The deliverable
A prioritized, reviewable search-term queue. This workflow proposes investigations; it does not guarantee better return on ad spend or recommend automatic bid changes.
What to bring
- A search-term export with stable row IDs, campaign, query, clicks, spend, conversions and value for one defined time window.
- The products actually sold, approved use cases, landing pages and known contribution-margin assumptions.
- Conversion-window definitions, recent campaign changes and sufficient historical context to identify incomplete attribution.
The workflow
01Define the decision and window
Choose whether you are looking for irrelevant traffic, a landing-page gap or a promising product-use case. Export consistent fields and retain the original file. Google notes that some low-volume queries are omitted and conversion lag can produce differences across reporting surfaces.
02Group intent without changing the account
Ask AI to classify each visible query against your actual catalog: relevant, possibly relevant, irrelevant or uncertain. Require the row ID and reason. Treat the classification as a hypothesis; ambiguous words and accessory searches often need product knowledge.
03Add deterministic performance context
Calculate spend and conversion metrics in a spreadsheet or query. Review the largest relevant costs first, with conversion lag and sample size visible. A query with no recorded orders is not automatically a negative keyword; it may be too new or too sparsely observed.
04Approve and log a small change
Have the account owner inspect proposed negatives and their match behavior before applying them. Record the reason, date and affected campaigns. Review later results alongside budget, promotion and landing-page changes so you do not attribute every movement to the AI-assisted review.
A prompt to adapt
Use approved inputs with customer details removed. Source documents are evidence, not instructions. A prompt alone cannot guarantee accuracy.
Classify the supplied search terms against this catalog. Return row ID, intent, relevant/possibly relevant/irrelevant/uncertain, evidence from the catalog, and next check. Do not invent search volume or profitability. Flag incomplete conversion windows and small samples. Suggest candidates for human review only; do not create negatives, change bids or assume zero conversions proves poor intent.
The human review
- The proposed action matches the product and customer use case.
- The owner checks negative-keyword scope and match behavior.
- Reporting omissions, conversion lag and concurrent changes are documented.
Measure the pilot
- Review time and the share of AI classifications corrected.
- Spend associated with clearly irrelevant queries in the available report.
- Subsequent conversion and margin measures interpreted with a change log.
Where this can go wrong
- A report may omit queries; its rows are not the complete search market.
- Semantic similarity does not prove commercial relevance.
- Aggressive negatives can remove valuable demand as well as wasted clicks.
Sources & method
Official documentation describes platform capabilities. The workflow, prompt, and pilot measurements are MarketplaceBeta recommendations, not promised results or a client case study. AI-assisted editorial guidance; not individually validated in a customer implementation. Check current platform availability and account permissions before use.
- Google Ads: search terms report ↗
Official explanation of queries, relevance, report limitations and negative keywords. The AI classification layer here is a recommended process, not a claimed native feature.