Reviewing Google Ads search terms: negative keywords with evidence

The decision logic and the conflict check below document the assessment row by row; only rows that pass that assessment yield a proposal – not every row does. Anyone who scans the search terms report once a month for the most expensive rows is checking a sample; anyone who reviews the available search terms systematically also sees new relevant demand and conflicts with existing lists. maitiq relies on systematic review of the available data so that new demand and conflicts become visible on an ongoing basis.

What does the search terms report show – and what does it not?

Google separates the two objects explicitly. A search term is what a person typed in. A keyword is what you have stored in an ad group in order to reach the right audience. Google’s matching sits between them.

For positive keywords, that gap is intended. According to Google, close variants include, among other things, misspellings, singular and plural forms, word stems, abbreviations, accents, reordered words with the same meaning, added or removed function words, implied words, and synonyms and paraphrases. All keyword options – broad match, phrase match and exact match – are eligible for close variants by default; which expansions actually apply depends on the keyword option and the language. Google does not describe a way to switch this off.

The search terms report closes that gap only partially. It shows queries with sufficient activity that triggered an ad. Google omits queries with too little activity for data protection reasons. An audit must therefore distinguish clearly: “not available” is not “observed zero”. Anyone who treats omitted queries as if they never happened underestimates both off-target traffic and demand at the same time — and justifies exclusions with a completeness that the report never promised.

Which fields does a defensible decision need?

Record at least the following per row – this list is a documentation requirement of the method and not, at every point, a product field:

• the search term in raw text and in a normalised form;
• the keyword that triggered it, plus the match type of the keyword and of the row;
• campaign, campaign type, ad group and landing page;
• impressions, clicks, cost and only the primary conversion confirmed as reliable;
• the date range, the account timezone and the buffer used for conversion delay;
• business relevance, search intent and campaign role;
• existing positive and negative keywords in scope;
• proposal, evidence, conflict result, the person reviewing and the verification date.

A single value does not support a decision. Cost without conversions is a warning signal when measurement is sound and the period is mature. With broken tracking or a long sales process, the same value is almost meaningless; whether the measurement holds up is clarified by the conversion tracking audit.

Why are zero conversions on their own not grounds for exclusion?

Before a row without a conversion counts as waste, six points are checked. Not every point is a precondition for every decision: obvious irrelevance, a strategic decision and an account in click mode do not need a mature conversion window. Click and hybrid are maitiq evaluation bases, not Google campaign types; hybrid mode may still need conversion evidence whose maturity has to be awaited.

1. Measurement quality: Is the primary conversion defined, triggered and imported correctly? It underpins performance-based assessments; a purely click-based judgement does not rest on it.
2. Data maturity – only where it is needed: A conversion-based judgement needs the delay observed or configured in the account and the evidence that goes with it – not a wait for the entire maximum Google conversion window before every performance decision. Obvious irrelevance and strategic cases can be justified earlier. In click or hybrid mode the judgement rests on click and engagement evidence; hybrid mode may still need conversion evidence whose maturity has to be awaited.
3. Campaign role: Is the row part of an approved test, a brand function or a research function?
4. Relevance: Does the intent match the offer, regardless of the result so far?
5. Materiality: Are clicks and cost sufficient to support any statement at all? How waste is evidenced across several sources is shown in the guide identifying Google Ads waste.
6. Conflict with active keywords: Would the planned negative keyword hit active, valuable keywords? A competitor exclusion that has been confirmed and is exact match remains in place as a decision even when the workflow suppresses a new candidate that conflicts with it.

If a check required for the specific decision fails or remains open, the result is not a performance-based exclusion but an observation task with a next review date. What cost or CPA threshold means “enough” depends on the account, its goal and approval by the responsible person or an explicit policy; there is no universally valid value for it.

Which search terms are reviewed first?

1. Obviously irrelevant: The person is looking for something you do not offer.
2. Material spend with no dependable result: With dependable measurement, clicks and cost justify a review.
3. Recurring patterns: Several queries share the same possibly unsuitable intent.
4. New relevant demand: Suitable terms point to new keywords, ads or content.

This fourth group is easily overlooked when only the most expensive rows of a sample are reviewed. The report is not purely a source of exclusions; it describes the language that interested people actually use – and thus provides pointers to new keywords, ad copy or content.

How do negative keywords work in 2026?

Four characteristics distinguish negative from positive matching and are regularly assumed incorrectly. They describe Google’s behaviour, not maitiq’s local text comparison:

• No close variants. According to Google, negative keywords do not match close variants or other expansions. You have to add synonyms and singular and plural forms separately.
• Capitalisation and misspellings, by contrast, are handled automatically by Google.
• Accent marks: accented and unaccented terms are distinct. A negative term with an accent and the same term without an accent count, according to Google, as two different negative keywords and are treated separately, not counted twice. For German spellings, this means treating the umlaut form, the form without an umlaut and the “ae/oe/ue” variant as separate candidates until proven otherwise. That is an inference, not a documented commitment, and it belongs in the account test; the local text comparison does not normalise these spellings.
• The 16-word limit. For queries with more than 16 words, the ad can appear if the negative term only follows after the 16th word.

At which level does a negative keyword sit?

Google states up to 5,000 negative keywords per list and up to 20 lists per account; the account-level list holds up to 1,000 terms and applies automatically to eligible Search and Shopping inventory in the relevant campaign types, including Performance Max. Negative keywords in Performance Max explicitly apply only to Search and Shopping inventory. “Account level” is therefore broad, but not synonymous with “everywhere”.

Three operational details determine how useful it is: a shared list that is not assigned to any campaign excludes nothing. Negative keywords do not apply retroactively to earlier reporting periods. And the keyword diagnostic tool tests the exact form of a keyword even if that keyword is “phrase match” or “broad match”; that limitation applies specifically to this keyword diagnostic tool, not to every use of Ad Preview and Diagnosis. It does not replace a follow-up check in the report.

How does the conflict check work before every exclusion?

Before every proposal, maitiq checks the candidate against the active positive keywords and, where data is available, against the existing negative lists. These are two separate comparisons. The positive comparison lowercases the text and normalises whitespace, then uses text equality or a substring in either direction. The coverage check also normalises capitalisation and whitespace, but requires for “broad match” that every word of the existing term appears in the candidate, for “phrase match” that the existing string is contained in the candidate, and for “exact match” text equality with an exact-match candidate; it also checks whether level, list and campaigns match. Both are a local text comparison, not a simulator of Google’s matching and not synonym detection.

1. Conflict with active keywords: If the proposal would block an active positive keyword, it is narrowed to “exact match”; if it is already “exact match”, it is discarded. Narrowing reduces the reach but does not prove that no valuable query can be blocked. If it is a confirmed exact-match competitor exclusion, that decision stays in place and the run reports it as a conflict case without a new proposal.
2. Existing coverage: If an existing negative list already covers the candidate in the same scope, no new proposal is created, so no duplicate arises. A confirmed competitor exclusion also remains in place as a decision when existing coverage suppresses its new candidate. Where no dependable negative-keyword export is available, coverage is unknown – not empty and not verified; the workflow logs that gap and judges only the data actually available.
3. Limits of the comparison: The local comparison recognises no synonyms, no umlaut or accent variants and no singular or plural forms. It does not prove that no valuable query can be blocked, and Google may assign queries differently from the local text comparison. This check remains manual and with the account test.

The result reads “no conflict evidenced”, “conflict – narrowed to exact match” or “conflict – proposal discarded”. Proposing itself is read-only: the run does not change the account in this phase. An overall run, however, can separately approve and apply eligible proposals only within an explicitly configured standing authority; Autopilot is off by default and strategic cases stay with humans. This text comparison does not check Google’s matching rules – close variants, accents, singular and plural; that check remains with a human and with the account test.

When do individual terms become a pattern?

Group queries by words, meaning and intent. Several independent queries are strong evidence of a pattern, but not an unconditional requirement for every negative keyword: a clearly documented business category outside the offer can justify a broader term after a scope and conflict review. A negative phrase is supported by several independent queries, consistently unsuitable meaning, sufficient evidence, a conflict check that has been passed, a clearly limited scope, and approval by the responsible person or an explicit policy. “Free” can often be unsuitable and still be part of a truthful offer – the pattern alone does not decide. Two things stay separate here: when estimating demand volume, variants are not added together into one market size; for observed cost, by contrast, non-overlapping rows reported separately may be summed even if the terms are variants.

Which decision stages does the method provide?

The “Redirect” stage prevents a common and expensive mistake: whether price, switching or audit queries are off-target traffic or purchase-ready demand depends on the offer being advertised. Anyone who excludes them across the board at campaign level can remove demand that merely landed in the wrong place. Competitor terms are a strategy and budget question with a decision of their own: a competitor exclusion that has been confirmed and is exact match is already a bounded grant for the linked, guarded negative-keyword implementation – no further approval step is needed for it. If the term, match type, list or campaign scope changes, the appropriate new grant is required; a broad exclusion needs one in any case.

The routine for this: at a cadence that matches the volume, the report is pulled with the maturity buffer the decision requires, new queries are classified, existing lists are checked for conflicts, positive opportunities are recorded, and approved changes are verified in both directions after implementation – unwanted queries that remain and relevant demand that was lost. These six stages are a manual decision framework of the method, not the exact set of statuses implemented in the product; the product supports only the steps that are individually represented there. Proposal, decision and implementation remain separate states throughout.

How maitiq takes over this review daily

This is how maitiq checks it on an ongoing basis: negative-keyword detection (the “Negative Keywords” workflow) runs this routine daily by default – report with maturity buffer, classification of queries (own brand, competitor, irrelevant or informational), text comparison of every negative-keyword proposal against active positive keywords and, where available, against existing negative lists. Proposals for new relevant demand (for example new keywords or ad groups) come from separate growth workflows, not from negative-keyword detection. For proposed rows you receive the action, the reason and the underlying data. Proposing is read-only; within an explicitly configured standing authority a run can approve eligible proposals and implement them in a guarded way. Autopilot is off by default, and strategic edge cases stay with a human.

Frequently asked questions

Sources and how to read them

The eight sources below are Google platform documentation on the search terms report, negative keywords, lists, matching and close variants. They describe documented platform behaviour and its limits – they are not legal, provider or association studies, nor a general market price or proof of success. Information about how maitiq works is available at maitiq.com.

Have maitiq review your specific case.