Branded AI Overviews
73%
73% of branded keywords showed an AI Overview in September, up from 61% in July.
Source: Ahrefs (US desktop)
Brands whose content AI cited when buyers were researching a category were almost twice as likely to be named when those buyers asked for recommendations. That relationship held up over three months, even as AI search named fewer brands and relied on fewer sources.
Named on consideration questions
Brands AI cited on research questions were named far more often.
Get the next quarterly search report the minute it drops.
Last quarter, we found that AI accurately described 96% of the brands we tested yet mentioned only a small fraction of them in its answers, and that the strongest signals associated with being mentioned came from a brand's presence across the rest of the web.
This quarter, we asked the same category questions about the same cohort of brands three months later, and we added a new question: does the content AI relies on when a buyer is still learning about a category predict which brands it recommends when that buyer is ready to choose?
The answer was yes, and it partly holds after accounting for brand size. AI also mentioned fewer brands than it did in Q2, concentrated its citations on a smaller set of sources in every industry, and raised the amount of third-party web presence a brand needs to be named at all.
We don’t enter into our quarterly research expecting a particular result. Instead, we go in with the questions we still had after last quarter’s findings, and we let the data lead the way.
The most startling finding was how closely informational content tracked with brand recommendations. If you assumed AI absorbs content into its answers with little benefit to the brand, our research found otherwise. Brands whose informational content AI cited when buyers were researching were almost twice as likely as other brands to be named when those buyers asked who to choose. Informational content earns a brand a place in the recommendation, and it belongs in any full-funnel strategy for AI search.
What strikes me about these findings is how familiar the pattern is. People have always trusted the company that helped them understand a problem before it tried to sell them a solution. AI search platforms appear to be following suit.
We asked AI platforms the same category questions about the same cohort of brands we measured in Q2, then compared each brand's AI visibility with its presence across the rest of the web.
We collected about 16,500 answers to category prompts across nine AI platforms at the end of September, using the same prompts as Q2. Citation results cover the six platforms that searched the web in our collection: ChatGPT, Perplexity, Gemini, Microsoft Copilot, Google AI Overviews, and Google AI Mode.
For every brand, we counted third-party web mentions, referring domains (the number of websites linking to the brand), organic rankings, and Google's organic results for the same questions we asked AI. To test whether the authority of a brand's mentions matters, we scored a random sample of 500 mention pages per brand.
We used Ahrefs Brand Radar, which records how AI platforms answer the questions US searchers ask, and classified those questions as research (informational) or consideration (commercial).
The cohort covers ecommerce and retail, financial services and insurance, healthcare, legal services, and SaaS and technology.
Brands whose content is cited in AI answers to research questions were almost twice as likely to be named in answers to consideration questions as brands that weren't present in research answers.
Questions US searchers asked AI platforms about each brand's category, recorded by Ahrefs Brand Radar.
Whether each brand's own content was cited in answers to research questions, and whether the brand was named in answers to consideration questions, across 161 brands and six AI platforms.
It explains part of it. After controlling for how often AI mentions each brand overall, brands cited for research questions were still more likely to be named on consideration questions.
We read this as correlation rather than cause. Being cited at the research stage and being named at the consideration stage appear to draw on the same underlying trust in a brand. That's also how a careful buyer behaves: the company that explained the problem has a head start when it's time to choose.
We compared each brand’s answers on July 1 with its answers in October. On July 1, there were 40 brands in our cohort that weren’t named on any consideration question. We split them into two groups:
By October, 81% of the first group had been named on a consideration question.
In the second group, 21% had.
21 brands in Group 1, 19 in Group 2.
Research-stage content has been treated as a traffic play. In AI search it looks more like a qualifying round. The brands AI trusts for explanations are the ones it appears to trust with recommendations.
Research answers still rarely name brands, and the sources behind them are shifting. Brand share of citations fell on 18 of the 20 research questions we track (four per industry), while government and association sites gained. In Q2 we asked whether research-stage citations would stay open to brands late to AEO. They have, but brands now compete with institutions for them.
Based on 17,217 citations in Q2 and 22,524 in Q3.
Research-stage content does two jobs. It answers the buyer's question, and it establishes the brand as a source AI can rely on before the buyer starts comparing providers. Because government and association pages are taking a larger share of research citations, brand content has to be more specific, more current, and better sourced than the institutional page on the same topic. The return on that work shows up later in the journey, when the same buyer asks who to choose.
That also changes what to measure. For each consideration prompt you track, identify the two or three research questions buyers ask before it and add them to your prompt tracker. Report whether your content is cited in those answers, not only whether your brand is named in the consideration ones.
The sources that gained share at the research stage are also placement opportunities. Industry associations and trade publications now account for a larger share of what AI cites when buyers are learning about a category. Membership listings, expert contributions, and guides published through those bodies put the brand inside a source AI already trusts, which supports being named once the buyer is comparing options. That complements, rather than replaces, the brand's own research content being cited.
In every industry we track, the 10 most-cited websites took a larger share of AI citations than they did in Q2. In legal services, they now account for about half.
| Industry | Top-10 share of citations, Q2 | Top-10 share of citations, Q3 |
|---|---|---|
| Legal services | 40.6% | 50.7% |
| Healthcare | 33.1% | 41.3% |
| Financial services | 30.3% | 32.5% |
| Ecommerce | 21.4% | 29.7% |
| SaaS | 20.6% | 24.5% |
The same category prompts as Q2, on the five platforms that searched the web, excluding AI Mode (see methodology).
The share of all citations that went to each industry's 10 most-cited websites, in Q2 and in Q3, measured on category prompts in both quarters.
Concentration rose most in legal services (up 10.1 points), followed by ecommerce and healthcare (about 8 points each). In legal services, three directories alone now account for about a third of all citations on category questions, up from about a quarter. SaaS and ecommerce remain the most open industries, with citations spread across thousands of domains, but both also moved toward consolidation.
The three industries in our cohort that Google classifies as Your Money or Your Life (YMYL) topics, legal services, healthcare, and financial services, are also the three most concentrated in Q3. That pattern is consistent with AI platforms leaning on established, vetted sources where the stakes for the reader are highest, though our data doesn’t test the cause.
The timing is covered in the market signals section.
AI platforms are narrowing the list of sources they treat as credible. As direct citations get harder to earn, brands should shift their effort toward earning placements on the sites AI already trusts.
Perplexity leans on trade publications and ranked lists. Google AI Overviews were the most stable platform in our data, with YouTube and Reddit as its top two sources in both quarters. Gemini changed the most: only 7 of its top 20 sources from Q2 remained in its top 20 in Q3. Copilot moved away from LinkedIn, which fell from 1.1% of its citations in our cohort to near zero.
Citations also differ more between platforms than between repeated answers on the same platform. Asked the same question, two platforms share only 4% to 10% of the websites they cite, while a single platform asked the same question twice reuses 25% to 74% of its sources.
The set of sources AI treats as credible in a given industry is getting smaller, and the brands that appear in AI answers are the ones present in that set. Identify the 10 websites that take the largest share of AI citations in your category and treat them as the priority list for coverage, listings, and expert contributions. That list is different for every industry and, increasingly, for every platform.
Because the platforms share so few sources, the practical move is to build the list for the platforms your specific buyers use, rather than use one list for AI search.
Of the 42 brands in our cohort that AI named in Q2, 15 were no longer named in Q3. Two brands were newly named. Most of the brands that dropped out were in the middle of the cohort by web presence.
The same category prompts about the same brands, three months apart.
Whether each brand was named in any AI answer in its industry, in Q2 and in Q3, and how that related to the brand's count of third-party web mentions.
The net loss of 13 named brands is unlikely to be chance (p about 0.002). Most of the losses were among brands with 2,000 to 10,000 third-party web mentions. In that range, the share of brands named by AI fell from 27% in Q2 to 12% in Q3. The brands named most often were the same in both quarters.
134 brands, same keywords and filters as July. No single range's change is statistically significant on its own.
The threshold for being named rose. In Q2, brands had a 50% chance of being named at about 20,000 third-party web mentions. In Q3, that point sits at roughly 24,000 to 28,000.
Third-party web mentions for a 50/50 chance of being named
Being named once is a low bar, though. Most named brands appeared on only a few platforms, so we also measured what it takes to be named on at least three of the six. That point sits at roughly 44,000 web mentions, about twice the threshold for being named at all.
brands named this quarter appeared on only one AI platform. 7 appeared on all six.
A single mention in one AI answer doesn't make a brand part of the conversation. The brands that show up across platforms have built a web presence large enough that every system finds them.
The number of websites that mention a brand and the number that link to it predicted which brands AI named. After those two counts were known, the quality signals we tested added almost nothing to the prediction.
| Signal we tested | Adds anything beyond mention and link counts? |
|---|---|
| Number of websites linking to the brand | YesCorrelates with being named at 0.49, in both Q2 and Q3. |
| Links from high-authority sites (DR 50+, DR 70+) | NoQ2's small additional lift (0.14) fell to 0.03 in Q3. |
| Followed vs nofollow links | No |
| Mentions on high-authority sites (random sample of 500 pages per brand) | SmallAt most a small relationship (0.16), no better than mention count alone. |
| Mentions on high-authority pages that receive traffic | No |
| Branded search volume | NoRelated to being named (0.33), but adds nothing once web mentions are counted. |
| Google Business Profile review count and rating | No |
Mentions and links each contribute on their own. About 6% of the pages that mention a brand also link to it. Among brands with similar link counts, those with more mentions were named somewhat more often (partial correlation 0.18); among brands with similar mention counts, those with more links were named somewhat more often (0.25).
Using the same measure as last quarter, 90.5% of brands were never named in answers to their own category’s questions, compared with 89.2% in Q2. On a broader measure that counts any answer in the brand’s industry, the share never named rose from 70.3% to 77.9%.
The middle of the market is where AI visibility is most fragile. A brand with a few thousand third-party mentions could appear in Q2 and drop out by Q3 without anything changing on its own site. One appearance in one AI answer is not evidence of visibility; a consistent presence across platforms is. That consistency comes from breadth: more publications, more lists, and more conversations that name the brand.
AI Overviews appeared on 97.5% of the category questions we searched on Google, up from 72.8% in Q2, and 37% of the sources they cited were pages ranking in Google's organic top 10 for the same question.
of category searches triggered an AI Overview, up from 72.8% in Q2
of AI Overviews citations ranked in Google's organic top 10 for the same question
the equivalent figure for ChatGPT, Perplexity, and Copilot
Each of our 34 category questions (seven each in financial services, healthcare, legal services, and SaaS, and six in ecommerce), searched on Google and asked on each AI platform.
Whether an AI Overview appeared, and what share of each platform's cited sources also ranked in Google's organic top 10 for the same question.
AI Overviews had the higher overlap with organic results on 29 to 33 of the 34 questions. Google’s top 10 shifted noticeably between searches run minutes apart, so the exact overlap figures carry some noise. In every industry, though, AI Overviews cited ranking pages more often than ChatGPT, Perplexity and Copilot did. Ranking remains a meaningful input for Google’s AI surfaces.
For Google's AI, organic rankings are the single largest input. Keep ranking work in the plan. For the other platforms, the lever is the third-party presence described in Findings 02 and 03, which also accounts for most of what AI Overviews cite.
Platform updates, algorithm changes, and industry developments that influenced the AI and organic search landscape during Q3 2026.
OpenAI released the GPT-5.6 model family across ChatGPT, Codex, and the API, along with ChatGPT Work, an agent for longer tasks.
Why it mattersA model change can shift which sources ChatGPT searches and cites. Your AI visibility can move without any change on your site.
OpenAI moved ChatGPT ads out of pilot in the UK, Japan, Brazil, Mexico, and South Korea, then extended them to 31 European countries on August 24.
Why it mattersBrands can now buy ads in ChatGPT conversations in more than 40 countries, which puts AI search in the ad budget as well as the organic plan.
Google said the Gemini app passed 1 billion monthly active users, and that 63% of users now talk to it by voice.
Why it mattersGemini now has an audience large enough to track alongside ChatGPT when you measure AI visibility.
Reddit's share of ChatGPT search citations fell from 3.83% to 0.52% in four days, according to Promptwatch data, with no similar drop in Google's AI products.
Why it mattersMeasure community presence platform by platform. One platform's sourcing change can shrink what a channel contributes to your AI visibility within days.
Google's third spam update of the year rolled out globally in two days and 16 hours. In an SE Ranking study of 100,000 US keywords, 16.71% of pages ranking in the top 10 fell out of the top 100 during the rollout, compared with 9.2% over a similar stretch in July.
Why it mattersIn SE Ranking's data, this update moved far more rankings than a normal week. A late-August drop can trace back to it even on sites that don't consider themselves spam targets.
Google confirmed that AI Overviews now open to full length on some searches without a click, which pushes organic results further down the page.
Why it mattersIf your rankings held and your clicks fell after late August, expanded AI Overviews could be the cause.
Rank-tracking company Nozzle found that from September 13 onward, 8% of the Google results pages it collected had most of their usual sites swapped out, compared with 0.17% before. The share peaked at 27% on September 29, and Google hasn't commented.
Why it mattersBefore you act on a late-September ranking drop from a third-party tool, check it in Search Console. Rank trackers may have logged drops that real searchers never saw.
Google's fourth spam update of the year took nearly two weeks to roll out. Google said the update reflects improvements to its automated spam detection.
Why it mattersGoogle released no core update in Q3. If your rankings moved in late August or late September, start with these two spam updates.
The market signals behind the headlines.
73% of branded keywords showed an AI Overview in September, up from 61% in July.
AI-driven traffic to US retail sites grew 127% year over year in August.
ChatGPT sent 95.1% of the referral traffic that came from AI platforms in August.
ChatGPT's ad business reached a $1 billion annualized run rate by the end of August, according to OpenAI.
Recurring themes from about 70 client and prospect calls.
Track it in two layers. The first is what AI is doing with your brand right now, which you get from a prompt tracker: how often AI names you, how often it cites your content, how it describes you, and your share of voice against competitors, broken out by platform and by funnel stage. These move first, so they’re your leading indicators.
The second layer is what that visibility is producing, which you can get from tools you already have: AI assistant sessions and conversions in GA4 (check whether GA4's native AI Assistant channel is catching Perplexity and Claude sessions, and add them with a custom channel definition if it isn't), branded search demand, direct traffic to the pages AI cites, and Search Console’s new AI impressions report, available to every site since August 31. These lag, and they’re the ones that connect to revenue.
Read both as monthly trends rather than single numbers, because any one month moves a lot. You won’t get full click-to-purchase attribution, since AI conversations are private, but these two layers together show whether the work is moving in the right direction. This quarter’s research adds one more leading indicator: your citation rate on research questions, which went with being named at the consideration stage (Finding 01). We walk through the full framework in Episode 13 of The Search Signal.
Because ranking is a Google signal, and most AI platforms don’t build their answers from Google’s results. ChatGPT, Perplexity, and Copilot run their own retrieval and draw on a different pool of sources, most of them third-party sites: publications, directories, review sites, and communities. A competitor that gets recommended over you is usually present in more of those places than you are, even if you outrank them.
The practical check is to ask the platform the question your buyers ask, then look at which sources it cites in the answer. If your competitor is in those sources and you aren’t, those are the sources to pursue. Finding 02 shows how concentrated those source lists have become, and Finding 03 shows how much third-party presence it takes to be named at all. The exception is Google’s AI Overviews, which do rely on rankings (Finding 04), so if you’re strong there and weak elsewhere, that’s the pattern you’d expect.
Listen: How To Create a Mention Building Program and How To Build a Brand Entity for AI Visibility
Yes, SEO still matters, and yes, AI Overviews are the most likely reason your traffic fell.
AI Overviews now appear on almost every category search (Finding 04). Since late August, Google has also been expanding them automatically on some queries, so more of the results page sits above the organic listings than it did three months ago. Steady rankings with fewer clicks is what that produces. Search Console’s new AI reports will show you how often your own pages appear in AI Overviews and AI Mode.
The rankings you’re holding are also what keep you in those AI answers. AI Overviews draw on ranking pages far more than ChatGPT, Perplexity or Copilot do, and the organic result is still there under the overview for the buyers who scroll. What has changed is that ranking alone no longer guarantees the click. Findings 02 and 03 cover the signals that get a brand named in the answer itself, and ranking is the foundation they sit on.
It depends on which platforms your buyers use. ChatGPT sharply cut its Reddit citations in August, and in our late-September collection, none of 1,014 ChatGPT answers cited Reddit, down from 223 of 1,017 in Q2. One collection can’t tell us whether that holds. Google's AI Overviews kept Reddit and YouTube as their top two sources in both quarters, and the other platforms barely moved. Community presence still counts toward the off-site footprint that predicts being named across platforms (Finding 03); what changed is that it is no longer a direct citation path into ChatGPT. Measure by platform before reallocating, and expect this to change again.
Every quarterly report raises new questions. These are the areas we'll keep collecting data on.
We answered the correlation question this quarter. Now, we’re running 20 fixed research questions weekly through December to see whether a brand's research citations appear before its consideration-stage mentions, and whether the relationship survives the next round of platform changes.
If research citations consistently come before consideration mentions, that predictability gives marketers more reason to invest in research content first and expect the recommendation to follow.
We’ve established that the sources AI cites for a question aren’t stable. Two answers to the same question, collected minutes apart, cite the same sites only 27% to 72% of the time, depending on the platform. Three months apart, the overlap falls to 6% to 16%. Our weekly panel will track how fast that drift happens in between.
A single check of an AI answer is a weak measurement. Knowing how quickly citations turn over tells marketers how often they need to measure for the reading to mean anything.
Most AI platforms now run several searches behind a single answer, and there are reports that ChatGPT's searches name specific sites. If a platform's own searches are aimed at a short list of directories and publications, those sites will be cited more often no matter how good any individual page is, and the concentration we measured would follow from the search, not the source. We captured the searches behind ChatGPT, Perplexity and Copilot answers this quarter. On our category questions, only about 3% of ChatGPT's searches targeted specific sites, far fewer than an outside analysis found in August. In Q4 we'll track whether that grows.
If concentration comes from how platforms search rather than which sources they judge credible, the list of sites to be present on becomes more predictable.
This quarter the number of websites linking to a brand predicted AI mentions, but the authority of those sites added little on top of it. We saved every brand’s Q3 link profile, so in Q4 we can compare brands that added high-authority links with brands that added ordinary ones and see which group gained AI mentions.
Link-building budgets are often spent chasing authority. If, as far as AI search visibility is concerned, authority still adds nothing once volume is accounted for, that money might be better spent on breadth.
The two products behave differently today. AI Mode overlaps with organic results about as much as AI Overviews do, yet the two share only about 1 in 10 cited websites for the same question. We’re tracking whether AI Mode grows as a share of Google usage, whether its reliance on ranking pages holds, and how much of its citations go to Google’s own properties.
If AI Mode becomes the default, the sources that keep brands in AI Overviews may not transfer, and Search Console’s new report lets you watch your own split between the two.
Identify the questions buyers ask before they shop, publish the clearest and best-sourced answer available, and track whether AI cites it. Finding 01 suggests those citations travel with the brand into consideration-stage answers.
of brands cited on research questions were named on consideration questions
See Finding 01 →Count the publications, lists, and communities that name your brand, and grow that number. Authority scores alone did not predict AI visibility this quarter.
third-party web mentions for a 50/50 chance of being named on at least 3 of 6 platforms
See Finding 03 →AI Overviews now appear on almost every category search and cite ranking pages far more than ChatGPT, Perplexity or Copilot do. Keep SEO and AEO in one plan.
of AI Overviews citations ranked in Google's organic top 10 for the same question
See Finding 04 →Our team works with marketing leaders who need to act with confidence. Let's talk about what the data means for your brand.