You hear every reader

What's on from day one#

Every published site collects these signals with nothing to install:

• "Was this page helpful?" — a Yes / No bar under every page, on by default; it is the Rate this page switch under Customize ▸ Content
• The same question beside the outline — off by default; turn on Rate Page under Customize ▸ Right sidebar
• Thumbs on AI answers — when AI chat is on, readers like or dislike each answer, and a dislike keeps the question it answered
• Searches that found nothing — every query in your docs search that returned no result
• Dead ends — visits where a reader searched, asked the AI or opened three or more pages, then left with nothing
• Frustration — a page opened three times in one visit, a jump to the next page and straight back, a second search after an empty one
• What gets read — the headings readers scrolled to, reading time per page, where each visit starts and ends
• People, not bots — crawlers never count as readers, and a reader sent by an AI answer is a channel of its own

Your own testing stays out of reader metrics too: visits from a network where you have the project open in the panel are left out.

What the agent does on its own#

Each signal can wake the Docsbook agent through a ready-made card under Triggers. Switch a card on and every run follows one loop:

1. A reader signal arrives — a downvote, an empty search, a week of exits
2. The agent reads the evidence against the rules of the expertise catalog
3. It changes the docs in a pull request, or files an issue when the finding needs your decision
4. It names the number to check — later the pull request reports As predicted, No effect, Went backwards or Cannot tell

Once a change merges, its pull request shows Reads, Dead ends, Rank and AI spend for the pages it touched, after the merge against before. Those pages are measured against the pages the change did not touch, so a site-wide swing is not credited to the edit.

The catalog's First screen, Navigation, Intent and Scanning axes hold the rules for these fixes. One of them: readers spend 74% of their viewing time on the first two screenfuls of a page (Nielsen Norman Group), so an answer below them is an answer most readers miss.

See it working#

These screens in the panel show the signals above:

• Analytics ▸ Feedback — each rated page with its thumbs-up and thumbs-down counts; switch the view to Votes for every vote with its date, or narrow to Page thumbs or Chat answers
• Analytics ▸ Chat — what readers asked the AI chat, by topic, each conversation marked Answered, Dead end or Unrated
• Activity ▸ Chat ▸ Content gaps — each search that found nothing and each question the chat could not answer, as it happens
• Graph — your docs as a map; colour it by Dead ends to see which pages end visits
• Analytics ▸ Insights — traffic, sources, audience and conversions, covered in Docs analytics
• Activity ▸ People ▸ Users — one row per reader: country, the translation they read, where they came from, the goals they reached

Tell your agent#

Say it in one sentence to docsbook_agent from Claude Code, Cursor or Codex, or in the panel chat. Tell your agent, get discovered shows how to connect.

Which pages did readers rate down this month? Fix the worst one.
Turn the searches that found nothing into pages.
Find where readers give up, and give those pages a next step.
When a reader rates a page down, fix the page.
Why did traffic drop last week?

FAQ#

Next steps#