Ai Overview Recovery

AI Overview Recovery Playbook

When Google AIO stops citing your article, run this recovery (per seo-geo pattern).

Phase 0: CONFIRM the loss is real (v3.35 — churn guard)

AIO citations churn ~45% per render (Ahrefs 2025: consecutive AIO renders of the same query share only ~54.5% of cited URLs). A single uncited probe is noise, not a loss — recovering from churn wastes a full rewrite cycle and poisons the journal.

• The automated trigger is already guarded: refresh_decision_router emits NOT_IN_AI only after 2 consecutive uncited probes ≥ 48h apart (observations persisted in projects/{slug}/audits/aio-observations.json; an in-between cited probe resets the clock). A router-emitted NOT_IN_AI action is therefore pre-confirmed — proceed to Phase 1.
• For a MANUAL /aio-recovery invocation: check aio-observations.json first. If the loss has only one observation, re-probe after 48h instead of starting recovery: python -m scripts.monitor.refresh_decision_router --site {slug} --json
• Evidence: references/seo/serp-feature-value-2026.md §4.

Phase 1: Measure Damage (Day 0)

• Check GSC: AIO impressions drop
• Run ai_search_probe to confirm
• Document: which queries lost AIO citation? Traffic impact?

Phase 2: Diagnose Skip Reason (Day 0-2)

4 query strategy classifiers:

• (A) Freshness — content >12 months old, new competitors fresher
• (B) Authority — competitors gained backlinks; we lost
• (C) Structure — H2/H3 structure no longer matches query intent
• (D) Entity — entity ambiguity (e.g., merger renamed brand)

Run:

bashCopy code
python -m scripts.fetch.ai_search_probe \
    --brand "{brand}" --domain "{site}" \
    --queries "{affected_queries_comma_separated}"

Phase 3: Targeted Rewrite (Day 3-5)

Per diagnosis:

• (A) Add dateModified + fresh stats + 2026 data
• (B) Build authority — add citations to Tier-1, get backlinks
• (C) Restructure H2s to match query intent
• (D) Add Organization schema with sameAs + add clarification copy

Phase 3.5: Record the change (closes the monitor verify loop)

After the targeted rewrite is republished, RECORD it so the verification loop can measure whether AIO citation recovers (Rule 6: the real executor for the "fixer skills should call --record" wiring, not a prose note). The diagnosis class (A/B/C/D) is the change --type:

bashCopy code
python -m scripts.monitor.optimization_journal --record \
    --site {site} --post-id {post_id} --url {live_url} \
    --type aio-recovery --query "{primary_affected_query}" \
    --before "{old_structure_or_stat}" --after "{new_structure_or_stat}"

Phase 4: Monitor (Day 7 / 14 / 28)

• Re-probe AI engines weekly
• Check GSC AIO impressions weekly (optimization_journal --verify --site {site} --window 14 reads the recorded baseline)
• If recovered: document what worked
• If still lost at Day 28: escalate to repair-orchestrator with from-scratch level

See also

• references/geo/ai-overview-recovery-playbook.md