2026-08-03 · Jack Stovell

AI keeps confusing you with someone else — how to disambiguate

AI models confuse you with someone else because nothing machine-readable tells them apart. The fix is to publish distinguishing signals: one canonical page about you, Person JSON-LD with sameAs links to your real accounts, and the same one-line bio — name, role, location — everywhere you appear. For answers built on live retrieval this works; confusion baked into training data fades more slowly.

Why do AI models mix people up?

Because a name is not an identity — it is a string, and models work with text, not birth records. When two people share a name, everything written about either of them contributes to one blurred statistical picture unless something in the text separates them. If the other Sam Wright is a footballer with a hundred times your press coverage, the model's default "Sam Wright" is the footballer, and your achievements get grafted onto him — or his onto you.

Retrieval has the same problem in real time. An assistant that searches your name gets a mixed page of results and has to guess which ones belong to which person. If no page says plainly who you are, where you are and what you do — in a form machines can parse — the guess is a coin toss weighted towards whoever has more coverage.

How do you tell machines which person you are?

You give them a canonical page and structured, distinguishing facts. Four things do most of the work. First, one stable URL that is unambiguously about you — your own site, or a profile page like the live example at i.hilyt.it/jack. Second, Person JSON-LD on that page with a sameAs array linking your real accounts, so machines can join your identities into one entity instead of several fragments. Third, a consistent one-line bio used everywhere — the repetition is the signal. Fourth, distinguishing facts the other person does not share: your location, your field, your employer, where you studied.

Here is what that looks like for a fictional Sam Wright — the designer, not the footballer:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Person",
  "name": "Sam Wright",
  "url": "https://samwright.design/",
  "jobTitle": "Product Designer",
  "disambiguatingDescription": "Product designer in Bristol, UK — not the footballer of the same name",
  "workLocation": {
    "@type": "Place",
    "address": {
      "@type": "PostalAddress",
      "addressLocality": "Bristol",
      "addressCountry": "GB"
    }
  },
  "alumniOf": {
    "@type": "CollegeOrUniversity",
    "name": "University of the West of England"
  },
  "sameAs": [
    "https://www.linkedin.com/in/samwrightdesign",
    "https://github.com/samwrightdesign",
    "https://dribbble.com/samwrightdesign"
  ]
}
</script>

Every field is doing disambiguation work: jobTitle and the address separate the designer from the footballer; disambiguatingDescription is a real schema.org property that exists for exactly this case; alumniOf adds a fact unlikely to be shared; and sameAs ties the page to accounts only the real Sam controls. For the full walkthrough of the markup itself, see the Person JSON-LD example guide. One caveat on sameAs targets: link your LinkedIn by all means, but do not rely on it as the canonical source — AI assistants generally cannot read LinkedIn profiles behind the login wall.

Then make the surrounding pages agree. The visible text on your site should carry the same facts as the JSON-LD, and your bios on other platforms should match it word for word where possible. Machines resolve identity by cross-checking; give them the same answer everywhere.

Does this actually work?

For retrieval-based answers, yes — over weeks rather than hours. When an assistant searches your name mid-answer, a canonical page with structured identity markup gives it a clean way to pick the right person, and consistent cross-platform bios confirm the pick. This is the path you can influence directly, and it is where most day-to-day confusion happens.

Confusion baked into model weights is slower to shift. If the training data blurred two people together, that blur persists until future models train on a web where the two are clearly separated — which is one more reason to publish the separation now. We are candid about timelines because we measure them: our own scoreboard at hilyt.it/geo currently records that models do not yet cite hilyt profiles for the queries we track. Disambiguation removes the reasons machines mix you up; it does not compel any particular answer. Necessary, not sufficient — but without it, the coin toss continues.

Create your AI-readable profile

Related guides

All guides · Free AI visibility checker · hilyt.it