Guide
AI for product discovery: from raw notes to a decision
Discovery produces a lot of unstructured material: interview transcripts, support tickets, survey comments, notes from sales calls. AI can help you read across it, as long as the link between evidence and conclusion is not lost on the way.
Synthesis first, solutions second
The most common mistake is to ask for feature ideas straight away. Start by asking what the material says. Upload the transcripts and tickets, then create a branch for each problem area and ask what evidence supports it, how often it appears and who it affects. Ask for quotes and where they came from, so you can check them.
Treat solutions as branches of a problem
Once a problem is worth solving, fork from it for each candidate solution. Each fork inherits the evidence above it, so the discussion of scope, risk and effort stays tied to the problem it is meant to solve. Three solution branches under one problem are easier to compare than three ideas scattered through a long chat.
The canvas is the decision record
Teams often write the decision document after the fact, from memory. On a canvas the record already exists: the problems considered, the options explored under each, and the point where one was chosen. Merge the chosen path into a summary for the write-up.
When you share the project, reviewers can open the branch behind any statement. That tends to shorten the “why not X” part of the review.
A caution
AI summarises what it is given. It cannot tell you what your interviews did not ask. Use it to organise and question the evidence, and keep the judgement about what to build with the people accountable for it.