Google AI Overviews Are Moving Deeper Into Buying Decisions
Google’s AI Overviews appeared for 88% of the commercial keywords tracked in a Semrush study in August 2026, up from 45.75% in September 2025. Search Engine Land’s coverage reports on that expansion into searches involving purchase decisions.
For a business publishing comparisons, pricing information, or product reviews, this raises a practical question: what does a prospective customer learn about you from the answer on the search results page?
That question belongs in a commercial content audit. A page’s ranking still deserves attention. So does whether the summary describes the offer accurately and preserves the conditions someone needs to make a sensible choice.
What the study measured
Semrush followed a fixed panel of 400 U.S. desktop keywords across banking, ecommerce, and software from September 2025 through August 2026. The groups covered alternatives, brand comparisons, “best X for Y,” pricing or cost, and reviews.
| Measure | Reported finding |
|---|---|
| Overall keyword coverage | Increased from 45.75% to 88% |
| Review searches | Rose from roughly 30% to roughly 90% |
| “Best X for Y” searches | Reached 96.12% coverage |
| Coverage weighted by estimated search volume | Reached 97.08% |
The spread narrowed as it rose. In September 2025 the modifier groups ranged from 30.14% for reviews to 63.41% for brand comparisons; by August 2026 every group finished above 80%, ranging from 82.05% to 96.12%. All three industries converged between 85% and 91% coverage, with ecommerce posting the largest industry gain, from 32.62% to 85.11%. Three of the fifteen industry-and-modifier combinations remained below 80%: alternatives for banking at 65%, price queries for ecommerce at 71.43%, and reviews for software at 75%. Even those climbed steeply; banking reviews rose from 13.79% to 89.66%. Semrush’s methodology and findings.
Keyword coverage counts each term equally. Volume weighting gives more influence to terms with greater estimated search demand. Both figures describe this sample; they aren’t estimates for every commercial Google search.
The research measured AI Overview presence. It didn’t establish changes in clicks, leads, or sales.
Why this deserves a place in a commercial content audit
Someone comparing products needs more than a product name. They may need to understand costs, compatibility, setup work, and which option fits their situation.
Consider a fictional scheduling tool that offers a low monthly price but charges separately for additional locations. A prospective buyer researching software for several branches needs that condition. If a summary presents only the starting price, it leaves out information that could change the decision.
That example isn’t a finding from the study. It’s the kind of problem to look for when reviewing the answers shown for a business’s important searches.
The fix is specific: check the summary, inspect its sources, and make sure the public pricing information explains the charge clearly. Then repeat the check later and record what changed.
Start with the buying questions already tied to your business
Begin with a short list of searches that sales conversations repeatedly bring up. For a software company, those might involve a competitor comparison, a use case, a pricing question, and an integration requirement. For a service business, project scope and provider selection might matter more.
For each search, record the date, country, device, and whether an AI Overview appeared. Save the answer and the cited URLs. Record the ordinary organic results too, so you can examine both parts of the experience.
Then ask three questions:
- Does the answer include the information a suitable customer needs?
- If it mentions our business, are the important facts correct?
- Which source supports each claim that needs attention?
Repeat the observations under similar conditions. A saved result gives you something concrete to investigate; several observations help you see whether an issue recurs.
Make comparison pages useful for choosing
A comparison page should explain the circumstances that make each option suitable.
For the fictional scheduling tool, a comparison should address the number of locations, staff permissions, reminder charges, calendar integrations, and the work involved in switching. Those criteria let a reader evaluate the recommendation.
Explain how you established the facts. A hands-on test should identify what was tested and when. A feature comparison based on documentation should link to that documentation and make the basis clear. Keep your own commercial interest visible.
Put the recommendation near its reasoning. For example, explain why a particular permission model would suit a business where branch managers control their own schedules. That gives the reader a useful decision rule and an opportunity to judge whether it applies.
Review existing comparison pages for unsupported claims before commissioning more of them. A well-supported explanation of one recurring customer decision is a manageable first improvement.
Keep pricing conditions and limitations beside the claim
Inspect the statements most likely to influence a purchase: starting prices, included capacity, supported integrations, implementation times, and eligibility requirements.
Keep the conditions beside the fact they qualify. An annual billing commitment belongs beside the advertised monthly equivalent. A location limit belongs beside the plan’s capacity. A feature that requires an add-on should identify that requirement where the feature is described.
This makes the page easier for customers to assess and gives you a clear factual reference when checking a summary.
Also confirm that Google can access the relevant page. Its AI features guidance says a supporting page must be indexed and eligible to be shown with a snippet, and states there are no additional technical requirements beyond that. It recommends making important information available as text. Eligibility still leaves source selection to Google.
Follow the citations when a description is wrong
An inaccurate statement needs a source investigation.
If the cited page belongs to your business, check whether the information is outdated, ambiguous, or missing a qualification. Correct the underlying explanation and document the revision.
If the answer cites a third-party comparison or profile, inspect that page directly. Establish what is wrong before requesting a correction, and provide a public reference supporting the current facts.
Sometimes the cited page is accurate while the generated answer loses an important condition. Save both versions. That distinction helps you describe the problem precisely and avoids unnecessary changes to accurate content.
Measure visibility and business outcomes separately
Google’s generative AI performance reports in Search Console provide dedicated visibility information, including impressions and the pages appearing in generative AI features. Those reports can help identify URLs worth investigating.
Pair that evidence with ordinary search performance, landing-page activity, and qualified inquiries. Keep the questions distinct: did the page appear, did someone visit, and did the visit contribute to useful work?
If impressions rise while inquiries fall, investigate the actual queries, visitors, and inquiry process before assigning a cause. Changes in rankings, demand, campaigns, and the website itself may also matter.
Likewise, a citation establishes that the page was used as a source in the observed answer. Assessing its commercial value requires more evidence.
Where this leaves commercial content
The sample is one fixed panel of 400 keywords, but the direction is hard to ignore: buying searches now mostly return an AI-generated answer, and that answer is the first thing a prospective customer reads. It is assembled from pages someone controls, including yours: the comparison criteria, the pricing conditions, the third-party profiles. The summary has become part of the sales surface, and it deserves the same accuracy standard as anything you publish.
The work is concrete rather than speculative. Pick one buying question that sales conversations keep surfacing, save the current answer, verify its claims against your own pages, and fix the single most misleading fact, whether that is a billing condition, a comparison criterion, or an outdated profile. Then recheck the same search under the same conditions and record what changed. The study measures presence, not outcomes, so visibility and revenue stay separate questions. But the gap between a correct page and a correct answer is now something you can see, investigate, and close.