Search term mining and negative keywords, reviewed by an AI first

Broad match and Performance Max widen reach, and they also match queries you would never bid on. An AI reviewer reads the search terms report the way a careful PPC manager would, sorts terms by intent, and proposes negatives and new keywords in a list a person approves.

01the problem and who owns it

The search terms report is where wasted spend shows up, and it is long. Reviewing it properly means reading thousands of queries, judging each one's intent against what you sell, and remembering which oddities were already handled. Most teams skim the top spenders and miss the long tail where irrelevant clicks add up.

The PPC manager owns the account, but judging intent requires product knowledge they may not have: whether a query naming a competitor, a job title, or a do-it-yourself phrase could ever become a customer.

02what the AI does, step by step

1. Export search terms with outcomesThe Google Ads API returns search terms with cost, clicks, conversions, and the keyword or asset group that matched them. Where possible, conversions are replaced with CRM qualified leads so judgment rests on real outcomes.
2. Cluster by intentThe model groups queries into themes such as job seekers, free or DIY intent, wrong product, competitor names, wrong location, and strong buyer intent. Clustering means one decision covers hundreds of similar queries.
3. Apply your business rulesA written brief describes what you sell, what you do not, service areas, and edge cases, for example that a competitor's name is welcome but its support phrases are not. The model checks every cluster against it.
4. Propose negatives at the right levelSuggestions specify match type and scope: account-level list, campaign, or ad group. Each suggestion carries the queries it would block and their spend, plus a check that it would not block converting terms.
5. Propose new keywordsHigh-intent queries that converted but are not yet explicit keywords are proposed as exact or phrase match additions, with the ad group they belong in.
6. Apply approved changes and log themAfter approval, changes are applied through the API or by hand, and logged with date and reason so the next review knows what was already decided.

03systems it connects to

• Google Ads. Search terms report and shared negative keyword lists through the Google Ads API; Microsoft Advertising works the same way.
• CRM. Lead quality data joined by click ID so negatives reflect revenue, not just form fills.
• Decision log. A sheet or table of every negative and keyword added, with the reasoning.

04human checkpoints

• Business brief approval. Sales and marketing sign off on what counts as wrong intent before the first run.
• Change approval. The PPC manager approves every negative and new keyword. Nothing is written to a live account automatically.
• Conflict review. Any proposed negative that overlaps a term with conversions in the lookback window is held for explicit review.

05what to measure

• Share of spend on irrelevant terms. Measured on each review by the clusters marked wrong intent.
• Cost per qualified lead. The account-level outcome, read over several weeks to allow for learning.
• Review coverage. Share of search term spend actually reviewed, versus top rows only before.
• Reversals. Negatives later removed because they blocked good traffic.

06risks and guardrails

• Over-blocking. Aggressive negatives can starve campaigns or block buyers who phrase things oddly. Prefer exact and phrase negatives, and always check against converting terms.
• Limited visibility. Google does not show every search term, and Performance Max reporting is more limited than Search. The process covers what is reported, not everything that ran.
• Stale rules. When you launch a product or enter a market, old negatives can block new demand. Review the shared lists after any offer change.

07build vs buy

Google's own recommendations and several PPC tools already suggest negatives, and scripts from the PPC community automate the export. For a small account, a monthly manual pass with a good script may be all you need.

A custom reviewer makes sense for accounts with heavy broad match spend, many campaigns, or intent rules that require knowing your product and lead quality in depth.

08related playbooks

Browse every marketing and paid ads playbook or the full library.