Repair an existing page around one reader decision

Choose one live page and one repair outcome

Intent repair works best as page surgery, not a simultaneous rewrite of the domain. Select an existing URL with a concrete symptom such as a misleading first screen, a buried answer, or a call to action that outruns the explanation. Preserve the current title, description, rendered opening, headings, events, and a dated sample of queries that reached the page. Write one repair outcome in reader terms—for example, ‘a visitor comparing tracking methods can identify the relevant trade-off before choosing a tool.’ Do not define success as ‘use the keyword better.’ Google’s people-first guidance and starter documentation support content organized for people and described clearly, while neither source promises a ranking gain from a rewrite. The bounded deliverable is an intent-to-section trace linking each visible promise to the exact section that fulfils it.

Evidence: Google Search Central; GOV.UK Service Manual

Preserve the page before surgery

Before touching copy, identify who can approve evidence, commercial disclosures, analytics definitions, and rollback. A preserved version prevents the repair from becoming an irreversible opinion.

Evidence: Google Search Central; World Wide Web Consortium

Sample language without treating it as mind reading

Collect a small set of page-level queries from a stable period, questions from support or community conversations, and the terms used by actual readers during a task interview. Group them by decision rather than by shared nouns. ‘What is last-click attribution?’ asks for a definition; ‘last-click versus data-driven attribution’ asks for comparison; ‘why did this order receive no commission?’ begins a diagnosis. The words overlap, but their completion states differ. Mark uncertain or mixed cases instead of forcing them into the dominant group. Search Console data is sampled and governed by platform definitions; it tells you what was reported, not why an individual searched. The working group should choose the job the page is qualified to serve, then write a result promise narrow enough that the excluded jobs are not baited into the click.

Add the chosen job and three excluded jobs to the trace so later optimization cannot silently broaden the page.

Evidence: Google Search Central; GOV.UK Service Manual

Place the usable answer before the evidence tour

Rewrite the opening in three moves: recognize the reader’s situation, state the bounded answer, and preview the evidence or steps that follow. Remove a preamble that merely repeats the title, recites the industry’s importance, or delays the answer until after a product block. Then mark the earliest sentence where the promised decision becomes possible. If it lies several screens down, decide whether context can move after the answer or whether the title promises too much. This does not mean every page should be short. Complex advice may require definitions and safety qualifications before action, but those prerequisites should be named rather than hidden. W3C’s headings guidance supports descriptive labels that let readers locate information and understand relationships. Use those labels to expose the page’s reasoning rather than to repeat a target phrase mechanically.

On the trace, connect the title to the opening answer and each prerequisite to the heading that supplies it. Any dangling promise becomes an edit ticket.

Evidence: Google Search Central; Google Search Central

Make every section earn its place

For each section, write the question it resolves and what a reader can do differently afterward. Keep it when that answer advances the chosen job, move it to another page when it serves an excluded job, and delete it when it exists only to lengthen the article. Map substantive claims to dated sources and add limitations near the claim rather than in a generic disclaimer. A product reference must answer a real comparison or implementation question; disclosure does not convert an irrelevant offer into useful content. Next inspect the final invitation. Someone who has just learned a concept may reasonably continue to a worked example, while someone who has completed a comparison may be ready for a decision worksheet. The CTA should not manufacture urgency or assume a purchase state that the article never established.

This section-by-section test makes the trace an operational artifact instead of a decorative outline.

Evidence: GOV.UK Service Manual; World Wide Web Consortium

Illustrative repair of an attribution explainer

Suppose a page titled ‘How affiliate attribution works’ opens with a tool recommendation, mixes cookie windows with analytics attribution models, and ends by urging readers to switch platforms. The repair outcome is narrower: explain which party records a referral, which rules can assign commission, and what evidence a publisher can preserve when a transaction is missing. The opening states that these systems differ and names the limits of publisher visibility. One section defines network tracking; another separates analytics reporting from merchant commission rules; a troubleshooting continuation points to a missing-order record. The product block moves to an optional implementation comparison whose criteria are explicit. No performance improvement is asserted. The intent-to-section trace simply shows that every visible promise now has an answer owner and that the final action matches the reader’s achieved state.

A later task check may contradict the design, which is why the preserved version and review date remain part of the repair.

Evidence: Google Search Central; Google Search Central; GOV.UK Service Manual

Publish the smallest reversible repair

Have an uninvolved reviewer read only the result title and first screen, state the expected job, and then use the page to complete a representative task. Record mismatches and evidence questions rather than coaching. Validate final headings, links, disclosure, and measurement events before release. The next action is one bounded edit package containing the dated baseline, intent-to-section trace, changed sections, and rollback owner. Observe an appropriate period, but do not call a ranking or conversion movement causal: demand, result presentation, competitors, consent, and other releases can change. A page may legitimately serve a mixed audience through clear branches; forcing it into one taxonomy label can be less useful than explicit routing. Stop when the chosen job is fulfilled and excluded jobs have honest destinations. Reopen the trace when query language, support questions, or direct task evidence reveals a recurring need the repaired page no longer handles.