Ask ChatGPT, Gemini or Perplexity which tool a buyer in your category should use, and one of three things happens: the engine names a company outright, it assembles an answer from pages that discuss the category without crediting anyone in particular, or it does not appear to know you exist. Most founders assume the job is to climb some ranking until the first outcome lands on them, and then pour their effort into the one surface they fully control, which is their own website. The argument of this piece is that they are polishing the wrong surface, because presence in a generative answer is built from the sources the model trusts, most of which sit off your domain, so the work that shifts it is the work of getting discussed, cited and reviewed elsewhere. The first task is therefore not to optimise a page but to find out which of the three outcomes you are actually getting, and the figures below explain why the middle one catches so many brands out.
Where brand presence in generative AI is actually decided
62%
of AI citations never name the brand
Your page can feed the answer and your name still not survive into it. Semrush.
40.1%
of all AI citations point to Reddit
The single most-cited source, so presence is often won in communities. Semrush.
<1 in 100
chance two identical questions name the same brands
Presence is a rate across many answers, not a fixed rank. SparkToro.
Presence comes in three states, not on and off
The instinct to treat presence as a switch, either the AI recommends you or it does not, is the first thing to drop, because in practice there are three states and the middle one is where most brands sit without realising it. An engine can read your page, lift a fact or a turn of phrase from it, and then hand the credit to a review site, a forum thread or nobody at all: 62% of AI citations never name the brand being cited (Semrush). The more common failure, in other words, is not being absent from the answer but being invisible inside it, with your words doing the work while someone else, or no one, collects the recommendation. That distinction changes what you go and fix, so the table below sets out the three states, what each looks like in the wild, and what actually moves you up a rung.
| The state you are in | What the AI does with you | What it costs you | What actually shifts it |
|---|---|---|---|
| Named | Recommends you by name as an option a buyer should consider | Nothing yet; the job is to hold the position across engines and reruns | Repetition of the mentions and reviews that earned the name |
| Cited, not named | Reads your page to build the answer, then credits a third party or no source at all | Your content fuels a competitor's recommendation | Earning mentions that attach your name to the claim, not just the fact |
| Absent | Neither reads nor names you; the answer is built entirely from others | The category conversation happens without you in it | Getting into the sources the models already read, from scratch |
The answer is written where you are not
Once you accept that presence is assembled from sources, the next question is which sources, and here the data is unusually blunt. Reddit alone accounts for 40.1% of all AI citations, the single most-cited source across the engines (Semrush), which tells a founder something uncomfortable and useful at once: the answer a buyer reads about your category is being shaped, more than anywhere else, in a place your marketing team neither owns nor can brief. Ahrefs' analysis of what correlates with AI visibility points the same way, finding that third-party mentions and video track it far more closely than any on-page factor does. The plain reading is that the homepage rewrite you were about to commission will move the needle less than a fortnight spent earning honest mentions in the communities and among the creators the models already read, and if the Reddit share surprises you, our explainer on why AI models lean so heavily on Reddit covers the mechanics.
This is also why the discipline has drifted away from classic search work and towards earning citations, a shift set out in full in our guide to what generative engine optimisation is and in the practical playbook for how to get cited by AI. The through-line is the same in both: you are not writing for a crawler that ranks pages, you are trying to become the thing an answer is built out of.
Presence is a rate, not a rank
If presence lived on your website you could screenshot it once and file the result, but it does not, so you cannot. The same buyer question returns a materially different answer roughly 70% of the time, and two identical questions asked back to back match the same list of named brands less than one time in a hundred (SparkToro), which means a single check that shows your name is closer to a coin landing heads than to a rank you have earned. Presence is therefore a rate, the share of answers across many runs in which the engine names you, and the only honest way to know yours is to sample the same questions repeatedly rather than trust the reply you happened to see this morning. This is the whole reason a single-figure visibility score has to be read with care, which our explainer on what AI visibility actually measures unpacks, and it is the reason a serious answer to whether generative engine optimisation works is always a trend line rather than a screenshot.
Want to get recommended by AI?
Check your AI search visibility, then let the Honeyb agent write, fix, and earn what gets you recommended. Free to start.
Which engines will show you their working
Sampling tells you the outcome; diagnosing the cause means reading the sources behind an answer, and the engines differ sharply in whether they let you. Perplexity and Claude expose their citations, so you can work backwards from a disappointing answer to the exact pages that fed it, while ChatGPT and Gemini usually keep theirs hidden, leaving you to infer the cause from the pattern of outcomes. That asymmetry makes Perplexity the honest mirror of the set, the fastest place to see in the open which forum post or review is teaching the model about your category, which is why our guide to getting your brand mentioned in Perplexity treats it as the engine to start with. The table below is the quick reference for where to look first.
| Engine | Shows the sources behind its answer | What that means for diagnosing your presence |
|---|---|---|
| Perplexity | Yes, citations shown in the open | Trace a weak answer straight to the pages that caused it |
| Claude | Yes, exposes its citations | Sources are visible, so you can reason from cause to fix |
| ChatGPT | Usually hidden | You see the verdict, not the working; infer the cause by sampling |
| Gemini | Usually hidden | Same again: measure the outcome repeatedly rather than read sources |
What to do about it this week
The practical sequence follows from the argument. Start by measuring rather than building: run a fixed set of the questions your buyers actually ask across ChatGPT, Gemini, Claude and Perplexity, several times each, and record how often you are named, how often you are merely cited, and how often you are absent, because that split tells you which of the three problems you are solving for. Then spend where the sources are, which means the communities and third-party sites the models cite, the reviews and comparisons your category lives in, and the video that Ahrefs finds tracks visibility so closely, rather than another pass over your own copy. It helps to know where the disciplines sit relative to each other, which our comparison of how SEO, AEO and GEO differ lays out, and whether the whole effort is worth it for your category, which our verdict on answer engine optimisation argues honestly either way. Honeyb, which is our product and so read this paragraph with that in mind, runs exactly that sampling loop on a schedule across the four engines so the volatility averages into a trend you can act on rather than a screenshot you have to trust.
The one thing not worth doing is guessing. Before you decide whether your presence problem is absence, invisibility inside the answer, or simple volatility, see what the engines say about your brand today with Honeyb's free AI visibility checker, which runs across ChatGPT, Gemini, Claude and Perplexity and returns the split by name rather than a single blended score, and which our walkthrough of auditing your presence in AI answers explains how to read. Whatever it shows, the fix will sit off your homepage, in the sources the models were reading all along.
