Google AI Overviews now appear on 88% of tracked commercial keywords. What should you measure?

What the October study actually found

Semrush published its study on October 21, 2026 after tracking 400 commercial keywords in the United States from September 2025 through August 2026. The panel covered five modifiers associated with purchase research: alternatives, best X for Y, price or cost, reviews, and brand-versus-brand comparisons. Across that fixed panel, AI Overview coverage rose from 45.75% in September 2025 to 88% in August 2026. When Semrush weighted the keywords by estimated search volume, coverage reached 97.08%.

Some groups changed sharply. The study says review queries rose from 30.14% to 89.04%, while ecommerce coverage rose from 32.62% to 85.11%. These are useful signals for teams that sell products or software: AI-generated summaries are no longer limited to broad informational questions in this sample.

The boundaries matter. This was a defined set of 400 US desktop keywords across selected industries and modifiers, not a census of all searches, countries, devices, or query types. Search-volume weighting also answers a different question from equal weighting: it gives more influence to queries estimated to be searched more often. Neither number tells a particular business whether its name appeared in an answer.

An AI Overview appearing is not the same as your brand appearing

A SERP-level coverage metric asks whether Google displayed an AI Overview for a keyword. It describes the search environment. Brand visibility asks whether the answer named your business, linked to a page you control, described your product accurately, or recommended you. Those are separate observations and should not share one denominator.

Imagine a hypothetical project-management SaaS tracking the query “best project management software for a five-person agency.” An AI Overview appears, but its answer names three competitors and cites an independent review site. The query counts as AI Overview coverage for the SERP, yet the SaaS has no brand mention in that answer. If its dashboard labels the whole event as “AI visibility,” it overstates what the buyer actually saw.

The reverse distinction matters too. A page can be cited as a source without the answer naming its brand. A brand can be named without a link to its own site. And a mention is not necessarily a recommendation. Report these events separately instead of turning them into one score that hides what changed.

• Feature presence: did an AI Overview appear for the query?
• Brand mention: did the answer name the company or product?
• Owned-source citation: did the answer link to a page on the company’s domain, and which page?
• Recommendation and accuracy: did the answer suggest the product, and was its description correct?
• Business outcome: did measurable referral traffic, branded demand, leads or sales move over time?

Why prompt selection can distort an AI visibility report

The same discipline is needed when measuring conversational answer engines. In research published on October 19, 2026, Momentic reported that its sample of 21 brands produced very different mention rates depending on prompt type: branded prompts mentioned the brand 85.8% of the time, comparison prompts 75.5%, and category prompts 11.3% across the measured September 1–October 18 window. Blending them produced 39.0%. The blended figure depends heavily on the mix of prompts in the report; it is not a universal industry benchmark.

That is why a visibility program should separate prompts that already contain your company name from discovery questions that do not. Branded questions help audit reputation, factual accuracy and caveats. Category questions test whether an unfamiliar buyer can discover you at all. Comparison questions reveal how the product is positioned against named alternatives. Combining these groups can make a report rise or fall simply because its prompt mix changed.

Sample size and repeatability matter as well. Momentic’s analysis estimated that about 100 prompts can put the margin of error near ±5 percentage points for a rate around 10% on one platform, under its stated assumptions. Its measurements also found that repeated answers to the same category prompt sometimes disagreed about whether a brand appeared. Treat that estimate as a planning guide from one study, not a guarantee that every 100-prompt audit is precise to five points.

A practical measurement plan for lean SaaS and ecommerce teams

Start with a small, stable set of real buyer questions. Draw them from sales calls, support tickets, site search, product comparisons and the language customers use before choosing a supplier. Include category, comparison and branded questions, but label each group before collecting results.

For Google, record AI Overview presence at the query level and separately inspect the answer, its cited URLs and whether your brand is named. Keep country, language, device and collection date consistent. For ChatGPT, Perplexity or Gemini, record the exact prompt and relevant search or account context when available. Do not assume that the same query on two products is a directly comparable experiment.

Repeat the same prompt set on a regular cadence before expanding it every week. Keep the raw answer, source URLs and capture date so a reviewer can check a claimed change. If the prompt wording, parser, platform interface or tracking method changes, annotate the break in the series. A new extraction feature can make a dashboard record more details without increasing the underlying visibility.

Report sample sizes beside rates. A move from 10% to 12% on a small sample may be noise; do not celebrate it as growth without more evidence. Prefer trends that persist across several readings and inspect individual answers when the business impact is material. For broad category visibility, increasing the number of distinct, well-chosen prompts often teaches more than repeatedly asking only a handful of questions.

Finally, place answer-level measurements beside business signals rather than claiming one citation caused one sale. Review AI referral sessions where analytics preserves a referrer, branded search demand, qualified leads and conversions over a longer window. Some influence will be invisible to referral analytics, so describe the relationship as a trend or hypothesis unless you have a credible attribution design.

What Google’s guidance means for content teams

Google’s Search Central guidance says pages must be indexed and eligible to appear with a Search snippet to be eligible as supporting links in AI Overviews or AI Mode. It does not specify a separate technical trick that guarantees inclusion. Google also describes its AI features as ways to surface relevant links and notes that AI Mode and AI Overviews may use query fan-out to explore related searches.

For a site owner, that points back to fundamentals: publish useful pages that answer a real task, make the content accessible to Search, explain claims with evidence, keep product and pricing facts consistent, and connect related pages with sensible internal links. For commercial searches, a clear comparison or alternatives page may help a buyer understand fit, but no page format guarantees that Google will select it.

Do not manufacture dozens of near-identical pages just because a modifier appears in a study. First ask whether the query represents a real customer decision and whether your page adds information a buyer cannot get from a generic summary. A smaller set of specific, accurate pages is easier to maintain and evaluate than a large collection of thin variations.

Turn the measurement into an editorial decision

A useful report should end with a decision, not a vanity score. If AI Overviews appear often for your target queries but your brand is absent, review which sources the answer uses and what information those pages provide. If your site is cited but the product is described incorrectly, fix the relevant source facts and documentation. If your brand is named but never recommended, investigate the comparison criteria, missing evidence or product limitations rather than publishing another generic explainer.

Ranki.ai helps SaaS and ecommerce teams plan and produce content from a repeatable editorial workflow. Use the findings from a measured query set to choose the next useful article, comparison or answer page, then publish it through your connected workflow. Keep the measurement method documented and do not assume that content production alone guarantees a mention, citation or ranking.

Start with Ranki.ai at https://ranki.ai/auth. The goal is not to claim that every AI answer is measurable with one dashboard; it is to make the content decisions you can control more consistent, and to judge progress using evidence that matches the question you are asking.