2026-08-03 · Jack Stovell

How to make your startup findable by AI assistants

AI assistants have never heard of your startup — it did not exist when the models were trained, so nothing about it is in their weights. That leaves live retrieval as your only path: an assistant searches, fetches your pages, and answers from what it can read in the moment. Being findable therefore comes down to three things you control this week: copy written in your buyer's words, Organization markup that ties the company to its founder, and corroboration outside your own site.

Why doesn't ChatGPT know your startup exists?

Because there are two ways an AI system can know an entity — baked into training data, or found by live retrieval — and a company founded after the training cutoff is only reachable by the second. Established competitors sit inside the weights; you do not, and waiting does not fix it quickly, since training influence accrues over crawl and release cycles measured in months. This sounds like a pure handicap but is partly an advantage: there is no stale, wrong version of you to correct, and the retrieval path works from the day your pages are readable. The full mechanics of that path — crawlability, structure, provenance — are in get your business cited by AI; this page covers what is specific to being new.

What should a startup's Organization schema include?

The three fields new companies usually omit: founder, foundingDate and sameAs. A week-old company has almost no footprint, but its founder usually has one — a GitHub account, a personal site, a profile with history. Naming the founder inside your Organization JSON-LD, and naming the company on the founder's own pages, ties the two entities together so each corroborates the other; the link has to run in both directions to do anything. A minimal, valid block to adapt:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Acme Pipeline Monitor",
  "url": "https://acmepipeline.example",
  "description": "Alerting for broken data pipelines, built for small data teams.",
  "foundingDate": "2026-02-01",
  "founder": {
    "@type": "Person",
    "name": "Sam Example",
    "url": "https://i.hilyt.it/samexample"
  },
  "sameAs": [
    "https://github.com/acme-pipeline",
    "https://www.crunchbase.com/organization/acme-pipeline-monitor"
  ]
}
</script>

Which properties AI systems demonstrably read is a moving target; the evidence-based subset is in Person and Organization schema AI actually uses.

Are you describing yourself in the buyer's words?

Retrieval matches query language against page language, and buyers do not ask assistants in positioning language. Nobody asks "who is redefining data observability" — they ask "tools that alert me when a data pipeline breaks". If your homepage leads with vision and invented category names, the sentence a buyer's question would match may not exist anywhere on your site. Keep the positioning if you want it, but make sure one plain sentence stating the category, the problem and the customer sits in crawlable text: what it is, who it is for, what it replaces. For most early-stage sites this single paragraph is the highest-leverage edit available.

Which corroboration sources can an early startup actually get?

A machine deciding whether to name you looks for agreement between independent sources, and early companies have real if unglamorous options: a Crunchbase profile (free to create, machine-readable, widely crawled), product directories relevant to your category, launch-announcement press, and the founder's established surfaces. Be realistic about the weight of each — these are modest signals, some directories are noise, and none substitutes for eventual customers and genuine coverage. But two or three consistent third-party pages move you from "only its own site says this company exists" to "independent sources agree", and consistency is the point: same name, same one-line description, same URL everywhere. Never buy fake coverage — models cross-check, and a caught fabrication costs more than the mention was worth.

None of this matters if the site is unreadable at step zero: a JavaScript-shell frontend, or a Cloudflare zone left on its default AI-crawler blocking, serves crawlers nothing regardless of your copy. Run the free checker before polishing anything above.

What won't this get you?

Recommended on demand. Readable, marked-up and corroborated removes the reasons an assistant cannot surface you; whether it does depends on retrieval ranking, competition for the query and model behaviour, which nobody sells honestly. We track this in public — our own scoreboard at hilyt.it/geo currently records that models do not yet cite hilyt profiles for the queries we track. The claim worth building on is narrower: a startup that is findable and verifiable can be discovered the day someone asks the right question. One that is not, cannot.

Create your AI-readable profile