You can now read intent, write bookmarkable pages, open the technical path, and plan page systems. Search rewards none of that once — it rewards the builders who watch what happens after publishing and tend the garden. This closing lesson teaches the measurement loop: publish, observe, improve, and retire without mistaking a spike for success.

Seven numbers, each answering one question

Open Search Console's performance tab and learn each metric as a question, with its limits:

• Impressions: "How often did we appear?" High impressions with few clicks means the packaging — title, snippet, result type — does not match the intent, or a bigger brand owns the click.
• Clicks: "How many visits did search send?" The only volume number that reaches your site. Never report it without what happened next.
• Position: "Where did we typically appear?" An average, not a promise — it blends many queries, devices, and countries. Moving from 22 to 9 matters; obsessing over 4.1 versus 4.3 does not.
• Branded vs. non-branded queries: "Did they search our name or their problem?" Branded ("Sonariq pricing") measures existing awareness. Non-branded ("how to cite a 10-K") measures discovery of strangers — the growth signal for this class.
• Conversion: "Did the visit complete the page's job?" Template download, sample-brief signup, booking, next-lesson click. Define it per page in the brief (Lesson 76.2); without it, traffic is theater.
• Decay: "Is an old winner fading?" Rankings and clicks slide when facts age, competitors improve, or intent shifts. A 2024-dated guide losing non-branded clicks is decay, not punishment.
• Refresh: "Did improving the page restore usefulness?" Track what changed — new evidence, reworked example, fixed intent mismatch — and what moved within 30 days.

The funnel diagnosis from Class 67 applies directly: no impressions → nobody asks this way or nothing points to the page; impressions without clicks → title/snippet mismatch; clicks without action → page fails the bookmark test; action without return → product or follow-up gap (Class 78's email territory).

The loop: find, publish, observe, improve, retire

Treat every page as a hypothesis with a lifecycle:

find a question → publish a useful answer → observe search + behavior
→ improve the actual usefulness → retire what no longer deserves attention

• Find: one question from the intent map, Console queries, support logs, or community threads. Prefer questions with evidence of demand and a contribution you can actually make.
• Publish: ship the briefed page with its evidence, links, and next step. Inspect the URL in Console on day one.
• Observe (30 days minimum): read queries, click-through, time on page, and the page's own conversion. Do not rewrite after a week of silence — new pages need crawl and evaluation time.
• Improve: strengthen the contribution, not the keyword density. Add the missing example, update stale facts, split a mixed-intent page, sharpen the title to the winning query phrasing.
• Retire: redirect or remove pages that mislead, duplicate, or decay beyond rescue. A smaller set of maintained pages outranks a graveyard of thin ones. Consolidate near-duplicates into one canonical answer with redirects.

AI helps the loop the same way it helped drafting: grouping queries, drafting refresh notes, flagging stale facts against sources. It never decides alone what to retire — a human checks whether the page still serves a real audience before deleting a doorway people use.

Exercise: run the monthly search review

Once a month, pick three pages and write SEARCH-REVIEW.md:

# SEARCH-REVIEW.md — [Month, Project]

## Page 1 — KEEP / IMPROVE
- URL + intent + job: ___
- Evidence (impressions, clicks, position, branded share, conversion): ___
- Decision + next action + owner + date: ___

## Page 2 — CONSOLIDATE
- URLs overlapping + canonical survivor: ___
- Redirect plan + internal-link updates: ___

## Page 3 — RETIRE / REFRESH
- Why it no longer deserves attention: ___
- Action (update with ___ or redirect to ___): ___

## 30-day learning agenda
- One question to answer: ___
- One page to publish: ___
- One page to improve: ___
- One vanity number we will not optimize: ___

Worked example — Research Desk, three pages: the filing guide (1,200 impressions, position 11, 18 template downloads) → improve with a fresh worked filing and sharper title, owner plus date; two overlapping "best research tool" posts → consolidate into one honest comparison with a redirect; a 2024 "AI model news" recap with zero non-branded clicks → retire with a redirect to the evergreen method page. Learning agenda: "does the comparison page earn trial starts?" Publish one city-events pilot, improve one title, ignore total impressions as a vanity line.

Decide with this rule of thumb: keep what converts strangers, improve what attracts but does not satisfy, consolidate what competes with itself, retire what nobody would miss. Three pages reviewed honestly beats thirty glanced at.

Finish line: a SEARCH-REVIEW.md with keep/improve, consolidate, and retire decisions — each with evidence, an action, an owner — plus a 30-day agenda naming one question, one publish, one improvement, and one number you will ignore.

Verify quickly: for each decision, cover the metrics and read the reasoning aloud. "Retire because the facts aged and a better canonical exists" is a decision. "Retire because traffic is low" is a shrug — low traffic on a high-conversion page can still be worth keeping.

Common failure mode: refreshing by rewording — swapping synonyms, adding a paragraph of filler, updating the date stamp without new evidence. Searchers and engines both see through cosmetic refresh. Change the usefulness or do not touch it.

Check your understanding

1. What does each of impressions, clicks, position, and branded share tell you — and what is each one's limit? 2. Why wait about 30 days before judging a new or refreshed page? 3. When should you consolidate two pages instead of improving both?

Next

Search is now a working loop, not a lottery ticket — intent mapped, pages worth bookmarking, plumbing open, systems gated, garden tended. Class 77 takes the same useful work to social ground: choosing the platform your audience already uses, structuring the first two seconds, and repurposing ideas without copying files.