How to get your product mentioned in AI answers

Open ChatGPT and ask it the question your best customer would ask. "What should I use to do X?" Read the answer.

Three or four products get named. Yours isn't one of them. And unlike a Google result, there's no page two to be on — there's the answer, and there's not being in it.

So you go looking for how to get your product mentioned in AI answers, and within ten minutes someone on LinkedIn has explained that you need GEO (generative engine optimisation), and that GEO means schema markup, an llms.txt file, and a monthly retainer. Some of that is true. Most of it isn't. This post is about which is which: what the work actually consists of, and — the part that gets left out of every pitch deck on this subject — what you genuinely cannot make happen no matter what you spend.

The short version, before the detail:

1. Work out which engine your buyers actually use. ChatGPT, Google's AI Overviews and Perplexity cite completely different sources.
2. Do the hygiene in one afternoon, then stop. llms.txt and schema markup are close to worthless for this, and Google says so in its own documentation.
3. Get discussed somewhere you don't own. That's the entire game.
4. For most B2B products, that somewhere is Reddit.

I wrote up which websites actually get cited in AI Overviews a few weeks ago, which is the evidence base for all of this. This post is the practical half. I'll link back rather than repeat the tables.

Step 1: Pick one engine — ChatGPT, AI Overviews and Perplexity cite different sources

The single most common mistake is treating AI search as one destination. It isn't. The engines have genuinely different reading habits, and the strategies that work on them barely overlap.

Here's what each one actually reads, going by the citation data:

• ChatGPT reads encyclopaedias. Wikipedia is roughly half of its top-ten citation share.
• Google's AI Overviews read social and video. Reddit and YouTube lead.
• Perplexity reads forums. Reddit is close to half of its top ten.

So "get mentioned in AI" isn't one project. It's three, and you probably only have the time for one.

If your buyers are in ChatGPT, the highest-leverage asset is a Wikipedia page you almost certainly can't have — Wikipedia's notability bar requires significant coverage in independent sources, and a seed-stage SaaS doesn't clear it. That's not a content problem you can outspend.

If your buyers are in Perplexity or hitting AI Overviews, the answer is community discussion, and that's a project you can start this week.

So before anything else: find out which one your customers use. Ask five of them. It takes an afternoon and it determines everything downstream. A Reddit campaign aimed at a ChatGPT-native audience isn't an AI visibility strategy — it might still be a decent lead-generation strategy, but those are different things and conflating them is how people end up disappointed with both.

Step 2: Clear out the busywork — llms.txt, schema, and "AI SEO" retainers

Before the real work, clear out the fake work. Most of what the "AI SEO" market sells you lives in this section, and it's cheap to do and worth roughly what it costs.

Does llms.txt work?

Essentially no. Ahrefs checked 137,210 domains receiving traffic in May 2026. About 28% published a valid llms.txt — and 97% of those files received zero requests that month.

Of the 3% that got any traffic at all, 96% was bots, mostly SEO audit crawlers checking whether the file existed. Only about a fifth of the bot requests came from AI tools of any kind.

The single heaviest AI consumer of llms.txt turned out to be Claude Code. Coding agents read them. Answer engines don't.

Google says the same thing in its own words. Its AI features documentation is unusually blunt: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary," and "you don't need to create new machine readable files, AI text files, or markup to appear in these features."

When the platform publishes that in its own docs and an agency quotes you £2,000 a month to do it anyway, you have all the information you need.

Does schema markup help you get cited?

Not meaningfully. Structured data helps engines parse a page they've already decided to read. It has no influence on whether they decide to read it.

None of this is harmful, to be clear. llms.txt takes twenty minutes, schema is good practice, and being crawlable is table stakes. Do all of it in an afternoon and then stop thinking about it.

It's hygiene, not strategy. The distinction matters because hygiene has a fixed, small cost and a fixed, small payoff, whereas the actual work is slow and never finishes — which is exactly why the market prefers selling you the first one.

Are you accidentally blocking the AI crawlers?

This is the one piece of housekeeping worth more than all the rest combined. GPTBot, PerplexityBot, ClaudeBot and Google-Extended are all controlled through robots.txt, and plenty of sites blocked them during the 2024 panic about training data and never revisited the decision.

If you're blocked, nothing else in this post can work. Check that before you do anything else.

Step 3: Get talked about somewhere you don't own

Be the thing that gets described, rather than the thing that describes itself. That's what all of this comes down to.

An answer engine asked "best X for Y" is trying to synthesise what informed people think. Your marketing site tells it what you think. It's not a useless input, but it's a self-interested one, and the model treats it accordingly.

Third-party discussion is treated as evidence. Your own copy is treated as a claim.

That inverts the standard playbook in a way most marketing budgets aren't set up for. Owned media compounds; rented land is risky — that's been the received wisdom for a decade. The citation layer says the reverse: the sources that get quoted are precisely the ones you don't control.

In practice, here's what to go after, best first:

1. Be recommended by other people in the forums your category gets discussed in. Reddit above all, because of the citation weight, but Hacker News, Stack Overflow and niche community forums all get read too.
2. Get real editorial coverage. A comparison post on an independent site that already ranks beats your own comparison page, because yours is a claim and theirs is evidence.
3. Show up in the roundups. "Best X for Y" listicles are heavily retrieved. Getting added to an existing one that already ranks is a far better use of an afternoon than writing your own.
4. Video, if your category is one people watch before buying. Underrated, given YouTube's share of AI Overview citations, and almost nobody in B2B SaaS does it.
5. Your own site. Last. Do it well anyway, because it's what converts the traffic the rest of this generates — but don't mistake it for the lever.

Notice that none of these are things you publish. They're all things that happen to you as a consequence of other people finding you worth mentioning. That's the uncomfortable core of the whole exercise, and it's why the honest version of this advice sounds so much less impressive than the version being sold.

Step 4: Getting mentioned on Reddit, specifically

Reddit deserves a section of its own because it's the only platform that shows up near the top for two of the three engines, and because it's the one where a small company can realistically get traction inside a quarter.

Here's how it actually plays out. Someone asks for a recommendation in your category. Several people answer. Six months later that thread is a retrieved source, and the products named in it are the products in the answer.

Threads persist and compound, too — a good thread from eighteen months ago is still doing work today, which is very different economics from an ad.

So the work is: be in those threads, be genuinely useful in them, and be named in them — ideally by other people, but also, honestly disclosed, by yourself.

The hard part isn't writing the comment. It's finding the thread while it's still alive. Most founders start well, get pulled into shipping, and surface four days later to find the question was answered on day one by whoever was watching. That gap is what I built ReplyMine to close — it watches the subreddits where your category comes up, ranks the threads by whether someone is actually asking for what you sell, and gets them in front of you while the thread is still worth replying to.

I want to be exact about what that does and doesn't buy you, because this is the point in every other article on this topic where the claims quietly detach from reality.

It gets you into the conversation. It does not manufacture a citation. No tool can do the second thing. What determines whether a thread ends up being retrieved is whether it's a good thread — upvoted, substantive, not obviously astroturfed. A thread full of vendor comments gets downvoted into invisibility, and an invisible thread is not a source anybody's model is retrieving. The mechanism only works when the recommendation is real. That's inconvenient, and it's also the reason the channel hasn't been spammed to death.

If you want the mechanics of doing this without getting your account banned, the full playbook is here, and the rules for mentioning your own product are the part people get wrong most often.

What you cannot do

Four hard limits. Each one is currently sold as a service by somebody, which is precisely why they're worth spelling out.

You cannot buy a citation. There is no placement, no partnership, no ad product that puts you in an organic AI answer. Anyone offering one is selling you either advertising labelled dishonestly or astroturfing labelled optimistically.

You cannot astroturf your way in. Set the ethics aside for a moment, because the practical objection stands on its own: fake recommendations get downvoted, deleted, and account-banned, and Reddit judges spam at the account level rather than the comment level. The content has to survive a community's scrutiny before it becomes a source. That filter is the entire reason the citation is worth anything.

You cannot fix a bad product with this. If the threads about your category exist and you're not in them, sometimes that's a visibility problem. Sometimes it's that nobody's been happy enough to mention you. Those look identical from a dashboard and have completely different fixes.

You cannot get a clean before-and-after. The same prompt gives different answers to different people on different days, and the underlying numbers move violently with every model update. Track it as a trend across many prompts over months, not as a metric you can move this week.

How long does it take to get mentioned in AI answers?

Slower than you want. The chain is: be discussed → the discussion gets indexed → the model's retrieval picks it up → it surfaces in answers. Every link in that has latency, and the last two aren't yours to control.

Expect months, not weeks. In the meantime, watch the thing that comes before the citation: is your product getting mentioned by other people more often than it was last quarter? You can see that week to week, and it's what the citation eventually grows out of.

Which is a reasonable way to think about the entire discipline, really. You're not optimising for the AI. You're doing the thing that used to be called having a good reputation, and the AI is just the new place where reputations get read out loud.

The one-paragraph version

Work out which engine your buyers use, because ChatGPT wants Wikipedia, AI Overviews want Reddit and YouTube, and Perplexity wants Reddit — three different projects, not one. Spend one afternoon on the hygiene (crawlable, schema, llms.txt if you like) and then stop, because 97% of llms.txt files are never read and Google's own documentation says you don't need one. Then do the slow thing: be independently, genuinely discussed in the places the engines read, which for most B2B products means Reddit. You cannot buy the citation, and you cannot fake it convincingly enough to survive the community filter that makes it valuable in the first place. You can only be worth mentioning, in the room where the mentioning happens, often enough that it starts to show up.

Start with one subreddit where your customers already are. Find a thread from today where you can help and have nothing to sell. That's the first move, and it's the same first move it was before any of this existed.