indranilbanerjee/seo-drift

/digital-marketing-pro:seo-drift

Purpose

Take two snapshots of SEO performance data — separated by weeks, a Core Update, a content refresh, or an algorithm change — and produce a structured drift report: top gainers, top losers, classifications (growth / decline / reshuffle / stable / new / lost), and diagnostic patterns. Works with classic GSC, the new GSC AI Performance Report, rank-tracker exports, and aeo-audit probe results.

Context efficiency

Heavy skill. Grep before Read any referenced file, then Read only matched ranges with offset + limit. List ${CLAUDE_PLUGIN_DATA}/<brand>/ before opening files. On re-invocation mid-session, skip files already in context.

When to Use

• Monthly performance review — last month vs the month before
• Core Update triage — pre-update vs post-update + settling window (use 14+ days after rollout-complete)
• AI Mode citation tracking — quarter-over-quarter aeo-audit outputs to see which queries gained / lost AI Mode citations (Google AI Mode citation diff is a leading indicator for organic decline)
• Content refresh attribution — before vs after a planned content update to attribute lift to the refresh vs other factors
• GSC AI Performance Report — month-over-month deltas on the new (3 Jun 2026) combined AI Overviews + AI Mode report
• Site migration audit — pre-migration baseline vs post-migration settling

Don't use for single-point-in-time analysis (use the source skill — seo-audit, aeo-audit, gsc-ai-performance).

Brand context (auto-applied)

1. Read ~/.claude-marketing/brands/_active-brand.json for the active slug, then load ~/.claude-marketing/brands/{slug}/profile.json
2. If no brand exists: ask "Set up a brand first (/digital-marketing-pro:brand-setup)?" — or proceed with defaults
3. Apply skills/context-engine/industry-profiles.md for industry-specific noise thresholds (YMYL industries should use higher --noise to filter out routine Quality Rater Guidelines volatility)

Inputs

| Input | Source | Required? | |---|---|---| | Baseline CSV | Older snapshot | yes | | Current CSV | Newer snapshot | yes | | Join keys | Auto-detected (query, keyword, page, url) or --join-on flag | optional | | Noise threshold | --noise (default 5%) — % below which a metric is "stable" | optional | | Top-N | --top (default 20) — gainers/losers per metric | optional |

Both snapshots must come from the same source. Mixing a GSC export with an Ahrefs export will produce nonsense — different sources count different things.

Process (10 steps, numbered-file output)

All outputs go to ${CLAUDE_PLUGIN_DATA}/{brand}/seo/seo-drift/{YYYY-MM-DD}/.

1. 00-input.md — capture baseline date range, current date range, source (GSC / GSC AI / rank-tracker / aeo-audit), brand context
2. 01-baseline.csv — copy baseline export here (so the drift run is reproducible months later)
3. 02-current.csv — copy current export here
4. 03-drift-run.json — run the script:

   python "scripts/seo_drift.py" \
       --baseline "${CLAUDE_PLUGIN_DATA}/{brand}/seo/seo-drift/{date}/01-baseline.csv" \
       --current  "${CLAUDE_PLUGIN_DATA}/{brand}/seo/seo-drift/{date}/02-current.csv" \
       --top 30 --noise 5 \
       --out "${CLAUDE_PLUGIN_DATA}/{brand}/seo/seo-drift/{date}/03-drift-run.json"
   

5. 04-quality-scorecard.md — read quality_scorecard from 03-drift-run.json. If status: needs_review, diagnose:
   • date_range_distinct: warn → script couldn't auto-validate. Manually confirm in 00-input.md that baseline and current cover non-overlapping windows.
   • sample_size: fail → either input has < 50 rows. Re-export without row limits.
   • metric_compatibility: fail → no numeric metrics in BOTH inputs. Column-name mismatch — re-export from the same source.
   • no_lookup_collisions: fail → duplicate keys in one input (e.g., same query × page row twice). Re-export with deduplication or use --join-on to add a distinguishing column.
6. 05-biggest-gainers.md — narrative on the top 10 gainers across impressions / clicks / position. For each: hypothesis on cause (new content? backlinks gained? Core Update favoured E-E-A-T? Featured Snippet rotation?). Hand off candidates to /digital-marketing-pro:content-engine for amplification.
7. 05-biggest-losers.md — narrative on the top 10 losers. For each: triage matrix — is_yMYL × had_recent_change × Core_Update_window → action (refresh content / restore reverted change / wait for next algo cycle / accept and reallocate).
8. 06-ai-mode-shift.md (only if input source is GSC AI Performance Report) — queries that LOST AI Mode impressions are a leading indicator. Cross-reference with /digital-marketing-pro:aeo-audit to verify citation loss in synthetic probes.
9. 07-classification-distribution.md — counts table:
   • growth / decline / reshuffle / stable / new / lost
   • If >40% in decline: likely Core Update or competitor catch-up. Run /digital-marketing-pro:seo-audit for diagnosis.
   • If >20% reshuffle: likely intent shift (AI Mode reweighting). Run /digital-marketing-pro:aeo-geo to align with new intent patterns.
10. PLAN.md — single-page summary: stats + scorecard + top 5 actions ranked by impact × effort, with owner suggestions (SEO lead / content lead / dev team).

Output format

${CLAUDE_PLUGIN_DATA}/{brand}/seo/seo-drift/2026-06-04/
├── 00-input.md
├── 01-baseline.csv
├── 02-current.csv
├── 03-drift-run.json
├── 04-quality-scorecard.md
├── 05-biggest-gainers.md
├── 05-biggest-losers.md
├── 06-ai-mode-shift.md       (only when input is GSC AI Performance Report)
├── 07-classification-distribution.md
└── PLAN.md

Quality scorecard (the four gates)

| Gate | What it checks | Why it matters | |---|---|---| | date_range_distinct | Baseline and current cover non-overlapping windows | Overlapping windows produce false-positive deltas — same data on both sides | | sample_size | Each input has ≥ 50 rows | Below this, drift is noise | | metric_compatibility | ≥ 1 numeric metric exists in both inputs | If columns differ (e.g., Ahrefs vs GSC), there's nothing to compare | | no_lookup_collisions | No duplicate keys within an input | Duplicates make the delta math ambiguous |

status: ready requires sample, metric compatibility, and no-collision gates pass (date-range-distinct is warn not fail — the script can't always autodetect dates).

Classification rules

Each row in the report falls into one bucket:

| Classification | Trigger | Interpretation | |---|---|---| | growth | ≥ 2 metrics moved up > noise%, no metric down > 10% | Clear win — investigate for amplification | | decline | ≥ 2 metrics moved down > noise%, no metric up > 10% | Clear loss — triage by YMYL × Core-Update-window | | reshuffle | Significant moves in opposite directions (e.g., impressions up, position down) | AI Mode signature — content is being shown more broadly but for slightly different intents | | stable | No metric moved more than noise% | No action | | new | Absent in baseline, present in current | New content or new SERP coverage — track | | lost | Present in baseline, absent in current | Content removed, deindexed, or fell out of tracking window |

Position is special: for position, lower numbers are better. The script automatically inverts position-delta direction for gain/loss ranking — you'll see -85.9% under position as a top gainer (page moved from position 12 to position 2).

Chain handoffs

This skill is typically a consumer + diagnostician:

1. /digital-marketing-pro:gsc-ai-performance or seo-audit or aeo-audit — generates the snapshots
2. /digital-marketing-pro:seo-drift — this skill
3. Branch by finding:
   • High decline → /digital-marketing-pro:seo-audit for technical-side check + /digital-marketing-pro:content-decay-scan for content-side
   • High reshuffle → /digital-marketing-pro:aeo-geo for intent realignment
   • High growth → /digital-marketing-pro:content-engine for amplification briefs

Tips & caveats

• Don't run during a Core Update rollout. Wait until Google announces "rollout complete" + 7–14 days of settling. Mid-rollout deltas are unreliable.
• Position deltas are noisier than impression/click deltas — pages bouncing between positions 8 and 12 produce ±30% position deltas that mean nothing. Trust impression/click moves more for diagnosis.
• GSC's data lag is ~3 days. When pulling "current month" data, use the date range ending 3 days ago, not yesterday.
• The GSC AI Performance Report (3 Jun 2026) has NO click data. drift on AI report = impressions-only drift. Don't try to compute CTR drift from it.
• For Core Update triage, run drift twice: pre-update vs day-after-rollout-complete (the "blast"), and pre-update vs 14-days-after (the "settled state"). The two often disagree, and the 14-day view is the one that matters.
• Reshuffle classification is a leading indicator — when reshuffle counts spike, intent reweighting is happening. The next quarter's drift will usually show clearer growth/decline. Don't react too fast.

Agents used

• analytics-analyst (primary) — interpretation + cause hypotheses
• seo-specialist — for technical-cause hypotheses on losers
• competitive-intel — when decline correlates with a competitor's win
• market-intelligence — for algo-update context (was there a Core Update in the window?)

See also

• /digital-marketing-pro:gsc-ai-performance — pull the GSC AI Performance Report (input source)
• /digital-marketing-pro:seo-audit — diagnose decline causes
• /digital-marketing-pro:aeo-audit — diagnose AI Mode citation loss
• /digital-marketing-pro:content-decay-scan — for content-side decline triage
• /digital-marketing-pro:content-engine — for amplifying gainers
• scripts/seo_drift.py — the underlying drift engine