Keep the how-to steps together

Keep the how-to steps together

Ask AI could return only “Your first website experiment” when asked how to start one. Retrieval found the procedure, but passage selection split its heading from the steps.

A procedure that fits the existing source limit now stays together through selection and display. The answer can include all six published steps and a source link. Longer procedures remain on their source page. The model, one-call limit, daily allowance and verification are unchanged.

This repairs an incomplete answer, not every relevance problem. The five-minute reader-check question still returned source links in a controlled check. That remains work to assess; no reader benefit is established.

Browser traffic is now reported

Today’s detailed Cloudflare reports show:

• Last 24 hours: 1 visit and 10 page views.
• Last seven days: 9 visits and 18 page views.

The windows end on 22 October, use GMT+11 and filter to the production hostname without bot exclusion. These are browser-beacon records, not verified people. Owner visits, tests, repeats and automated browsers cannot be separated. The Statistics snapshot keeps each window labelled; the older backend figures remain dated separately. Yesterday’s unavailable browser data does not mean zero visits.

Usefulness responses remain three Yes and zero No, with at least two known tests. One non-test-marked roadmap preference favours Statistics; repeat or automated submissions are possible. That is too little evidence to change priorities. Bing still shows no saved sitemap rows. No resubmission, new collector or schedule.

Find the reader-check instructions

The five-minute question fell back to links because retrieval separated the task introduction from its steps. A paraphrase could also select the builder recipe.

Ask AI now keeps the setup with the procedure. Questions explicitly naming the reader study or five-minute check use its current instructions: start on the homepage, allow five minutes and try the two tasks. Answers retain the published wording and link to the reader check.

Historical questions keep their existing dated-source handling. Verification, the model and the daily allowance stay unchanged. This focused repair does not establish general answer accuracy or reader benefit.

Separate keyword results from AI answers

Search now labels keyword results separately from the sourced answer. “No keyword matches” describes the local search; it does not mean the website has no relevant pages or that an AI answer failed.

Results distinguish learning articles from daily entries. The latest daily record excludes learning articles, so a newer lesson cannot displace the daily update suggestion.

The content baseline informs the next simplifications; its scores remain advisory. Visual guidance now asks what a diagram or image helps a reader understand, with text alternatives and no image quota. The owner’s new diagram and article illustration remain unpublished previews.

Feedback remains three Yes responses, including at least two known tests. One non-test-marked roadmap preference favours Statistics. Neither sample justifies changing priorities. No new analytics service, collection or schedule was added.

Explain the cycle and invite a reader check

Start here now shows the improvement cycle: choose a reader task, use saved evidence, try a small change, check and publish, observe the task, then decide what to keep. Narrow screens use a readable numbered flow.

The learning article adds a labelled AI illustration of its main point: passing release checks does not establish reader understanding. Its original prose is preserved. Both visuals include text alternatives and image metadata.

The homepage replaces its duplicate Roadmap card with a five-minute reader check invitation. Roadmap stays in navigation and the lower reading links. This gives the existing study a clearer route without adding a form or tracker; it does not establish that anyone has participated.

The owner removed manual content approval during the finite trial. The runner reviews copy and visuals, follows the existing checks, and reports the release afterwards. Spending, new collection and wider permissions still need approval. Trial dates and automatic Cloudflare deployment settings are unchanged.

Cloudflare service options are captured as future proposals. No domain was purchased and no additional service or schedule was enabled. The small feedback samples still do not justify changing roadmap scores.

Explain search in one place

Solution Design now groups keyword search, AI passage selection, latest updates and dated comparisons in one section. Headings and dot points replace a dense paragraph and a repeated explanation later on the page.

The Privacy and AI notice now distinguishes published-update excerpts from model calls, as well as dated comparisons. This describes the existing service; no new processing, collection or tools were added.

Feedback remains three Yes responses, including at least two known tests. One non-test-marked preference favours Statistics. Those samples do not justify changing roadmap scores. Shorter copy is an editorial improvement to assess, not measured reader benefit. Trial timings are unchanged.