Skip to content
whitead

Search query mining — workflow for negatives & intent clusters

playbook google ads updated 2026.05.06 6 min read
// contents · 7 sections
  1. What "search query mining" actually means
  2. Top causes of mining decay (ranked by audit frequency)
  3. Diagnostic checklist
  4. Fix paths
  5. When to escalate
  6. Related concepts
  7. Sources

Search query mining is the disciplined habit of turning the search-terms report into a structured negatives + ad-group-split program — not a once-a-quarter purge, but a 30-90 day rolling workflow that keeps broad match honest.

What "search query mining" actually means

The search-terms report shows the actual queries Google matched to your keywords. Mining it means systematically classifying every high-impression term as converting, adjacent, or off-intent — and acting on each bucket. With broad match now Google's recommended default for Smart Bidding (Google Ads Help — About broad match), the matching surface is wider than phrase or exact ever was, so the cleanup cadence has to keep up.

Three operational definitions:

  • Irrelevant query share = (impressions on off-intent queries) ÷ (total search-terms impressions). Above 15% on commercial campaigns is a warning; above 25% is a Smart Bidding problem.
  • Mining cadence = how often you cycle the workflow. Healthy accounts run weekly on high-spend campaigns, monthly on long-tail.
  • Coverage gap = high-impression queries (e.g. ≥100 impressions/30d) that match neither a positive keyword nor a negative — these are the silent leaks.

Note that since 2020, the search-terms report only shows queries with "significant" volume — Google's privacy threshold hides ~28% of clicks for some accounts per third-party measurement (Search Engine Land — Search terms report changes, 2020-09-02). Mine what you can see; budget for what you can't.

Top causes of mining decay (ranked by audit frequency)

  1. No cadence — mining happens reactively when CTR drops, not on a calendar. By the time symptoms appear, the bidder has already absorbed weeks of noisy signals.
  2. Negatives added at campaign level only — should usually live on a shared negative list so new campaigns inherit them.
  3. Singular vs plural / close-variant blindness — Google matches close variants automatically since 2018 (Google Ads Help — About close variants); your negatives need to too.
  4. No intent clustering — engineers paste 200 individual negatives instead of grouping by theme (job-seeking, free, DIY, competitor brand).
  5. Mining the wrong window — 7 days is noise; 90 days catches seasonal drift the bidder has already learned.
  6. Confusing "added/excluded" filter with the full report — the default UI filter hides matched terms that aren't in your keyword list, which is exactly where the leaks are.

Diagnostic checklist

Run before opening the negatives editor. Stop at the first "yes" — that is where to start mining.

#CheckWhere in UIThreshold
1Irrelevant query share over 15% of impressionsInsights & reports → Search terms → 90d>15% = mine now
2Shared negative list missing or under 50 entriesTools → Shared library → Negative keyword lists0-50 = critical gap
3High-impression queries (≥100/30d) not classifiedSearch terms → sort by impressions descTop 50 should all be classified
4Mining cadence longer than 30 days on broad-match campaignsAudit log / change history>30d on broad = stale
5Job-seeking / free / DIY / competitor terms unblockedSearch terms → text filterAny present = priority adds
6Singular and plural negative coverage unevenNegative list inspectionMismatch = close-variant leak

Fix paths

Run in this order — out of sequence wastes Smart Bidding learning:

  1. Export the 90-day search terms report. Filter to campaigns over the spend threshold, export to Sheets. In Sheets, build a pivot: rows = query, values = impressions / clicks / cost / conversions. Sort by impressions descending.
  2. Cluster by intent, not by character match. Tag each top-50-by-impression query as convert, adjacent, off, or competitor. Group off rows by theme — free, job, tutorial, DIY, competitor-brand, wrong-industry. This step is what separates a 30-negative pass from a 200-negative pass that actually holds.
  3. Bulk-add negatives at the right scope. Generic off-intent themes (free, salary, job) → shared negative list. Brand-of-competitor terms → campaign-level. Adjacent-but-wrong-product → ad-group-level so the rest of the campaign keeps eligibility. See Negative Keywords for match-type guidance.
  4. Split ad groups when one keyword pulls two intents. If crm software and free crm share a group, the bidder cannot price them differently. Split, give each its negatives, and let the new structure run 14 days before judging.
  5. Automate the cadence. Optmyzr's n-gram analysis or a Sheets-import script with a weekly trigger surfaces new high-impression unclassified queries automatically; manual quarterly mining drifts within weeks (Optmyzr — search terms n-gram playbook, 2025-06-18).
  6. Re-check after 14 days. Irrelevant query share should drop below 10%, CTR lifts 15-30%, CPC drops 8-15%. If not, the problem is upstream (keyword choice, intent, landing page) — not mining.

Methodology note. Whitead's mining-readiness signal combines three measurements taken on the 90-day search-terms window: (a) irrelevant-query-impression-share computed by tagging top-200-by-impression queries against the campaign's positive keyword theme, (b) shared-negative-list size and inheritance coverage, and (c) median age of the campaign's last 50 negative additions. Accounts where (a) exceeds 15% and (c) exceeds 60 days are flagged as search_query_mining_stale with HIGH priority — the rule fires upstream of CTR, KQS, and budget-limited rules because mining cleanup typically resolves a fraction of those downstream symptoms before they need their own fix path. Threshold values are calibrated against Google's documented broad-match guidance and benchmarked against industry CTR medians; sample-size disclosure is in the per-rule audit metadata, not synthesised here.

When to escalate

Escalate beyond mining when:

  • Irrelevant query share stays above 15% after two consecutive mining passes 14 days apart — the keyword set itself is wrong (intent mismatch, not coverage gap).
  • Mining cleanup raises CTR but conversions drop — the off-intent traffic was actually converting at low margin; check landing-page intent before continuing to prune.
  • The search-terms report is mostly (other) — Google's privacy filter is hiding the long tail; switch the diagnostic to keyword-level performance and consider tightening match types.

Sources

← Back to Wiki