How to use these playbooks
Work top-down through the pipeline — eligibility and candidate selection before chunk edits. Jumping straight to rewrites when the page never enters the candidate set wastes effort.
Playbook 1: Competitor cited, you are not
Symptoms: Prompt runs show competitor URLs in citations; your domain absent. Visibility near 0% for that prompt.
Ask:
- Are you in the answer at all? Check brand mentions. If named but not cited, the problem may be mention positioning — not page structure.
- Is any of your site cited for related prompts? If never, suspect candidate page selection — weak page-level aboutness signals, diluted topical intent, or authority filter excluding the domain for this query type.
- Is a relevant page cited elsewhere but not here? Query-dependent selection — you may need a dedicated URL for this intent, not a homepage or generic hub.
- Open cited-page eval for your closest URL. If Evergreen Health is low, move to Playbook 3.
Likely actions: New or refocused landing page for the intent · improve title/H1/page centroid · fix indexation/crawl issues · copywriting on the best candidate URL.
Playbook 2: Visibility up, owned citations flat
Symptoms: Brand mention quality or visibility trend improves; your URLs still rarely linked.
Ask:
- Split mention quality from citation quality in prompt detail.
- Are mentions positive but passive ("Brand X also offers…") without a link? Optimise for citable passages, not just name inclusion.
- Check which competitor URLs get cited — often FAQ chunks, comparison tables, or definition paragraphs.
Likely actions: Add self-contained, quotable sections · improve chunkability · ensure the cited-worthy content is on the page you want linked, not buried in nav or tabs.
Playbook 3: Page cited (or should be) but scores are weak
Symptoms: URL appears in cited pages list; Evergreen Health below 6; Precision or Recall flagged in page tables.
Ask:
| Pattern | Likely issue | First move |
|---|---|---|
| Low Semantic Recall, decent Content Precision | Missing subtopics or thin coverage | Expand structured sections; fill topic gaps |
| Low Content Precision, decent Semantic Recall | Generic or rambling sections | Tighten passages so each chunk makes one idea |
| Low Content Uniqueness | Commodity content | Add proprietary proof, data, or workflows |
| High Topical Opportunity | Stale claims or dated references | Refresh facts, dates, product details |
| Low Questions Answered | Topic covered but not in query language | Align headings and FAQs to how people ask |
Likely actions: Open Copywriting for that URL · generate reviewed plan and cards · work pinned items · re-evaluate after meaningful accepts.
Playbook 4: Strong page scores, weak prompt visibility
Symptoms: Evergreen Health 7+ on evaluation; prompt-level visibility still poor.
Ask:
- Wrong page for the query — evaluation scores the URL; prompts may require a different URL entirely.
- Candidate selection never selects this page — authority or page-level aboutness mismatch. Rankings and traffic do not guarantee generative candidacy.
- Platform split — strong on one AI provider, absent on another. Filter prompt views by model.
- Sample size — few runs yet; wait for scheduled cadence before restructuring strategy.
Likely actions: Map prompts to intended landing URLs · create query-specific pages · broaden prompt set to test adjacent intents.
Playbook 5: Topic-level prioritisation
Symptoms: Too many weak prompts; limited content team bandwidth.
Sort topics by:
- Business importance — revenue-adjacent intents first
- Visibility gap — high competitor presence, low yours
- Cited-page health — you have a URL in the game but scores are fixable
- Trend direction — declining over 4+ runs beats static low
Avoid optimising prompts where you lack any plausible candidate page — fix page strategy first.
When to trust a signal
| Signal | Guidance |
|---|---|
| Single run | Hypothesis only — read the full answer log |
| 3–5 runs, same direction | Worth investigating |
| Scheduled cadence, weeks | Prioritise in content roadmap |
| Re-eval delta after publish | Strongest page-level confirmation |