How AI assistants identify people

Two sources of knowledge

An assistant answers from two places. The first is its training data: a snapshot of the web from months or years ago, compressed into the model. The second is a live web search, which ChatGPT, Claude, Gemini and Perplexity run by default for questions about people. Training data explains why old roles linger; live search is where new, clear sources make a difference quickly.

The signals that tie pages to one person

• The name, with its variants: full name, short name, maiden name, transliterations.
• The context around it: employer, job title, city, field. "Dana Levi, cardiologist in Haifa" is a different person from "Dana Levi, designer in Berlin".
• Links between accounts: a website that links to a LinkedIn profile that links back, or profiles marked with rel="me". Two-way links are strong evidence that accounts belong to the same person.
• Structured data: schema.org Person markup states a person's name, role, employer and accounts (sameAs) in a form machines do not have to guess at. An @id gives the person a stable identifier across pages.
• Knowledge bases: Wikidata and Google's Knowledge Graph hold identifiers for well-known people; most professionals are not in them.
• Source authority: a company's own team page or a verified profile counts for more than a scraped directory.

Why namesakes get merged

Most people share their name with someone. If the pages about you rarely mention your employer, city or field together, an assistant has little to separate you from a namesake, and it may blend the two. The more famous namesake usually wins, because there are more pages about them. The fix is context that travels with your name: the same title and employer on every page you control, and one explicit line that says how to tell you apart.

Why some people are missing entirely

Many professionals have most of their public record on LinkedIn, which requires a sign-in and limits crawlers. To an assistant, such a person can look like they barely exist. An open profile page that lists the same facts, and links to LinkedIn as a verified account, fills the gap.

What a machine-readable profile looks like

A SelfBadge profile gives assistants each signal in the form they read best:

• A plain-language summary at the top, and the facts again as short sentences.
• schema.org JSON-LD in the page: a ProfilePage whose main entity is the Person, with a stable @id such as https://selfbadge.com/<handle>#person.
• A Markdown version (/<handle>.md) for language models, and a JSON version with each fact's verification level.
• Only accounts that were proven to be the person's in sameAs.
• A "to tell them apart" line that names what separates the person from namesakes.
• A public API and an MCP server, so AI agents can look a person up directly.

See an example profile, read the developer documentation for the JSON and Markdown versions, or check what AI says about you today.