How to get mentioned by AI: five off-site sources, and the one you cannot buy

How to get mentioned by AI comes down to one mechanic: an answer engine can only name something it has already seen named, by someone else, next to the thing being asked about. Your own site cannot supply that sentence. Five kinds of off-site page can, they take between a day and a quarter to land, and only one of them is impossible to buy.

Read this first

This chapter is about the half of the work that happens on other people's domains. The other half — declaring who you are on your own pages so an engine can attach the two together — is entity SEO for AI search, and it should already be done. A mention that an engine cannot connect to a real organisation is a string, not a signal.

Two words are used precisely here and are not interchangeable. Mentioned means your name appears in someone else's text. Cited means an engine attributes a statement to a URL you own and usually links it. This chapter is about the first one. The second is decided by what is on your page, not by who talks about you.

Why a mention works differently from a link

A link is a machine-readable edge between two documents. A mention is a plain string sitting inside a sentence, and its value comes from the sentence, not from the markup. "We use Framer for the marketing site" carries a category, a use case and an endorsement in eleven words. A bare link carries none of that.

That difference sets everything else in this chapter. You are not trying to accumulate edges. You are trying to get your name written into sentences that already answer the question you want to be part of — which means the target is not a domain with high authority, it is a specific page that already answers that specific question.

You cannot write the sentence that gets you mentioned. You can only make it easy for someone else to write it.

The five sources, and what each one costs

Every off-site mention worth chasing falls into one of five shapes. They differ on the only two axes that matter in practice: how much control you have, and how long it takes.

The order in that table is deliberate. Reference records are first because they are the only source where the criteria are published, the outcome is binary, and nobody has to like you. Community answers are fourth because they are the highest-value and the only one with no lever at all.

Reference records: the boring one that works

A registry entry is a structured statement that a thing exists, has a name, and belongs to a category. Package registries, app stores, code hosts and open knowledge bases all produce one. They are dull, they are mostly forms, and they are the only mentions in this list whose acceptance criteria you can read in advance.

Two rules. Fill the category field with the words a person would use, not your marketing phrase. And never create a record about yourself where the rules say you should not — encyclopaedic projects in particular have conflict-of-interest policies, and getting removed is a worse outcome than never appearing.

Third-party lists: cheap, slow, and easy to overdo

Someone maintains a page listing tools in your category. Getting added is usually a pull request, a form or an email. The cost is an hour; the risk is doing three hundred of them, which produces a footprint that looks exactly like what it is.

Pick lists that a person would actually read to the end. If the page is a wall of two hundred entries with one sentence each, being entry 187 does not put you in anyone's sentence.

Integration pages: the most durable mention you can earn

When another product documents that you work with them, your name lands on a page that they maintain, that their users read, and that stays current because it has to. This is the mention that ages best, and the only way to get it is to build the integration.

Community answers: the one you cannot buy

Someone asks which tool does X, and a stranger names you. This is the most valuable shape and the only one with no procedure attached. What you control is upstream: whether the thing you built is specific enough that a person can recommend it in one sentence without hedging.

Do not manufacture these. Beyond the ethics, threads that read like planted recommendations get removed by the communities themselves, which turns a mention into a deletion log entry.

Primary coverage: slow, and not the place to start

Podcasts, newsletters and press produce mentions inside long-form text with real context. They also take months and depend on someone finding you interesting. Treat this as the last of the five, not the first.

How to get mentioned by AI in five steps

The sequence matters more than the effort. Most of the wasted work in this area comes from step three being skipped.

1. Write the sentence you want someone else to write, in under 20 words, naming the category and the specific job. If you cannot, no source in the table above will help.
2. Find the pages that already answer that question today. Search the question as a person would ask it, and list the pages in the first two result pages that a machine reading for an answer would use.
3. Sort those pages by whether you can legitimately be added to them. Skip anything that would require you to be a different company.
4. Do the one action that source type accepts: submit the record, open the pull request, build the integration, ship the thing, send the pitch.
5. Verify the mention exists in the page's raw HTML, not just in your browser.

Step five catches a failure mode that looks like success. Plenty of directory and community pages render their content in the browser, and a mention that only exists after JavaScript runs may not be in the copy a crawler stored. Fetch the page with curl and search the returned HTML for your name. If it is not there, the mention exists for humans only.

The deliverable: a mention record

Keep one row per attempt. Five columns, filled in when you act and again when you verify.

The last column is the one people skip and the one that pays. Six months in, reading twenty of those sentences side by side tells you what other people think you are — and if that description is not the one you wanted, the fix is upstream in the product or the positioning, not in more submissions.

Three ways this goes wrong

All three are versions of optimising the wrong unit — the domain instead of the page, the volume instead of the sentence, the outcome instead of its cause.

1. Chasing domains instead of pages. A mention on a high-authority site's unrelated page does nothing for the question you care about. A mention on a small forum thread that directly answers it might.
2. Bulk submissions. Two hundred directory entries in a week produce a uniform, mechanical pattern and no sentences worth reading. The work is per-page or it is not work.
3. Counting mentions as citations. They are different outcomes with different causes. If you want to be cited, the lever is your own page structure, described in how AI engines pick sources.

What this chapter cannot tell you

We have not measured that any of this moves AI citations, and we are not going to imply it. Nobody outside those companies can see what a model was trained on or what its retrieval layer indexed, so the causal chain from "a stranger named you in a thread" to "an engine names you in an answer" is not observable from where we sit.

Two more limits worth writing down. We do not know whether linked and unlinked mentions differ for AI systems the way they do for classic search. And this approach does not apply at all if your category has no third-party pages yet — if nobody has written the roundup, there is nothing to be added to, and your time goes into making the category legible instead.

What you can observe is whether the number of answers that name you changes over time, which is a measurement with its own procedure: how to measure AI share of voice by hand. Run it before you start, or you will have nothing to compare against. Automating the whole loop is what QueryWin is being built to do.

Common questions

How do I get mentioned by AI without a marketing budget?

Start with reference records and integration pages. Neither costs money, both have published criteria, and both land on domains that stay maintained. Press and podcasts are the expensive ones, and they are last in the table for that reason.

Do I need a link, or is being named enough?

For being mentioned, the name in the text is the thing that matters, because the sentence is what carries the meaning. Whether the link adds anything on top for AI systems is not something we can measure, so take the link when it is offered and do not walk away when it is not.

How long before a mention shows up in an answer?

We do not know, and anyone giving you a number for this is guessing. What we can say is that the page has to be crawlable and stable first, which is why step five exists.

Is this just link building with a new name?

It overlaps, and the tactics look similar from a distance. The difference is what you optimise for: link building counts edges, this counts sentences, and a page with one clear sentence about you beats five pages with your name in a footer list.