What is an AI twin — and what makes one trustworthy?

An AI twin is an AI version of a specific real person, built from that person's own material — their writing, talks, videos, and notes — that answers questions the way they would. Where a chatbot represents a company or nothing at all, a twin represents someone: their knowledge, their positions, their way of phrasing things.

That "someone" is exactly what makes twins useful, and exactly what makes them dangerous to do badly.

The three ways AI twins go wrong#

1. They make things up in a real person's name. A generic chatbot that hallucinates is annoying. A twin that hallucinates puts invented opinions, prices, and commitments into the mouth of a real human being who then has to live with them.

2. They're built without consent. If anyone can publish "an AI version" of anyone else, the category collapses into impersonation with better grammar.

3. They hide what they are. A visitor who doesn't realise they're talking to software has been deceived, however good the answers were.

Any product in this category should be judged on how it handles those three failure modes — ours included.

What a trustworthy twin looks like#

These are the mechanics we build hiy around, and the checklist we'd suggest for evaluating any twin, including ones that aren't ours:

• Grounding with receipts. Answers come from the person's actual material, and the twin can show the exact passage behind a claim. If a twin can't cite, it's improvising.
• An honest "I don't know." When the material runs out, the twin says so — in the person's tone — rather than guessing. The unanswered question goes to the person, whose answer teaches the twin permanently.
• Corrections that win. When the person fixes something, the fix overrides every older source that disagrees with it. A twin that can't be corrected can only decay.
• Identity, enforced. You may only twin yourself or a product you own. Public twins are reviewed before they're reachable, and every page carries a report link.
• Disclosure, always. Every page says it's AI, that it may make mistakes, and that the creator can read the conversation. Quietly, but never absent.
• The person stays in charge. Read everything it knows, unpublish instantly, export or delete everything. The twin is theirs, not the platform's.

What an AI twin is not#

It's not a replacement for the person — it's the person's material, made conversational. A good twin says "that's yours to decide" when asked to commit to something, routes stuck visitors to the human, and treats "check with the real me" as a feature, not a failure.

If you want to see the difference in practice, talk to Demo Sam — including asking him something his material doesn't cover — or read how hiy works under the hood.