Those used to be search-bar questions. Now a large and growing share of them get typed into ChatGPT, and the person typing isn't doing anything exotic. They're using it as a super intelligent Google, and taking the answer it gives back.
People who build with these models tend to forget that. We think about agents and workflows and what the technology could become, and most of the people using it are asking what to buy. The answer comes back as a paragraph. It names two or three products, says something about each, and the buyer picks from that paragraph the way they used to pick from a shelf.
I ran consumer brands on the shelf, on the results page, and on Amazon, and the questions I ask of the paragraph are the same two I asked of every shelf before it. Are we being recommended? And what are they saying about us, and is it even right?
The second question is the one that got under my skin. Ranking inside an answer is a fight I recognize. Getting the answer to say what's true is a new kind of fight, and here's the difference. On a package, you control the words. A customer reading your box is reading language you chose, letter for letter. When an LLM sits in the middle, the buyer never reads your words at all. The model reads your site, interprets it, and projects its interpretation to the buyer, and wherever your record is thin, it fills the gap with whatever it has. You control the packaging, not the paraphrase. And when the paraphrase gets your brand wrong, you can't correct it. There's no listing to edit. There's no support ticket. You can't necessarily see where the wrong fact came from, and it gets repeated in the same confident voice the engine uses for everything else. That is a strange position for a brand to be in, and it's the position every brand is in now.
There's no listing to edit. There's no support ticket.
You can see where you stand in about a minute, and I'd do it before reading on. Ask ChatGPT the question a buyer in your category would ask, without naming your brand, and read the paragraph twice: once for whether you're in it, once for whether what it says about you is true. Write down what you find. Everything that follows is about that paragraph.
Why a wrong answer costs more than a lost ranking
A search result that was wrong about you cost you a click. An AI answer that is wrong about you costs you the whole recommendation, because the buyer trusts it more. That's my read from the operator's chair, and the data agrees: Semrush found that the average AI search visitor is 4.4 times as valuable as the average visit from traditional organic search, measured by conversion rate (Semrush, We Studied the Impact of AI Search on SEO Traffic, 2025). The person who arrives from an answer has already compared, already narrowed, and is most of the way to buying. That is the traffic worth being described correctly for, because the paragraph that sends it is the same paragraph that describes you.
It compounds from here, and not only because more buyers ask. Agents are starting to do the asking on the buyer's behalf, with more autonomy every year, and an agent doesn't browse. It reads what the engines know about you and moves on. If the engines and the agents don't know who you are, or know the wrong thing, it's going to hurt, and you'll never see it happen.
Ranking on a page of links still matters. A new layer now sits on top of it: being the brand the engine describes correctly, and cites. That layer is what Fama owns, your visibility inside the answer itself, and the rest of this piece is what owning it looks like in practice.
One limit belongs next to that promise, before anything else. Nothing pushes a correction into ChatGPT, Fama included; the engines change what they say on their own schedule. What can be done is to find where they're wrong, put the true thing where they read, and measure whether they picked it up. I'll come back to that at the end.
One number, computed in the open
Fama starts by asking. Your real buyer questions, the unbranded ones, go to the six engines your buyers actually use: ChatGPT, Perplexity, Claude, Gemini, Grok, and Google AI Overviews. The real answers come back, and every one of them is read.
The reading is what makes the number worth anything. Each answer gets a prominence score from 0 to 5, and the scale is plain enough to argue with. A 0 means you weren't mentioned at all, and a 1 means you appeared only in a citation list at the bottom. A 2 is named in passing, mid-answer. A 3 is named as a real option in the body. A 4 is named in the engine's first sentence, and a 5 is named first and called a top pick or the best. Those blend into one number from 0 to 100. That number is your Brand IQ.
- Score 0 · not mentioned. Your brand doesn't appear in the engine's answer at all. The buyer reading it has no idea you exist in this category.
- Score 1 · citation list only. You appeared only in a citation list at the bottom, under the answer, and not in the text the buyer actually reads.
- Score 2 · named in passing. Your brand is named in passing, mid-answer, without being put forward as something the buyer should consider.
- Score 3 · a real option. Your brand is named as a real option in the body of the answer, alongside the others the engine is recommending.
- Score 4 · the first sentence. Your brand is named in the engine's first sentence, which is the part of the paragraph a buyer is most likely to read.
- Score 5 · the top pick. Your brand is named first and called a top pick or the best. This is the highest rung on the scale, not a guarantee of a sale.
Two things about that Brand IQ I'd want to know if I were buying it. First, no AI grades the AI. The prominence reader is a set of rules over where your name lands in the text, so the same answer scores the same way every time, and your Brand IQ can be explained in one sentence: a 4 because you were named in the first sentence. Why a scorer that is itself a model would inherit the volatility it exists to measure is an argument I've made on its own, so I'll leave it there. Second, if an engine doesn't answer on a run, the report says so and carries a flag for it. The gap is never quietly smoothed over into a Brand IQ that looks complete.
The free audit and the paid plans score the identical set of questions. The free number is the real number, which is a discipline of its own and gets its own piece.
Which engine knows you, and which one is making things up
A prominence score tells you how loudly the engines talk about you. It says nothing about whether what they say is true, and the two are kept deliberately separate, for a reason I'll get to.
So the audit also runs a Knowledge Check. Field by field, what your brand is, your mission, your founder story, your hero product, your audience, your values, did each engine get it right? Each cell comes back as correct, wrong, missing, fabricated, or vague, and the output is a side-by-side grid of which engine knows you best and which one is inventing a founder for you.
The engine got this field right. It matches what your brand has confirmed to be true, so there is nothing here to fix.
The engine stated something about your brand that isn't true. This is the urgent kind: an engine actively spreading a falsehood.
The engine had nothing to say about this field. Not damaging in itself, but it is a gap somebody else's answer can fill.
The engine invented the answer rather than reading it anywhere, which is how a brand ends up with a founder it never had.
The engine hedged, saying nothing specific enough to be either right or wrong, which is its own kind of absence from the answer.
The wrong answers are ranked before they're shown to you. A wrong claim and a misattributed claim are urgent, since that is an engine actively spreading a falsehood about your brand, and they're ordered worst first by reach times damage: how many engines repeat the claim, times how harmful that kind of claim is. To make the ordering concrete, a fabricated ingredient in a supplement would sit above a wrong founding year.
For every question you're absent from, Fama also reads who the engines cited instead of you and ranks those domains. That list is the AI-era version of the retail placement list I used to carry into stores, and it separates two findings that get conflated everywhere else, someone else was cited and nobody was cited at all, because an empty answer is a different opportunity from a full one and you'd work them differently.
Confirm it once
This is where the word "fix" comes in.
When the audit finds a falsehood, you confirm it once and give the real answer. That confirmed correction is respected by every future audit, every accuracy grade, and every piece of content the product generates for you, and the same claim is never flagged again. You don't re-litigate your own founding date every week.
And a correction can never move your Brand IQ. That sounds like a strange thing to promise until you've seen the alternative. If confirming corrections raised the score, the score would drift toward whoever confirmed the most. A number you can nudge is a number nobody should trust. So the two questions, how loudly the engines talk about you and whether they get the facts right, are kept structurally separate, and we test for it on purpose: your Brand IQ is computed with and without your confirmed corrections and has to come out identical. That's also why Fama can say an engine is wrong, which is a different finding from saying you weren't mentioned. It has your verified brand record to check the answer against.
| Computed with your confirmed corrections | Computed without them | |
|---|---|---|
| Your Brand IQ | Identical. The two numbers have to match, and a test checks that they do. | Identical. There is no lever here: confirming more corrections can never raise the score. |
The corrections then become content. Every gap the audit finds turns into a page-level rewrite plan: new versions of the pages you already have, checked against your voice and your compliance rules before you see them, plus the pages engines look for and most brands never wrote, an FAQ, an About page, product prose an engine can actually quote. The playbook is a story for another day, so one line here: the audit ends in content, because a to-do list is where audits go to die.
Fama also writes the machine-facing layer: structured data so search results understand your pages, and a crawl file that tells AI crawlers what they're welcome to read. Neither one is the content. They're the plumbing that lets engines read what you wrote.
What it looks like in your week
The first audit is on demand and it's free. What happens after it is your call, because monitoring is off by default on a fresh account. There is no fake always-on tracking, and no trend line that fills itself in with nothing behind it. When you switch monitoring on, the audit re-runs weekly or daily per brand, and movement that matters, a new falsehood, a score shift, lands as an alert. The week's movement arrives in one email. The digest is a piece of its own.
On demand, and free. It scores the identical set of questions the paid plans score, across the same six engines.
Switching monitoring on re-runs the audit weekly or daily per brand. Until you do, nothing runs on a schedule at all.
Movement that matters, a new falsehood or a score shift, lands as an alert, and the week's movement arrives in one email.
The lift is computed run over run. You published a correction; did the engines change what they say? That question has a measured answer now.
Then you publish the fix, and you re-run. The lift is computed run over run, which is the only kind of lift I'm willing to put on a slide. You published a correction; did the engines change what they say? That question has a measured answer now.
Fama can't push a correction into ChatGPT. Nobody can, and that's the limit worth knowing before you buy anything in this category. What it can do is find the falsehood, rank it, generate the true content, put it where the engines read, and measure whether they picked it up. The engines change on their own schedule, and some corrections take a while to land. The work is slow the way getting your facts straight has always been slow. The difference is you can finally see whether it worked.
Fama can't push a correction into ChatGPT.
What it replaces
You could hire an expensive GEO agency instead. A retainer gets you a slide deck of what the engines say and a list of recommendations. You're re-audited when you re-engage. The writing is on you, and the methodology lives in the deck.
Fama runs the same read continuously, generates the corrected content itself, and hands you receipts: the score line, the per-engine grid, and the test that proves the number can't be bought, not even by us.
| A GEO agency | Fama | |
|---|---|---|
| What you get | A slide deck of what the engines say, and a list of recommendations. | Receipts: the score line, the per-engine grid, and the test that proves the number can't be bought. |
| How often it runs | You're re-audited when you re-engage. | The same read runs continuously. |
| Who writes the fix | The writing is on you. | The corrected content is generated for you, from the gaps the audit found. |
| Where the method lives | In the deck. | In the open, in a score you can explain in one sentence. |
There's a second category it replaces, the AI-visibility monitoring dashboard, and the distinction is the product. Monitoring tools report what the engines say, and stop there. Fama grounds the same read in your verified brand record and then writes the content that changes what the engines say, which is the difference between knowing you were misdescribed and doing something about it.
I've spent my career fighting for placement on shelves I could walk into. This is the first one I can't, and the first one that describes my brand whether I show up or not. The buyer asking about their dog's supplement is going to get an answer either way. Fama exists so that when they do, you're in it, and it's right.
The buyer asking about their dog's supplement is going to get an answer either way.
Can anything make an AI engine correct what it says about my brand?
Fama can't push a correction into ChatGPT. Nobody can, and that's the limit worth knowing before you buy anything in this category. What it can do is find the falsehood, rank it, generate the true content, put it where the engines read, and measure whether they picked it up.
Which engines does the audit read?
Six: ChatGPT, Perplexity, Claude, Gemini, Grok, and Google AI Overviews. If an engine doesn't answer on a run, the report says so and carries a flag for it. The gap is never quietly smoothed over into a Brand IQ that looks complete.
Is the free audit a watered-down version of the paid one?
The free audit and the paid plans score the identical set of questions. The free number is the real number, which is a discipline of its own and gets its own piece.