Negative keyword suggestions should be reviewed as decisions, not accepted as cleanup. Keep the original search query, likely intent, proposed match type, evidence, and approver together because a broad or ambiguous exclusion can remove useful demand.
Octocrew — managed AI marketing agents — configures an AI Paid Media agent to prepare this package within the accepted platform, permissions, and approval mode; current pricing and the AI agent comparison explain the wider model. Google Ads work is configurable after a fit check and should not be described as supported until its platform package and acceptance are complete.
Which search-term evidence should you start with?
Start with the search terms report because it shows searches that triggered an ad, but do not treat it as a complete query record. Google explains: “Some search terms that don’t have enough query activity are omitted from the search terms report in order to keep with our standards on data privacy.”
Set the review window before making decisions. Keep the campaign, ad group, keyword, match type, cost, conversions, conversion value, and destination page beside each search term. If conversion tracking is missing, stale, or disputed, mark that gap instead of treating the absence of conversions as proof that the query is waste.
The first pass should classify the query:
- Clearly irrelevant. The search asks for a product, location, audience, or job the business does not serve.
- Relevant but weakly matched. The query belongs to the offer, but the keyword, ad, or landing page may be too broad.
- Ambiguous. The words can signal more than one intent and need context.
- Useful demand. The term is relevant and should remain eligible even if the current result is disappointing.
This separates targeting errors from offer, creative, landing-page, or measurement problems.
What intent does the query show?
A search query is evidence of language, not a complete explanation of intent. Read the full phrase. Compare it with the offer and destination page. Look for modifiers such as free, jobs, course, template, repair, wholesale, nearby, or a competitor name, but do not assume that every modifier has the same meaning for every account.
For example, “free” may be irrelevant for a paid service, or it may belong to a legitimate research query that later converts. A competitor term may be intentionally targeted, incidentally matched, or prohibited by the campaign plan. The account’s accepted strategy decides which interpretation applies.
Record the reason in plain language: “The query asks for employment; this campaign sells local repair services.” That note is more useful than a label such as “bad traffic.”
Which negative match type is the narrowest safe choice?
The narrowest safe choice is the match type that removes the verified mismatch without excluding additional useful demand. Google Ads supports negative broad, phrase, and exact match, and their behavior is not identical to positive keyword matching.
- A negative broad match can block a search when all negative terms appear, even in a different order.
- A negative phrase match blocks searches that contain the negative terms in the same order.
- A negative exact match blocks the exact terms in the same order without additional words.
Google states: “Negative keywords won’t match to close variants or other expansions.” Synonyms and singular or plural versions may therefore need separate entries, while casing and misspellings are handled automatically and do not need separate negatives.
The decision should therefore contain two fields: the text and the match type. “Add running shoes as a negative” is incomplete. “Add "running shoes" as a negative phrase at the campaign level because the account sells only tennis shoes” is reviewable.
Use the narrowest exclusion that resolves the verified mismatch. A wider negative requires stronger evidence because it affects more future searches.
Where should the negative keyword apply?
Apply the negative at the narrowest level that enforces the verified rule. A negative can apply to an ad group, campaign, or shared list, and the larger the scope, the larger the possible impact.
An ad-group negative can separate two closely related intents. A campaign negative can remove demand that the campaign should never serve. A shared list can enforce a stable business rule across campaigns, but it should not become a dumping ground for one-off reactions.
Before approving the level, ask:
- Could another ad group serve this query correctly?
- Does another campaign intentionally target it?
- Is the exclusion a permanent business rule or a temporary diagnosis?
- Would the same word mean something different in another product category?
If ownership is unclear, keep the suggestion pending.
Google Ads’ published account limits allow at most 20 negative keyword lists per manager account and 20 per child account, 5,000 keywords per list, and 10,000 negative keywords per campaign. These are capacity ceilings, not targets or evidence that a larger exclusion list is better.
Which decision state fits the term?
Use one of four states because a practical review needs more than approve or reject.
| State | Use it when | Required record |
|---|---|---|
| Exclude | The mismatch is verified and the scope is safe | Negative text, match type, level, reason, approver |
| Keep | The query is relevant or intentionally targeted | Reason and any follow-up test |
| Investigate | Intent or measurement is unclear | Missing evidence, owner, next check |
| Reroute | Another campaign or page should serve the demand | Proposed destination and change owner |
This prevents an uncertain term from becoming a permanent exclusion by default.
How should an approval-gated workflow operate?
An approval-gated workflow separates evidence preparation from platform execution. An AI Paid Media agent can collect terms, prepare evidence, and propose narrow exclusions; a named operator approves the exact negative text, match type, and level before any eligible change is applied.
Google Ads remains configurable only after a fit check, platform package, permissions, and acceptance tests are complete. When execution is accepted, platform readback confirms the applied state; this is a role-boundary example, not a client result.
What should an AI Paid Media agent return?
A useful negative-keyword review is a decision package:
- the exact search query and date range;
- campaign, ad group, keyword, and current match type;
- cost and conversion evidence with tracking caveats;
- the proposed negative text, match type, and level;
- one plain-language reason;
- status: exclude, keep, investigate, or reroute;
- reviewer and final platform readback.
Octocrew’s AI Paid Media agent is configured around accepted data sources, permissions, spend boundaries, and approval modes. It can prepare qualified changes and apply only those that fall inside the recorded authority.
What should you verify before removing demand?
Negative keywords are powerful because they change future eligibility. Keep the original query visible, treat missing data as a blocker, and make the match type and scope explicit. The goal is not the largest exclusion list. It is a smaller set of defensible decisions that the account can explain later.
FAQ
Does the search terms report show every query?
No. Google omits some low-volume queries for privacy, so record the report window and treat the visible rows as useful but incomplete evidence.
Which negative match type is safest?
Use the narrowest type that resolves the verified mismatch. A broader exclusion needs stronger evidence because it can remove more future demand.
When should a shared negative list be used?
Use one only for a stable rule that should apply across the named campaigns. Do not turn an account limit into a list-size target or use a shared list for one-off reactions.
Can an AI Paid Media agent apply the change?
Only when the exact change falls inside an accepted platform, permission, approval mode, and recorded authority. Google Ads remains configurable until its platform package and acceptance are complete, and any eligible execution requires platform readback.