Quick answer

Review search terms by intent, not just by individual wording. Exclude themes that are irrelevant, wrong-stage, unserviceable or commercially weak, then feed the patterns back into campaign structure and landing-page decisions.

The goal is not the smallest possible query list. It is a search-term mix that produces enough qualified demand to learn and scale.

What I would check first

Start with the queries receiving spend, conversions and high impressions. Classify them as relevant, adjacent, irrelevant, competitor, research-only, job-seeking, DIY or location-mismatched.

Check whether the conversion action being optimized is trustworthy. A weak lead event can make poor traffic look productive and delay the right exclusions.

Decision framework

Exclude a theme when the business cannot serve it, the intent is clearly incompatible, the economics are unacceptable or repeated evidence shows poor downstream quality. Keep an adjacent term when it may be valuable but needs a different landing page or campaign.

Use shared negatives only for themes that are wrong everywhere. Keep campaign-specific negatives for differences in service, geography or buying stage.

Implementation steps

Export a regular search-term review, tag each row by intent, add negatives at the narrowest sensible level and record why each exclusion was made. Recheck close variants after changes because matching is intent-aware rather than purely literal.

Create a small negative taxonomy such as jobs, training, free, DIY, parts, irrelevant locations and unsupported services. This makes maintenance repeatable across accounts.

Measurement and business implications

Measure qualified conversion rate, sales acceptance and customer value by intent theme. A campaign with fewer total conversions but cleaner downstream outcomes may be healthier.

Pair this guide with the Google Ads keyword match-type guide, the Google Ads Performance Calculator and the Launch Checklist. Google’s current search systems make actual search-term evidence more important than relying on keyword lists alone.