answer text directly. A model asked “was this brand mentioned” is agreeable and will sometimes say yes to be helpful, and the one number this product cannot get wrong is that one.

We publish our measurements with the method and the limits attached. The figures in this article come from studies you can read, including the parts that do not flatter us: our own coverage in our own category is poor, ChatGPT has named us twice in 61 days, and we wrote that up rather than quietly leaving it out. We also corrected one of our own numbers in public when a first analysis turned out to be wrong in the direction that flattered us.

What we do not claim: that we are the cheapest, because free graders exist. That we can see what people ask ChatGPT, because nobody can. Or that we know why a model chose what it chose, because nobody outside those companies does, and a list of “AI ranking factors” is a guess with a confident face on it. What can be measured honestly is the question, the pages retrieved, and the names that came back. That turns out to be plenty to act on.

Where to start

Run the ten questions by hand if you want to understand the problem, or run our check if you want the same thing in a couple of minutes with the sources attached. Either way, the number to write down is not whether you appeared. It is in how many of your buyers’ questions you appeared, across more than one assistant, so that next month you can tell whether anything you did worked.