Ask ChatGPT which tool a buyer in your category should pick, and the answer that comes back is rarely built from your website. It is stitched together from a Reddit thread, a comparison page, a review site and a YouTube walkthrough, most of which you did not write and some of which do not mention you by name. Which means the honest way to change what ChatGPT says about you is not to rewrite your homepage, it is to change what ChatGPT can read, and the first step in that is the least glamorous one: actually seeing the sources sitting behind the answer.
That source list is closer than it looks and better hidden than it should be. ChatGPT will show you its citations if you ask it the right way, but the app trims what it displays, so the version you see is a partial view of what the model actually read. Learn to pull the fuller list and the rest of the job, getting your brand into it, stops being guesswork and starts being a content plan.
What sits behind a ChatGPT answer about you
40.1%
of all AI citations point to Reddit, the single most-cited source
The likeliest place ChatGPT is reading about you is a thread you never posted in. Semrush.
62%
of AI citations never name the brand
The answer is built from pages that discuss you without crediting you. Semrush.
~60% to ~10%
Reddit's ChatGPT citation share, inside one fortnight
The sources feeding your answer reshuffle in weeks, late 2025. Semrush.
The numbers behind that are unsentimental. Reddit alone accounts for 40.1% of all the citations AI engines make, according to Semrush's analysis of roughly 150,000 citations, which means the single likeliest place ChatGPT is forming a view of you is a forum thread you have never posted in. And because 62% of those citations never name the brand at all, the answer a buyer reads is often assembled from pages that discuss your category, and you, without ever crediting you. The list does not hold still either: Reddit's share of ChatGPT's citations fell from roughly 60% to roughly 10% inside a single fortnight in late 2025, so a source feeding your answer this month may be gone the next.
What ChatGPT actually cites, and how much
Start with the reassuring part, which is that ChatGPT does cite, and generously. When it runs a web search it returns more sources per answer than any of its rivals: an average of fifteen in our July 2026 test of buyer-intent prompts, against roughly ten for Gemini and eight or nine for Claude and Perplexity.
Sources per answer
Average sources cited per answer, by engine
Fifteen sources sounds like transparency until you remember that the consumer app usually shows only a handful of them under a tidy 'Sources' line, so the citation you can see is a curated subset of the citations the model used. That gap is where most brands lose the plot: they check the visible list, do not find themselves, and conclude ChatGPT cannot see them, when the truth is usually that ChatGPT read a page about them and simply did not surface it. For the fuller picture of what ChatGPT shows, hides and occasionally invents, our companion piece on whether ChatGPT cites sources lays out the mechanics.
How to see the sources behind an answer
Seeing the list is a manual job in the app and an automated one in a tool. The manual route works engine by engine, and it is worth knowing all four, because a buyer rarely asks only one of them.
| Engine | Sources visible by default | How to see the fuller list |
|---|---|---|
| ChatGPT | Sometimes, and fewer than the model used | Turn web search on, then expand the citations shown under the answer and open each linked page |
| Gemini | Often tucked behind a sources toggle | Open the sources panel, then cross-check whether the linked pages actually name you |
| Perplexity | Yes, numbered inline | Read the footnotes; every claim carries one, which makes it the easiest place to audit citations |
| Claude | Yes, when it has searched the web | Check the sources it lists after a web search rather than assuming it answered from memory |
None of this is hard. All of it shares the same flaw, which is that it shows you one answer, at one moment, for one phrasing of the question, and ChatGPT's answer changes about seven times in ten between identical runs, as SparkToro's repeat-query work found. So the most misleading thing you can do is read the sources once and treat them as the sources; they are a sample of one. Our test of whether ChatGPT gives everyone the same answer shows how far a single reading can lead you astray.
Why it names your competitors and not you
When ChatGPT recommends a rival instead of you, the cause is almost never that your product is worse, it is that the pages ChatGPT trusts have more to say about the rival than about you. AI visibility, as Ahrefs found across its large study of what moves the needle, correlates most strongly with third-party mentions and video, and barely at all with the on-page tweaks that consumed the last decade of search work. That reframes the task: the lever is not your homepage copy, it is how often, and how well, the wider web talks about you.
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.
In plain terms, being named is downstream of being discussed. If the review sites, forums and roundups in your category cover three competitors in depth and mention you in passing, ChatGPT will name those three, and it will do so in a confident sentence that reads like a verdict rather than an accident of coverage.
How to get into the sources ChatGPT cites
The playbook follows the citations. If community threads, third-party reviews and video carry the bulk of what AI engines cite, that is where the work goes, roughly in that order of weight.
| Source | Where it sits in AI citations | How to earn a place there |
|---|---|---|
| Reddit and community threads | The single most-cited source, at 40.1% of all AI citations | Be genuinely useful in the subreddits your buyers already read, rather than posting your own launch notes |
| Third-party reviews and roundups | Where a buyer-intent question most often lands | Get listed and fairly reviewed, with your real strengths and caveats stated, on the comparison pages your category is judged on |
| Video, meaning YouTube and demos | One of the two source types AI visibility tracks most closely | Publish explainers that answer real buyer questions out loud, not brand films |
| Your own site | Read, but rarely the deciding citation | Keep it open to GPTBot and write your claims to be quoted verbatim |
Reddit comes first because it is the largest single source and the one brands most often get wrong. The move is not to post your own announcement, it is to earn a place in the threads your buyers already trust, because AI models weight the community's verdict rather than your press release. We set out why in why AI models cite Reddit and the practical version in getting cited on Reddit by AI.
Third-party reviews and roundups come next, because they are where a buyer-intent question most often lands, and a fair listing on the comparison pages your category is judged on does more for an AI answer than any amount of homepage polish. The broader version of this argument, across every engine rather than just ChatGPT, is in our guide to getting cited by AI.
Your own site matters least for the citation and most for the crawl. ChatGPT rarely makes your homepage the deciding source, but it cannot read a page it is blocked from, so the one technical job worth doing is making sure GPTBot can reach you and that your claims are written to be lifted whole. Our GPTBot crawler guide covers the robots.txt and rendering details that quietly decide whether you are readable at all.
Watch the list, do not glance at it
Because the source set moves week to week and phrasing to phrasing, the only honest way to know which sources ChatGPT cites for your category is to sample it repeatedly and log what comes back, which is precisely the job no founder wants to do by hand at eleven at night. That is what citation-tracking tools automate: they run your buyer prompts across engines on a schedule, record the sources behind each answer, and show which ones are stable, which are new, and which competitor has just picked up a citation you have not. Our roundup of AI citation tracking tools compares the options, Honeyb, our own tool, included and disclosed as such.
The point of the automation is not the dashboard, it is catching the fortnight when Reddit's share of your answers halves, before your pipeline notices the same thing a quarter later.
The bottom line
The verdict is simple enough to act on this week. ChatGPT's answer about you is assembled from other people's pages, most of them on Reddit, review sites and video, so the way to change the answer is to change what those pages say and to keep watching which ones ChatGPT is reading. Start by pulling the source list behind the three buyer questions that matter most in your category; if your brand is missing from the pages ChatGPT cites, that gap is your content plan for the quarter, already written. The fastest way to see where you stand is to run the check: our free AI visibility checker shows which prompts name you, which name your competitors, and the sources sitting behind both.
