The valuable output is the list of questions you cannot answer
An FAQ written from the inside answers questions the team finds interesting. Users have different ones, clustering in predictable places: limits, pricing edges, failures, and what the product does not do. Documentation is thinnest there.
Generating questions from the documentation alone reproduces the blind spot. Generating them from someone who has not used the product does not, because that reader asks about boundaries rather than features.
Then every question gets answered strictly from the source, or not at all. A question the documentation cannot support is not a failure of the flow. It is the finding.
Two outputs emerge: a publishable FAQ and a ranked list of what the documentation is missing. The gap list decays quickly and is more valuable.
How the ActionFlow runs on the canvas
Documentation enters the flow through Document Upload or a Scrappey node when the docs are published on the web. PDF4me handles those that arrive as PDFs.
A Generate Object node produces candidate questions, instructed to write from the position of a first-time evaluator rather than a maintainer. It returns them as an array with an estimate of how common each is.
A Loop node walks through the questions. Inside it, a Generate Text node attempts an answer using only the supplied documentation and returns an explicit empty when the source does not support one.
A Filter node splits answered from unanswered questions. Answered pairs go to Notion as a publishable FAQ. Unanswered questions go to a Linear or Jira issue as documentation work, ordered by the commonality estimate.
A Validate node sits before publication and rejects any answer that contains a claim without supporting passage, which is the check that prevents a generated FAQ from quietly inventing a limit.
Running this after each significant release keeps the gap list current. The documentation changes but the questions do not. That is why the gap list is the part that moves.
Nodes this ActionFlow uses
- Document Upload (Data): ingests documentation files. Scrappey (Util) fetches published docs instead.
- PDF4me (Data): extracts contents from PDF documentation.
- Generate Object (AI Core): produces candidate questions from an outsider's position, as a ranked array.
- Loop (Control): walks the questions one at a time.
- Generate Text (AI Core): answers strictly from the source, returning empty where it cannot.
- Validate (Data): rejects answers that assert anything the source does not support.
- Filter (Data): splits answered pairs from the gap list.
- Notion (Integration): receives the publishable FAQ. Linear or Jira receives the documentation gaps.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.