Search terms cleanup

Updated November 12, 2026

What the search terms cleanup routine is

The cleanup routine is the weekly loop behind negative keywords: pull the last 7 days of search terms, find the queries wasting money, cluster them by intent, propose negatives at the right level, apply after approval, log.

It exists because the gap between the keywords you bid on and the searches you actually pay for has grown every year. Broad match, close variants and Performance Max all widen the match between your keywords and real queries, so every Search account continuously accumulates junk matches, and the search terms report is the only place they are visible.

Cleanup is the classic highest-return maintenance task in Google Ads: pure waste removal, measurable in dollars per week, and mechanical enough that skipping it is purely a time problem. Which is exactly the profile of work that should be automated with an approval gate.

What the search terms report shows, and hides

The search terms report lists the real queries that triggered your ads, with each query's impressions, clicks, cost and conversions, alongside the keyword and match type that caught it. It is the ground truth of what you are buying, as opposed to the keywords, which are what you asked for.

It also hides things, and knowing the gaps keeps you honest. Google omits queries below privacy thresholds, so a slice of your spend sits in terms you cannot see individually; the visible report is a large sample, never the full census. Performance Max reports search themes and category-level insights rather than a full per-query view, which is one reason PMax waste is harder to catch by hand.

The practical conclusion: clean what is visible weekly, and treat persistent spend-without-results in the invisible remainder as its own diagnostic signal, often pointing at match types or PMax settings rather than individual queries.

Where junk queries come from

Knowing the sources explains why cleanup never ends:

Broad match matches on meaning, as Google models it, so 'crm software' can catch 'what does a sales team do all day'. Smart bidding reins this in over time when conversions are tracked well, but exploration is the design, and you pay for the exploration.

Close variants affect even exact match: plurals, misspellings, reorderings and 'same meaning' rewrites. Mostly fine, occasionally a meaning drift that costs real money.

Performance Max matches across Search among its channels, and its query-level controls are coarser, so wrong-intent matches accumulate quietly.

Your own ambiguity: words like 'trainer' (shoes? person?) or brand names that collide with common words invite junk structurally. These are the cases where exact negatives earn their keep.

The weekly loop in detail

What the routine does each week, step by step:

1. Pull the last 7 days of search terms across Search campaigns, plus Performance Max search themes where available.
2. Score each query: spend, clicks, conversions, and conversion history over a longer lookback, so a query having one slow week is not misjudged.
3. Flag two kinds of offenders: spend-without-conversions past a threshold scaled to the account, and wrong-intent families regardless of spend (jobs, free, DIY, courses, unrelated brands).
4. Cluster flagged queries by intent into families, so you review 'free-seekers: 11 queries, $48, 0 conversions' once instead of 11 rows.
5. Draft the negatives: match type and level per cluster, phrase negatives for junk phrases, exact for ambiguous single words, shared list placement for universal junk.
6. Check the drafts against safety rules: never a query with conversions, never a conflict with active keywords, never a fence wider than the evidence.
7. Deliver the proposal; on approval, apply and log. On autopilot, apply inside the same rules and log identically.

The whole loop compresses to a few minutes of your review per week, and the daily note confirms what was applied.

What a cleanup proposal looks like

Each proposal is a table of clusters, and each cluster expands to its queries. The columns carry everything needed to judge it:

You can approve a whole proposal, approve some clusters and reject others, or edit a negative's text, match type or level before approving. Edits teach you nothing is sacred: the routine drafts, you decide.

Do the cleanup manually

The manual equivalent, useful on its own and for auditing any tool's work:

1. Google Ads, Campaigns, Insights and reports, Search terms; date range: last 7 days; sort by cost descending.
2. Read down to your spend floor. For each query ask two questions: could this person buy, and did these clicks ever convert (check a 90-day window before condemning).
3. List the junk in a sheet with its spend, and group it by intent family as you go.
4. Write the negatives: phrase match for phrases, exact for ambiguous single words. Choose ad group, campaign or shared list per family.
5. Add them in the Negative keywords section (or directly from the search terms view), and attach shared lists to any campaign missing them.
6. Log the week's total blocked spend. The running number is your evidence the habit pays.

Budget a half hour weekly for a mid-size account. The search terms cleanup answer page has a condensed version of this checklist.

Mistakes that undo the value

• Judging on a 7-day conversion window alone. Queries with long consideration cycles convert late; condemning them on last week's zeros negates future buyers. Always check the longer lookback, as the routine does.
• Cleaning only the biggest campaign. Junk spreads account-wide; shared lists plus an account-wide weekly pass beat deep-cleaning one campaign quarterly.
• Writing negatives wider than the evidence. The query 'free crm for nonprofits' justifies negating 'free', as a phrase or broad negative, only if nonprofit freebie traffic never converts for you; it might. Fence what the data condemns.
• Never revisiting old negatives. Businesses change; a negative that was right two years ago can block today's product line. An annual review of shared lists catches these.
• Ignoring the invisible remainder. If totals show spend the visible terms do not explain and performance sags, the lever is match type and campaign settings, not more negatives.

How AdsGuru handles it, honestly

This routine is deliberately the easiest one to trust: its inputs are visible in your own search terms report, its proposals carry the exact queries and stats, and its actions are cheap to reverse. That is why it is the usual first candidate for autopilot after a few weeks of clean approvals, and why we point trial users at it first: verification takes minutes.

The honest caveats: AdsGuru needs its 7 to 14 days of baseline before thresholds calibrate to your account, PMax query visibility is limited by what Google exposes, and the routine removes waste rather than creating growth; it funds growth by reclaiming spend. Approval mode is the default, a query with conversions is never proposed as a negative, and every applied negative is logged with its before and after state.

Common questions

Related