Why your AI twin should say “I don't know”

Ask a language model something it doesn't know and, left to its defaults, it will answer anyway. Fluently. Plausibly. Wrongly. For a general-purpose assistant that's a known hazard; for an AI twin — software that answers in a specific person's name — it's disqualifying.

If your twin invents your project rates, your availability, or your opinion of a competitor, the damage lands on you. The visitor doesn't remember that an AI got it wrong. They remember that you did.

The bar is different for twins#

A general chatbot is graded on helpfulness: did the user get an answer? A twin is graded on fidelity: did the user get your answer? Those two goals point in opposite directions precisely at the moment the material runs out. Helpfulness says: fill the silence. Fidelity says: the silence is the answer — "I haven't written about that" is true, and anything else isn't.

That's why we treat the designed "I don't know" moment as hiy's most important feature. When a visitor asks something outside your material, your twin says so plainly, in your tone, and offers what it can speak to. No smooth improvisation, no papering over.

The gap is where the twin gets better#

Here's the part that surprised us building it: the honest gap isn't just harm-prevention. It's the growth loop.

Every "I don't know" is a real question, from a real member of your audience, that your content doesn't answer yet. hiy queues it for you. You answer once — a couple of sentences is enough — and that answer becomes an authoritative source: your twin uses your words from then on, and your correction outranks anything older that contradicts it.

Run that loop for a few weeks and something interesting happens. Your twin's knowledge stops looking like "what I happened to have written" and starts looking like "what my audience actually asks." The gaps are the content strategy. Competitors bolt "unanswered questions" reports onto chatbots as an analytics feature; in a twin built around honesty, the report generates itself, and every entry comes with a one-tap way to close it.

"Won't refusing make it look dumb?"#

The fear is understandable and, in our experience, backwards. Visitors don't lose trust when a twin says "that's not something Sam has covered — but ask me about pricing strategy, that I can answer properly." They lose trust the first time they catch a confident answer being wrong — and after that, every correct answer is suspect too. Trust compounds; so does its opposite.

An expert who says "I don't know, let me find out" reads as more credible, not less. The same is true of their twin — with the advantage that the twin actually will find out, because the question goes straight to the person who can answer it.

You can watch the honest gap happen live: ask Demo Sam about his refund policy. He doesn't have one written down. He'll tell you exactly that.