5 Places AI Learns About Your Brand (Your Website Isn’t One)

Somewhere this week, a buyer asked an AI assistant to recommend a company like yours. The answer arrived in seconds, named three or four brands and moved on. Your website had almost nothing to do with what it said.

Founders keep asking me how to show up in that moment, and most arrive with the same plan: rewrite the homepage again. The research keeps pointing somewhere else. Models form their picture of a brand from what other people publish about it, so the useful question is where they do their reading.

This is a different map from the one SEO trained us on, where every road led back to your own domain. The new one runs through territory you do not control, which is uncomfortable and also the point. Ranked by the weight of the current evidence, it looks like this.

1. Journalism and earned press

Nothing else comes close. Muck Rack analyzed more than 25 million links cited by ChatGPT, Claude and Gemini and found that earned media accounts for 84% of AI citations, with journalism alone at 27%. Paid placements barely register at 0.3%.

The spread is wider than people assume. Category questions pull from trade publications and niche industry sites as often as from famous mastheads, which puts useful coverage within reach of companies that will never see a national headline.

The practical work here has not changed in decades. Answer journalist requests quickly, hand writers a usable quote instead of a brochure and publish original numbers people can repeat. Coverage you cannot buy is exactly the coverage the machines trust.

2. Video and podcast transcripts

When Ahrefs studied 75,000 brands, YouTube mentions turned out to be the strongest single signal it measured, ahead of every traditional metric. The reason is plain: transcripts are text, and models read text at scale.

You do not need a channel of your own. Guest on the niche shows your buyers already follow and say your company name out loud, clearly, more than once. A host who calls you “my guest’s company” for 40 minutes has made you invisible to every transcript that follows.

3. Review platforms

For buying questions, review sites punch far above their size. SE Ranking’s research found that a small group of them dominates the review links AI answers cite, even as their human traffic falls. Neutral, structured opinion from strangers is the kind of source a cautious model prefers over your sales page.

Claim your profiles on the two platforms your category uses, ask satisfied clients for reviews at a steady pace and reply to what comes in. For business software, that usually means G2 or Capterra, while consumer brands live and die on Trustpilot and its peers. Replies are text too, and they show the system a company that answers.

Reddit, Quora and the niche forums of your industry now feed AI companies directly through licensing deals, and models lean on candid discussion when marketing copy reads too polished. This is the slowest channel on the list and the easiest one to ruin.

Participate as yourself, under your real name, answering questions you can answer well. Six months of honest answers beats six days of seeded threads, and the models keep the receipts either way. Promotional posts get removed by moderators and discounted by models for the same reason: everyone can tell.

5. Reference sites and roundups

The encyclopedic layer still matters. Wikipedia, industry wikis, directories and credible best-of lists are where systems check who is who, and you cannot write your own Wikipedia entry without it backfiring. Entries get written from press coverage, which loops this list straight back to number one.

While you are at it, make sure every directory spells your company one way. A business listed under three name variations reads to a machine like three small strangers.

And your own website

Every study above lands on the same footnote: what you publish about yourself carries the least weight. Your site is where machines verify, not where they discover. So make verification easy. Give every important claim a date and a plain sentence a machine can lift, keep your names consistent and stop expecting publishing volume alone to buy visibility.

The audit takes 20 minutes. Write down the five questions a buyer would ask an assistant about your category, run them through two different tools and note where every cited source came from. Then compare that list with where your marketing budget went last quarter. For most companies I meet, the two lists barely overlap, and closing that gap quietly decides who gets named for the next few years.

Somewhere this week, a buyer asked an AI assistant to recommend a company like yours. The answer arrived in seconds, named three or four brands and moved on. Your website had almost nothing to do with what it said.

Founders keep asking me how to show up in that moment, and most arrive with the same plan: rewrite the homepage again. The research keeps pointing somewhere else. Models form their picture of a brand from what other people publish about it, so the useful question is where they do their reading.

This is a different map from the one SEO trained us on, where every road led back to your own domain. The new one runs through territory you do not control, which is uncomfortable and also the point. Ranked by the weight of the current evidence, it looks like this.