AI Assistant v2 sticks to your docs

Biel.ai chatbot answering strictly from the docs instead of guessing

A chatbot that invents an answer is worse than one that admits a gap, so AI Assistant v2 ships a redesigned default prompt that stays inside what your docs actually say.

A few ways this plays out:

A user asks about a feature you haven't documented and gets "the docs don't cover this" instead of a confident guess
Answers stick to retrieved context, so they cite real pages rather than blending in the model's training data
Edge-case questions get a narrower, more honest reply rather than a plausible-sounding wrong one

This is the new default for every project, so there's nothing to switch on. If you've written your own prompt, the v2 behavior is documented in the custom prompt docs so you can fold the same grounding rules into yours.

One tip: re-test your trickiest questions after the change. Answers that used to come back confident and wrong should now come back honest, and that's the signal you want.

Seeing an answer that still drifts? Hit reply and send us the question.