An AI Overview is the AI-generated summary that Google displays above the classic results. According to a study published by Semrush on 21 October 2026, in August 2026 it appeared on 88% of a panel of 400 commercial queries tracked in the United States on desktop, up from 45.75% in September 2025; weighted by search volume, those queries represent 97% of the panel's searches. Comparisons, alternatives, reviews, prices, best X for Y: the first opinion a prospect reads about an offer is now, most of the time, a text assembled by Google from the company's website, its competitors' sites, review platforms and videos. AI Overviews are available in Luxembourg, in French, German and English, and in the three neighbouring countries. Ranking well remains necessary; being cited, and correctly described, in the answer becomes the issue. Here are the figures, their limits, and what a Luxembourg company can check right now.
The facts
- The Semrush study (Luke Harsel, 21 October 2026) tracks every month, from September 2025 to August 2026, a fixed panel of 400 commercial keywords split across three sectors (banking, software, e-commerce) and five query types used just before a purchase: alternatives (sofi alternatives), brand vs brand (hubspot vs salesforce), best X for Y (best credit union for car loan), pricing and cost (price antivirus software), reviews (chase checking account review). Google United States data, desktop only. The share of keywords triggering an AI Overview rises from 45.75% to 88%. Weighted by estimated search volume, coverage reaches 97.08%: the 12% of keywords without an AI Overview carry less than 3% of the panel's searches.
- Review queries record the largest increase, from 30.14% to 89.04%, moving from the least exposed type to the second most exposed; best X for Y is the most covered, at 96.12%; brand vs brand grows the least (63.41% to 86.59%) because it was already the most covered type a year earlier. By sector, the gap closes: from 32.62% (e-commerce) and 68% (software) in September 2025 to a range of 85% to 91% in August 2026. Of the fifteen sector and query type combinations, twelve exceed 80%, three reach 100% (best X for Y in banking and in software, alternatives in software) and three stay below (alternatives in banking 65%, pricing in e-commerce 71.43%, reviews in software 75%). Banking reviews go from 13.79% to 89.66%, e-commerce pricing from 4.76% to 71.43%.
- PPC Land (22 October) recalls the limits: cells of 20 to 30 keywords where a single keyword moves a cell by three to five points, volumes modelled by Semrush, a vendor that sells the tracking tools its conclusions recommend, and a labelling error in the key takeaways (82.81% attributed to brand comparisons instead of alternatives). The study measures the presence of an AI Overview; it measures neither clicks, nor traffic, nor conversions. It extends a Semrush study of 2 July 2026 on more than 600,000 keywords, which measured an average 71% increase in AI Overviews on commercial-intent result pages between November 2025 and April 2026; the panels differ, the direction is the same. On 15 October, Google announced more inline links in its AI answers, placed higher, and a test of a dedicated section for official sites at the top of the answer, without publishing any click data (Search Engine Land).
Google's AI answers now cover almost every query that precedes a purchase
Studies on AI Overviews had so far mostly measured informational queries: PPC Land's timeline cites Pew Research Center (July 2025, a click on a link in 8% of cases with an AI summary versus 15% without) and Ahrefs (February 2026, click-through rate of the first position estimated down 58% when an AI Overview is present). The October study moves the question down the funnel, to the moment when a prospect compares offers, looks for a price or reads reviews, and where revenue is decided. Semrush concludes that a first organic position will probably bring less traffic than a year ago; that is the vendor's opinion, which no traffic data in the study supports. What the study establishes is the presence of a generated answer before the results on almost every commercial query in the panel, with a US panel, on desktop and in three sectors. The table below sets out the five query types.
| Query type | September 2025 | August 2026 |
|---|---|---|
| Reviews (amazfit smartwatch review) | 30.14% | 89.04% |
| Best X for Y (best smartwatch for women) | 42.72% | 96.12% |
| Pricing and cost (coffee maker under 100) | 43.59% | 82.05% |
| Alternatives (crowdstrike alternatives) | 48.44% | 82.81% |
| Brand vs brand (keurig vs nespresso) | 63.41% | 86.59% |
| Whole panel | 45.75% | 88% (97.08% volume-weighted) |
The first opinion a prospect reads is assembled by Google from sources the company does not choose
Semrush gives one example, to be read as an example and not as a statistic: on the query mailchimp vs klaviyo, the AI Overview cites Klaviyo's own guide for the argument choose Klaviyo if, and a third-party site, Zapier, for the argument choose Mailchimp if you are a small generalist business; a YouTube video and another third-party site appear among the sources put forward. One brand made its own case, the other was described by a third party. According to Semrush, the answer to a your brand vs a competitor query can draw on YouTube videos, Reddit threads, review sites, your site and the competitor's. The study provides no citation statistics across the 400 keywords.
Semrush's five recommendations follow from this: track visibility in AI Overviews alongside rankings (mentions, citations, sentiment); map the sources cited (review platforms, sector roundups, videos, forums, competitor sites); keep comparison and pricing pages current and structured so that Google extracts clear passages, since outdated information leaves room for an inaccurate description; work with YouTube creators; earn customer reviews on the platforms cited, within their rules. We recognise in this list what we apply on the sites we support: Google's answer is built from what the web says about the company, and the best-maintained page of that web is the one the company controls.
In Luxembourg, the same mechanism applies in three languages, with more desktop searches and cross-border prospects
No public data measures the share of AI Overviews on commercial queries in Luxembourg, nor in French or German: the 88% figure describes a US panel and does not transpose. What is established can be read in Google's help pages consulted on 23 October: Luxembourg is listed among the countries where AI Overviews and AI Mode are available, like France, Belgium and Germany; the list of AI Overview languages includes French, German, English and Portuguese, and Luxembourgish is not on it. In France, the launch dates from 22 July 2026 according to Google France. According to Statcounter, Google accounted for 80.86% of searches in Luxembourg in September 2026 (Bing 8.84%, DuckDuckGo 6.76%), and 54.48% of searches were made on desktop against 44.24% on mobile: the study's desktop-only scope, a limit on many markets, is closer to the Luxembourg reality.
Three lines of reasoning follow, to be taken as such. The study's sectors speak to the local economy: banking and financial services, software and B2B, e-commerce, with queries such as best bank for, reviews of this bank or software A vs software B. Multilingualism multiplies the answers to watch: the same company can be described differently depending on whether the query is asked in French, German or English, because the available sources differ from one language to another, as we showed for the languages of customers in Luxembourg. On a small market, the ecosystem of local comparisons, reviews and media is narrower than in France or Germany: Google may then draw on French, Belgian, German or generic sources, and the company's own content and customer reviews gain relative weight. Finally, part of the prospects search from France, Belgium or Germany, three countries where AI Overviews are active: a cross-border worker sees the same generated answer, whatever the company's country.
Six checks to run on your purchase queries
The checks below are our own practice and take up Semrush's recommendations; they are within reach of an SME as much as a large account.
- List your purchase queries. Comparisons (X vs Y), alternatives, reviews, prices, best for, in each of your customers' languages: French, German, English, and Portuguese if your customer base justifies it.
- Look at the answer, as well as the position. For a sample of 20 to 50 strategic queries, asked from Luxembourg in each language: does an AI Overview appear? Is your brand cited? How is it described? Which sources does Google use? Record the result, it will serve as a starting point.
- Update your commercial pages. Pricing pages, comparisons, product FAQs: clear, dated and easily extractable information limits the risk of an inaccurate or outdated description.
- Work on your reputation beyond your site. Customer reviews on the platforms actually cited in the answers, presence in your sector's comparisons and media, videos on YouTube.
- Adjust the measurement. Track visibility in AI answers alongside rankings and traffic, and accept that a good position no longer guarantees the same volume of clicks as before.
- Think SEA and SEO together. Commercial queries are also where paid ads concentrate; the study does not measure the effect of AI Overviews on ads, and we claim nothing on that point, but both teams must look at the same result pages.
Google repeats that optimising for its AI features is a matter of search engine optimisation, with no special markup. The fundamentals remain the base; what changes is where the work is read: in the generated answer, and no longer only in the list of links.
AIxH's view
We audit our clients' AI visibility this way: query by query, language by language, noting whether Google answers, whether it cites the company, what it says and where it gets it from, then correcting first what depends on the company (its commercial pages, its reviews, its presence in the sources cited). That is the work of our GEO service, carried out with search engine optimisation since both rest on the same pages. We do not promise a citation share: we measure a starting point and what moves afterwards. If your purchase queries have not yet been looked at from this angle, in French, German and English, now is the time.
