AdAce AdsTriage search intent using business fit, not a list of forbidden words
A query is a short description of a task, not a complete customer profile. Calling every search without a conversion irrelevant can remove useful demand. A better first pass asks whether the business can satisfy the task, what would count as progress and what evidence is missing. The following triage is an editorial method for review before any keyword changes.
Three dimensions of a search task
Use three independent dimensions. Service fit asks whether you actually sell the requested product or solve the problem. Decision stage distinguishes immediate action, comparison, education and support. Commercial feasibility checks the service area, delivery model, minimum order and other real constraints. A query can be educational but relevant, or urgent but impossible to fulfil. Keeping the dimensions separate avoids one simplistic good-or-bad label.
Consider a fictional equipment rental business as an educational example. 'Rent floor cleaner this weekend' describes a transaction. 'Floor cleaner rental versus purchase' describes a comparison you may be able to support with a useful page. 'Floor cleaner replacement motor' requests a product the business does not sell. 'How to use a rented floor cleaner' may be customer support rather than new demand. These labels are hypotheses about the task; they do not prove the identity or eventual value of the searcher.
Give ambiguous queries an investigation lane. 'Cheap floor cleaner' could mean purchase or hire. Look for the ad shown, destination content and the aggregate outcomes of similar tasks. Record whether the current offer answered the query. If people wanted weekend collection and the page only described monthly commercial contracts, the failure might be your message or product fit rather than a word that must be excluded everywhere.
Google documents that Search negative match types have distinct blocking behaviour. Classify intent first and examine any potential exclusion second. The meaning of a query and the reach of a proposed negative are different questions. An intent label should not silently become a campaign or account-level action, especially when the same word is useful to another service.
Support, investigate or review an exclusion
- Support: the task fits the business and the page can answer it. Keep the reason and the expected business outcome visible, rather than treating every click as equally valuable.
- Investigate: fit is plausible but evidence is incomplete. Name the missing information and a review date, so uncertain demand is not permanently stranded in a spreadsheet.
- Exclude candidate: a confirmed mismatch exists. Document the fulfilment constraint and the narrow scope to review; a poor recent CPA alone is not that confirmation.
Run an intent review that another person can repeat
- Ask sales or operations which requests they can fulfil. Translate their answer into explicit service, location and offer criteria before inspecting individual searches.
- Take a dated sample that includes high spend, successful outcomes and ambiguous queries. Selecting only apparent waste produces a biased picture of intent.
- Assign the three dimensions and the triage lane. Allow 'unknown'; include a short reason and the evidence date so another reviewer can challenge the judgement.
- Compare similar tasks against qualified outcomes after enough time for processing. Keep unmatched or missing outcomes separate from confirmed unqualified enquiries.
- Ask a second reviewer to classify a subset independently. Resolve disagreements by improving the business criteria, not by counting which analyst is more confident.
- Turn the findings into a page improvement, a bounded test or an exclusion review. Make the next decision reversible where possible and measure business outcomes, not only cleaner-looking query rows.
Keep classification separate from action
- Search reporting can be incomplete. Present the spend covered by the review and leave an explicit residual rather than assigning unseen queries a made-up intent.
- This process does not justify a universal ban on terms such as free, jobs or reviews. Their meaning depends on the business and the complete query.
- A read-only AI request may help suggest labels, but a business owner should review ambiguous cases. Request text analysis only, without propose_change or tool writes.
Sources and further reading
Try it on your own accounts
Create a workspace, connect Google or Meta in a couple of clicks and see your accounts clearly. Changes follow your approvals or the policy you configure.
Create your workspace