Search Engine Land contributor Ana Kostic proposed a three-bucket framework for paid search around AI Overviews: fight for some queries, influence the answer around others, and generate demand elsewhere.
It is a useful planning prompt. It is also practitioner analysis, not a Google policy or a rule that can move a service business’s budget on its own.
The better question is smaller: what job should this query do, and what evidence would justify paying for that job?
Start with the query, then add the business result
Build the first review set from search terms that account for meaningful spend, qualified leads or both. For each one, save four pieces of evidence:
- Intent: Is the person researching, comparing options or ready to inquire?
- Current result page: Does an AI Overview appear, and where are the ads and organic results around it?
- Your visibility: Is the business named, cited, advertised or absent?
- Business outcome: Did the term produce a qualified inquiry, appointment or sale in the CRM?
Check the result page on a consistent date, device and target location. Search layouts change. The purpose is to create a dated observation that can sit beside the account data, not to declare a permanent result for the keyword.
If the CRM outcome is missing, mark it missing. A conversion in Google Ads is not enough to classify a term when the business still cannot tell whether the inquiry was useful.
Glory’s paid advertising, landing-page and CRM work connects those records: the search term, the ad and page, and the inquiry in the pipeline. That gives the budget decision a business result rather than a platform conversion alone.
Give each bucket an evidence test
Kostic calls the first bucket Fight. These are commercial queries where a paid click still feeds real leads or sales. For a service business, keep the term in this bucket only when the query shows buying intent and the account can connect the spend to a qualified result. Position matters after that test, not before it.
Influence covers questions where the AI answer helps shape the decision. Paid search, useful page content and organic visibility can overlap here. The first task is to inspect the answer and its sources. Then decide whether the gap belongs in the ad, the page behind it or the wider content on the site.
Generate demand is the third bucket. The author’s argument is that some informational queries have become poor capture channels because the answer ends on Google. Treat that as a hypothesis to test in your own account. A budget move needs repeated weak capture, a reason the topic still matters and a way to measure the channel receiving the money.
The labels help because they force a decision. The evidence keeps the label honest.
Do not inherit somebody else’s benchmark
The contributor article includes an intent split from one B2B or industrial account. That example says something about that account, not the normal mix for every service business.
It also cites an SE Ranking study of 50,032 US commercial keywords collected on June 30, 2026. That was a snapshot of ads in AI Mode, a different search surface from AI Overviews. Its ad-frequency and cost-per-click patterns should not become AI Overview budget benchmarks.
Your own term, result page and lead record are closer to the decision. Use outside research to form a question. Use account evidence to answer it.
Turn the classification into a controlled change
Once a query has a provisional bucket, write the next action and the measure beside it.
- For a Fight query, test the bid or ad only after setting the qualified-lead cost you are willing to accept.
- For an Influence query, fix the missing answer or proof on the relevant page, then check the same result journey again.
- For a Generate demand query, move a limited test budget with a defined audience, message and follow-up measure.
- For a query with weak tracking, repair the measurement first and leave the budget decision open.
Set a review date. A query can move between buckets when the result page, cost or lead quality changes.
The three-bucket framework is most useful as a worksheet, not a verdict. Give each query one job, one business measure and one review date. Then move the budget only when the search account and the CRM point to the same decision.
Source note: The fight, influence and generate-demand model is Ana Kostic’s contributor framework in Search Engine Land. The linked SE Ranking numbers come from a single US AI Mode snapshot, not an AI Overviews study. Glory wrote the evidence test and operating method.