Brands in AI answers, October 2026 in review: named by others, cited for a day

At mid-month our first reading of October argued that a brand can be cited by an AI assistant and still not be the brand it recommends. The second half of the month did not overturn that. It added three harder findings.

First, most of what assistants cite about a brand is not on the brand's site: Muck Rack's October count puts owned media at 15.2% of AI citations. Second, a citation is not a position but a reading that changes daily: Ahrefs found half or more of day-one citations gone on day two on every engine it checked, and SISTRIX showed that one prompt can miss a brand's true share by 34 points. Third, the answer around the brand is being sold: Google took ads in AI Mode to ten more countries and into follow-up questions, while OpenAI will rewrite ChatGPT ads by default.

This analysis follows those threads by theme, not by date, and ends with what to watch in November.

October's evidence in one table

October was unusually rich in measurement. The table lists the studies this analysis relies on, with the number that matters most for a brand. Their samples, markets and methods differ, so they are best read side by side rather than added up.

Named is not cited, and naming starts in memory

The split between being a source and being the answer was the headline of late September, and October's studies widened it. Vaer AI's study for Lightspeed Commerce, published on 29 September, analysed about 500,000 shopping answers. Shown a large and a smaller retailer without labels, models picked the larger one 90–94% of the time, and about ten chains took 40–44% of all recommendations, yet small retailers appeared in the cited sources far more often than in the recommendations. Asking for independent stores lifted small retailers from about a third to nearly four-fifths of those named; local and near me helped far less, because a nearby branch of a chain counts as local.

Semrush's first UK benchmark, on 13 October, found the same gap at national scale. Fewer than half of the most-mentioned brands were also among the most-cited domains. Tesco and Asda were both top-100 mentions on all four platforms every month, yet in July Tesco ranked 25th among domains ChatGPT cited and Asda 606th.

Graphite, in a study dated 16 October and written up on 26 October, offered an explanation for why the gap exists. It analysed 3,115 entity-and-question pairs in logged-out ChatGPT and about 957,000 fan-out queries, and compared what ChatGPT says about a brand without search with what it says with search.

In plain terms: what the model already believes about a brand shapes whether it names that brand once it searches. Brands ChatGPT rarely names from memory "reach only 11% median visibility with search when their visibility without search is in the lowest band, 0–25%", even when they appear on the pages it cites. Graphite sells answer-engine optimisation, which Search Engine Journal noted, and the study does not control for brand awareness, which is the obvious alternative explanation.

On 6 October a third layer appeared on top of naming. Brodie Clark spotted ChatGPT testing labels on organic product results: a green "best all-rounder" for one product and grey labels such as "best for beginners" for others. OpenAI has not announced it. If it ships, an answer will not only list brands but rank them, and a brand can keep its citation share and still lose the green label.

Our reading: a brand's presence in an answer is now at least three results — cited, named, labelled best — and each has different inputs. Citations follow pages. Naming follows what the model has absorbed about the brand over time, which is mostly what others have written. Labels follow criteria nobody has published.

What assistants read: mostly other people

The month's most direct answer to "where does AI get its view of a brand" came from Muck Rack's October edition of "What Is AI Reading?", last updated on 28 October. It sorts the citations in AI answers by who published them. Between May and September journalism's share fell from 27% to 22%, while brands' own sites rose from 13.7% to 15.2%. Third-party sources together still make up the large majority.

Which third parties count depends on the engine and the question. Ahrefs compared 2.4 million US ChatGPT answers from March with 3.8 million from September, reported on 12 October: Reddit's share of answers citing it fell from 19.1% to 2.3%, Wikipedia's from 14.1% to 2.9%, PubMed rose from 64th to 4th, and Britannica, Investopedia and the New York Post left the top 200. Ahrefs reads it as ChatGPT "prioritizing original sources over sites that summarize them".

Searchable, reported by Press Gazette on 22 October, ran 30,000 prompts about 100 UK towns and cities through ChatGPT, Perplexity, Gemini, AI Mode and AI Overviews and logged 244,725 citations. ChatGPT used council or government sources for 81.3% of questions and local news for 0.9%; on crime questions, police websites supplied 57.8% of answers. For a local business or a regional brand, the official source is the competitor for the citation, not the local paper.

Access matters too. Muck Rack found that "among major publishers blocking Claude's search crawler, 81% received almost no Claude citations", against 12% of publishers that allow it.

The two months of data point the same way. Original material — first-hand data, official records, tests — gained in ChatGPT; summaries lost. Third-party coverage outweighs owned pages everywhere Muck Rack looked. A brand's own site is necessary, but on these numbers it supplies about one citation in seven.

Facts that arrive without a link

Two findings this month concern what reaches an answer without a visible source. On 28 October Dan Petrovic of DEJAN published research suggesting that AI Mode receives whole Knowledge Graph biographies of people and companies as results without a URL. His summary: "A fact that only exists in the Knowledge Graph reaches AI answers without a link to you." Google has not commented, and Petrovic himself notes that the schemas show "the data structures exist", not how Google uses them. If the knowledge panel about a company is wrong, this is one more reason to fix it: the error may travel into answers with no page to trace it back to.

On 23 October Brodie Clark noticed that the citation cards beside AI Overviews sometimes show a short AI-written description of the cited page instead of the page's own text, "summarised based on the information that Google thinks is most relevant to the query." Search Engine Roundtable reproduced it on some queries and not others; Google has not commented. Combined with the 15 October change in which Google said it had increased inline links in AI Mode and AI Overviews and was testing a section that highlights links to official sites, the pattern is mixed: more links, higher up, but described more often in Google's words than the brand's.

A citation lasts days; a measurement needs dozens of questions

Profound set the frame on 30 September: following about 883,000 pages across seven engines for a year, the median page lost half its peak citation share 11 days after the peak, 78% of page-engine pairs had a half-life of two weeks or less, and correlations between most pairs of engines ran from 0.03 to 0.09. October measured the same instability at the scale of a single day.

Ahrefs, on 29 October, asked five AI tools the same questions every day for 20 days and followed every site cited on the first day, about 37,500 citations. Half or more were gone the next day on every tool.

Most of the churn is rotation rather than loss, which is why Ahrefs advises: "Don't judge AI visibility from one check" and "Count how many of the last seven days you were cited." An arXiv preprint submitted on 11 October found the same at page level: of the pages ChatGPT retrieved, cited or not, 40.8% on GPT-5.3 Instant and 44.5% on GPT-5.5 Instant appeared in only one of 20 runs of the same question. The paper has not been peer-reviewed.

SISTRIX, on 28 October, asked how many questions a measurement needs. It simulated a brand whose true share of answers is 25%. One prompt asked on 20 days "could land anywhere between 0 and 59 per cent"; the typical error is "±34 percentage points for one prompt, ±9.5 for 20 prompts and ±4.2 for 100." For ±5 points it puts the need at around 70 prompts, for ±3 at around 200. Two further findings matter for anyone who reports visibility to a client: of all brands that appeared in a prompt's answers, "70 per cent show up on fewer than one in five days", and when a brand chooses its own prompt set, "the brand performs better in its own set in three out of four cases, usually by a clear margin."

Put together, October's studies retire the screenshot: a handful of a brand's own prompts, checked once, measures mostly luck.

The brand's own query now opens with Google's summary

The one results page a brand used to control changed fastest. Chris Long reported on 30 September that 93% of about 100 brands he tested had an AI Overview on their brand-name search, usually partway down the page below the brand's own result, and DemandSphere said AI Overviews on branded queries "tripled in September". SE Ranking measured it at larger scale on 22 October: 1,000 US and 1,000 German brand keywords across 17 industries, each checked three times on desktop in October.

Two details matter more than the headline. Smaller brands get more AI Overviews on their own name than the largest ones, so it is the smaller brand whose own name most often opens with Google's summary. And the Overview cites different pages from the ones that rank: "Homepages make up 71.1% of the brand-site URLs in the organic top 10, but only 36.8% of brand-site citations in AIOs." Product pages, About pages and help pages are quoted instead.

The same month brought a reminder that the brand's query is shared territory on Bing too. Topics Portfolio, added in beta to Bing Webmaster Tools on 13 October, groups the queries a site is cited for in Copilot and Bing's AI answers and lists where other sites are cited and yours is not. Microsoft's own example shows a competitor's brand under "Running Shoes" as an opportunity, and Microsoft asks site owners to "treat the numbers as directional signals for planning, not as a precise ledger."

Named often, clicked rarely

Visibility and traffic stayed far apart. Marie Haynes, on 21 October, looked at 12 months of GA4 data from 100 websites with more than 145 million sessions: "sessions from ChatGPT amounted to just 0.62% of the volume driven by Google Organic." ChatGPT visitors converted at 13.8% against 3.02%, but Google still drove 96.5% of all conversions. Kevin Indig, working with 12 sites, wrote that "we also can't yet find a reliable relationship between AI Visibility and branded search or direct traffic"; we saw that finding only in Search Engine Roundtable's summary.

Even that thin traffic was miscounted for a week: after GPT-6 arrived on 7–8 October many ChatGPT links lost their tag and referrer until a fix began on 15 October; Adobe's Cedric Huesler says OpenAI confirmed to him that it was a bug. Our analytics analysis covers it. For brands the point is simpler: an answer can name a brand without a click, and a click can arrive without a name. What an answer says about you is measured in the answer.

The answer becomes ad space

Paid placements moved into AI answers on both sides during the month, and several of the changes bear directly on how a brand is presented.

On ChatGPT, OpenAI emailed advertisers on 20 October that "creative enhancements" will be on by default for self-serve campaigns from 29 or 30 October, depending on which OpenAI page you read. The system can rewrite headlines and descriptions, translate ad text, resize images and send the click to what it judges a more relevant page on the advertiser's site. One toggle in Ads Manager turns all four off.

On 22 October OpenAI described an auction that can use earlier turns of the conversation, and on 26 October some ChatGPT ads were reported to be bought through Amazon DSP.

Google moved further. On 17 October it was reported testing ads in AI Mode for Personal Intelligence users, a surface that had been ad-free. On 28 October it announced ads in AI Mode for ten more countries — Australia, Canada, India, Kenya, Malaysia, New Zealand, Nigeria, Pakistan, the Philippines and Singapore — with Brazil, Mexico and the UK planned for the fourth quarter, and multiturn ads on follow-up questions in the US. Its case to advertisers is that "click-through rates are already similar between sponsored and organic product list items within AI Mode", a claim without a number. On 29 October a test mixed sponsored and organic products in one AI Mode unit; Google said "All sponsored listings are labeled". The same week Alphabet's Philipp Schindler said the growth in AI Mode queries had weighed on Search revenue "because of the lower monetization".

Paid slots also describe brands in new words. Google's AI review summaries in Search ads, documented on 26 October, are "positive only"; Microsoft Advertising, on 28 October, lets advertisers set up to 40 guidelines "to help steer generated copy and protect your brand". And from 15 October Universal Cart keeps items from many stores in one cart across Search, AI Mode and the Gemini app, which makes accurate product feeds part of agentic commerce visibility.

Our reading: for a brand, the competitor's ad and the platform's rewrite of your own ad are now both part of the answer about you. A brand that tracks only organic citations sees half of that answer.

Who gets paid to be the source

The month opened with Judge Amit Mehta's opinion, signed on 30 September, dismissing Penske Media's and Chegg's federal antitrust claims over AI Overviews: "an expectation is not an agreement." It left their California unjust-enrichment claims — the ones about using content without payment — undecided, and came with a final, appealable order.

Payment is arriving by deal instead. The Information reported on 30 September that Google's AI contribution pilot pays about 100 publishers, and for several small and midsize sites "less than one-tenth of 1% of their advertising revenue." On 15 October Google said its News AI pilot covers more than 400 publications, with CNN, Axios and the Financial Times joining. USA Today Co sued OpenAI on 9 October.

Brands sit on both sides of this. Their own sites may be crawled under terms they did not negotiate: Meta's revised Commercial Terms let it crawl domains associated with any business that uses Meta products; in mid-October Meta removed "train" from the clause, kept the crawl, and later moved the start from 30 October to 20 November. And the publishers that carry a brand's earned coverage are the ones whose access to each assistant is being priced, blocked or litigated. Muck Rack's Claude finding points to the mechanism: a publisher that blocks a crawler largely disappears from that assistant's citations, and with it the coverage of every brand it wrote about.

Manipulation: easy to do, now against the rules

October produced the clearest evidence yet of how easily an AI answer can be steered, and the clearest rules against doing it. A preprint posted to arXiv on 8 October by Qi Liu and colleagues, which Search Engine Journal attributes to Fudan University, posted about invented subjects, including a "Cactus Gravity Protocol", on sites AI search tools cite often and a new account can post to, such as Medium, YouTube and Sohu. Within seven days, eight of ten tools were citing the posts. The paper has not been peer-reviewed.

Placement mattered more than volume: forty posts on a self-hosted WordPress site drew no reference in seven days, while a post on a source the tools already cite drew the first reference every time it was tested. The tests predate Google's 15 May statement that its spam policies cover attempts to manipulate AI responses.

The rules caught up in the same weeks. Anthropic's usage policy, published on 8 October and effective 12 November, forbids using Claude to "manipulate the sources from which search engines or AI systems draw answers by seeding them with content that misrepresents its origin, authorship, or independence", with networks of sites posing as unaffiliated sources as the example. Google's helpful-content guidance, rewritten on 1–2 October, now calls fabricated creator profiles "a form of deception". DeepSeek, opening its search engine to developers on 17 October, says it filters "GEO manipulation" without defining it.

The lesson for a brand cuts both ways. A brand's citation share can move because of what strangers post on Medium or YouTube, so the sources an assistant cites for its category are worth watching, not only its own pages. And any agency offering a network of "independent" review or news sites is now offering what Anthropic's policy names outright and Google's spam policies already cover. Generative engine optimisation that earns coverage is one thing; manufacturing the appearance of coverage is another.

Links and digital PR: what still moves AI citations

Link building had its own October story. On 26 October Majestic said it had removed a wave of spam built to get into backlink indexes themselves: "This year, we've seen a HUGE increase in toxic link spam that specifically targets backlink indexes like Majestic." More than 10,000 sites appear to have been deployed to insert adverts for link building services, and "the number of backlink URLs removed from the index is in the hundreds of billions." Many sites will see referring domain counts drop and Trust Flow shift, though "the core Trust Flow mechanics remain unchanged." Asked three days later about spammy links shown in a third-party tool, John Mueller said: "if these are random spammy links on the web, I'd just ignore them."

The more important question for brands is what links do for AI visibility. Ahrefs' stability study answered it directly. Long-lived and short-lived citations came from similarly strong sites, typically Domain Rating 64 to 78, and "Domain Rating and links to the page made little difference" to how long a citation lasted. What did matter on Google's surfaces was ranking: on AI Overviews and AI Mode, pages that ranked anywhere in Google for the same question "were about seven times more likely" to keep their citations for at least a week.

How far a Google ranking carries depends on the engine. TJ Digital's study, dated 25 October and written up by Search Engine Journal on 28 October, covered 193,507 answers to 4,167 buyer questions that name no brand. Of the pages in Google's top ten for a question, each engine cited a different share.

So the chain from a backlink to an AI answer, on October's data, runs through search rankings and mostly through Google's own answers. For ChatGPT the stronger lever is the one Muck Rack measured from the other side: third-party coverage, which is what digital PR earns, is most of what AI engines cite. The Fudan preprint adds that the venue matters more than the count. For a PR or link-building programme, that means coverage in the publications an assistant already reads and original data others will cite are worth more than link volume; links that help a page rank still help, on Google's surfaces most.

What it adds up to

This section is our reading, not a measurement.

• The evidence about a brand is mostly off its site. Owned media is about 15% of AI citations; third-party corporate content, journalism and aggregators make up most of the rest. AI visibility is a PR and reputation outcome as much as a site outcome.
• A citation is a reading, not a position. Half or more vanish overnight, three in four return, and one prompt can miss the truth by 34 points. Only repeated measurement over many questions, per engine, says anything.
• Naming starts before search. What a model already believes about a brand predicts whether it names it after searching. That belief is built slowly, from coverage, and cannot be fixed with one page.
• The brand query belongs to Google's summary. With AI Overviews on more than four in five US brand searches, and more often for smaller brands, the cited About and help pages now do the work the homepage used to.
• The answer is now ad space. AI Mode ads in the US and ten more countries, and in follow-up questions, ChatGPT ads rewritten by default, sponsored and organic products in one unit. A brand's organic position and its rival's ad share the same answer.
• Links help through rankings; coverage helps directly. Domain strength did little for how long a citation lasted. Ranking in Google helped on Google's AI surfaces; coverage in cited sources helps everywhere.
• Manufactured independence is now named. Anthropic bans faking independent sources, Google calls fabricated authors deception, DeepSeek says it filters "GEO manipulation". The Fudan test shows why the rules are needed.

What to watch in November

• 12 November: Anthropic's usage policy on seeding AI answer sources takes effect. Watch whether other model makers write the same line.
• 20 November: Meta's revised Commercial Terms, with the crawl clause, take effect unless delayed again; the UK's CMA said it will keep scrutinising them.
• AI Mode ads in Brazil, Mexico and the UK, planned for the fourth quarter, and whether Google reports AI Mode separately from ordinary Search and Shopping campaigns.
• ChatGPT's "best" labels and creative enhancements: whether the labels ship and how they are chosen, and how often rewritten ads send clicks to pages the brand did not pick.
• AI-written citation cards in AI Overviews: whether the test becomes the default, replacing a page's own description with Google's.
• The next Muck Rack edition: whether journalism's share keeps falling and owned media keeps rising.
• Penske and Chegg: whether either appeals the antitrust ruling or refiles the state-law claims.

To read what three assistants say about your brand, across the buyer questions rather than one, and compare it with your own page, Brandscope runs them in one reading.

Sources

Every claim above is dated and linked. Where a number comes from a measurement rather than an announcement, whoever measured it is named.

• When you ask AI where to shop, it sends you to the big guys — Vaer AI for Lightspeed Commerce, 29 September 2026
• The half-life of an AI citation is 11 days — Profound, 30 September 2026
• Google AI Overviews Return For Many Large Brand Names — Search Engine Roundtable, 30 September 2026
• OpenAI Tests New Product Labels On ChatGPT Results — Search Engine Roundtable, on Brodie Clark's report, 6 October 2026
• ChatGPT's Most-Cited Sites Tilt Toward Primary Sources, Data Shows — Search Engine Journal on Ahrefs, 12 October 2026
• Semrush 2026 AI Visibility Index: first UK benchmark — Semrush, 13 October 2026
• Parametric knowledge optimization — Graphite, 16 October 2026
• Some Brands On ChatGPT's Cited Pages Still Got Few Mentions — Search Engine Journal, 26 October 2026
• What is AI reading? October 2026 — Muck Rack, last updated 28 October 2026
• AI answer engines often bypass local news outlets — Press Gazette on Searchable, 22 October 2026
• Knowledge Graph biographies in AI Mode — DEJAN, 28 October 2026
• Google AI Overview Citation Cards With AI-Generated Summary Snippets — Search Engine Roundtable, 23 October 2026
• More ways to connect with the web from Search — Google, 15 October 2026
• How stable are AI citations? — Ahrefs, 29 October 2026
• Understanding ChatGPT Citations: Evidence from Production Web Search Across Model Generations — arXiv preprint, 11 October 2026
• OpenAI's API Only Partly Matched ChatGPT's Citations, Paper Finds — Search Engine Journal, 18 October 2026
• Measuring AI visibility — SISTRIX, 28 October 2026
• AI Overviews Appear on 82.5% of US Brand Searches — SE Ranking, 22 October 2026
• Topics Portfolio (beta) in AI Performance — Bing Webmaster Blog, 13 October 2026
• GEO is a distraction — Marie Haynes, 21 October 2026
• 3 New Studies Suggest AI Visibility & Search Not Worth Investing In — Search Engine Roundtable, 22 October 2026
• OpenAI Drops UTM Parameters From ChatGPT Results (GPT-6) — Search Engine Roundtable, 14 October 2026, updated 15 October
• OpenAI Auto Enabling Creative Enhancements In ChatGPT Ads — Search Engine Roundtable, 20 October 2026
• OpenAI to rewrite ChatGPT ads by default as advertisers get days to opt out — PPC Land, 21 October 2026
• Inside ChatGPT's Ad Auction: How Relevance, Bids, And Memory Decide Placement — Search Engine Journal, 22 October 2026
• ChatGPT Ads With Learn More Arrow Link; Some Served Through Amazon DSP — Search Engine Roundtable, 26 October 2026
• Google Tests Ads in AI Mode for Personal Intelligence Users — The Keyword, 17 October 2026
• Ads in AI Mode are expanding globally — Google, 28 October 2026
• Google AI Mode tests sponsored and all products together — Search Engine Roundtable, 29 October 2026
• Google: AI Mode ads earn less — Search Engine Roundtable, 29 October 2026
• Google Ads AI review summaries help document — Search Engine Roundtable, 26 October 2026
• What's new in Microsoft Advertising this October — Microsoft Advertising, 28 October 2026
• Google's Universal Cart reaches US shoppers as Meta flips browsing on — PPC Land, 16 October 2026
• Penske AI Overviews lawsuit dismissed because 'no formal bargain' struck with Google — Press Gazette, 1 October 2026
• Judge dismisses Chegg's and Penske's AI Overviews cases: an expectation is not an agreement — The Decoder, 4 October 2026
• Google Paying ~100 Publishers 0.1% Of Ad Revenue In AI Contribution Pilot — Search Engine Roundtable, on The Information's report, 30 September 2026
• 400 publications are now partnering with us on our News AI pilot program. — Google, 15 October 2026
• CNN, Axios and other publishers join Google's AI news programme — Bloomberg via BNN Bloomberg, 15 October 2026
• News publisher AI deals and lawsuits tracker — Press Gazette, updated 12 October 2026
• Meta Drops Training From Its Crawl Clause And Keeps The Crawl — Search Engine Journal, 19 October 2026
• Meta Delays The Terms That Let It Crawl Business Sites — Search Engine Journal, 26 October 2026
• From Public Posts to AI-Search Citations: Measuring the Fragility of AI Search — arXiv preprint, 8 October 2026
• AI Search Tools Cited A Made-Up Concept Within A Week, Paper Finds — Search Engine Journal, 19 October 2026
• Usage Policy — Anthropic, effective 12 November 2026
• Updating our Usage Policy — Anthropic, 8 October 2026
• Google Adds Fake Author Warning To Helpful Content Guidance — Search Engine Journal, 2 October 2026
• DeepSeek Search API — DeepSeek API Docs, 17 October 2026
• Fighting referring domain spam — Majestic, 26 October 2026
• Search Console does not show all links — Search Engine Roundtable, 29 October 2026
• 2026 AI citation study — TJ Digital, 25 October 2026, and Search Engine Journal's write-up, 28 October 2026