When people search your brand name on Google, they now usually see an AI-written summary before they see your website. AI Overviews appeared on 82.91% of US desktop branded-keyword results on 29 September 2026, up from a 61.3% monthly average in July (Ahrefs, 232.8 million search crawls, October 2026). For many brands, the first description a searcher reads is no longer one you wrote.
That changes what "owning your brand search" means. This piece covers how fast the shift happened, what those answers typically draw on, where they go wrong, and how to check and correct them.
How common are AI Overviews on branded searches now?
Very common, and recently so. Ahrefs compared branded and non-branded keywords across US desktop results from July to September 2026.
Branded queries caught up. Monthly averages for branded keywords rose from 61.3% in July to 65.1% in August and 73.3% in September. Non-branded keywords moved from 73.5% to 78.3% over the same period. The gap between them narrowed from 12.2 points to 5.
A year ago it was rare. In Ahrefs' November 2025 study, AI Overviews appeared on 13.1% of branded queries (a different sample, so approximate). Brand searches were long treated as the safe zone where Google sent people straight to the official site. That assumption no longer holds.
Bare brand names too. In a separate panel of 100 global brand names searched on their own, AI Overviews appeared for 5 of 98 usable results on 1 July and 63 of 100 by 30 September. In 48 of those 63, the AI Overview sat in position one.
Two caveats from the study itself: it covers US desktop results only, and Google can change when AI Overviews trigger at any time. The 82.91% figure is a single-day reading, not a monthly average. Google also has a second surface to watch: AI Mode, which answers differently and cites different pages (see our explainer on AI Mode vs AI Overviews).
Where does AI get its information about your brand?
Mostly from sources you do not control.
In the Ahrefs brand-name panel, the most-cited external domains in branded AI Overviews were Wikipedia (cited in 71.9% of the 57 AI Overviews that listed sources), YouTube (38.6%), LinkedIn (22.8%), Yahoo Finance (15.8%) and Facebook (10.5%). With 57 AI Overviews, treat these shares as directional. That is a sample of large global brands, so smaller companies will see a different mix, but the pattern is clear: third-party profiles carry a lot of weight. Our piece on whether Wikipedia matters for AI visibility covers what that means if you do not have a page.
Every domain on that list is a third-party platform, not the brand's own website. In this panel, Google built brand summaries largely from profiles and pages the brand does not fully control. Review profiles are another third-party source worth keeping current, which we cover in reviews and AI visibility.
How accurate are AI answers about brands?
When AI engines answer, they are usually right on basic facts. The bigger problems are refusals, stale details and weak comparisons.
Seer Interactive, a digital agency, tested 1,562 prompts about its own brand across six AI platforms, collecting 28,123 responses daily from 1 May to 9 June 2026 (published August 2026). Its findings:
Refusing was more common than lying. Platforms declined to answer 31.2% of prompts on average and gave an inaccurate answer 2.8% of the time. When they did answer, they were accurate 95.9% of the time.
Engines differed a lot. AI Mode had the highest total accuracy at 74.2% of all prompts. Claude declined most often (46.3% no-answer), so its total accuracy was lowest at 52.6%, even though it was 98% accurate when it answered. AI Overviews had the highest inaccuracy rate at 5.7%.
Comparisons were the weak spot. Comparative prompts, the "who are the best agencies for X" kind, were accurate only 18.8% of the time and went unanswered 80.3% of the time.
The errors were stale, not invented. Mistakes included an old office location, services the company does not offer, an inaccurate headcount, and a narrative traced back to a single review from 2018.
This is one company studying itself, with its own large web footprint, so treat it as directional. Smaller brands with thinner coverage tend to have more gaps for the engine to fill, which is where invented details creep in.
The stakes are real. In Moffatt v Air Canada (2024), a Canadian tribunal held the airline liable for a refund policy its own chatbot invented. AI errors about your brand are your problem even when you did not write them. Our guide on when AI gets your brand wrong walks through the causes and fixes.
Why does this matter more than generic queries?
Branded searches come from people who already know you. They are comparing, checking, or about to buy. An AI answer at that moment can:
Reframe your positioning. If the AI Overview calls you a "general provider" when you are a specialist, that is the first impression a warm prospect gets.