Market analysis

· 14 min

AI assistants in October 2026: the questions spread out, the clicks did not

Three model generations shipped in five weeks, Gemini and Claude took a third of US prompts from ChatGPT, and ChatGPT still sends most of the clicks. The market is splitting along a line most analytics cannot see.

AI
AI56%GOOGLE & SEARCH10%PPC6%ANALYTICS14%BRANDS14%
Topic radarSelect a topic to filter the archive. The radar shows how this issue's words divide across the topics.

Signal readout

Topic mix

  • AI: 56%
  • Google & Search: 10%
  • PPC: 6%
  • Analytics: 14%
  • Brands: 14%
  • AI56%
  • Google & Search10%
  • PPC6%
  • Analytics14%
  • Brands14%

Share of the issue's words about each topic, counted section by section.

Why it matters

Measure AI visibility in the answers, not in your analytics. Referral clicks come overwhelmingly from one assistant, while questions are moving to assistants that rarely send a click. A brand that is named in Gemini or Claude answers will not see it in its traffic report, and a brand that disappears from them will not see that either.

The short answer

What
A reading of the AI assistant market in mid-October 2026: which models shipped, how the prompts and the referral clicks divide between assistants, how answers are changing shape, and where ads, agents, publishers and regulators stand.
Why
Because the share of questions and the share of clicks have come apart. A brand that measures AI only through referral traffic is watching the assistant that is losing share of questions and missing the ones gaining it.
Who
Marketers, SEO and GEO practitioners, and anyone deciding where to measure or spend on AI visibility.
Where
Global, with US panel data where only US data exists (Comscore), and EU and UK regulation where it applies.
When
July to 16 October 2026, against the trend since January.
How we know
From the vendors' own announcements (OpenAI, Google, Anthropic, Meta), panel and clickstream data (Comscore, Similarweb, Statcounter, BrightEdge, Chartbeat), trade press that reported the rest, and court and regulator documents. Every source is listed at the end.

In plain words

A year ago almost everyone who asked an AI a question asked ChatGPT. Now many people ask Google's Gemini or Anthropic's Claude instead. But when an AI answer sends someone to a website, it is still mostly ChatGPT doing the sending, because the other assistants keep people on their own page. So a website that only counts its visitors sees a much smaller and older picture of AI than the one its customers actually live in.

Between 22 September and 13 October four of the five largest assistants changed their main model. Anthropic released Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 in three weeks; OpenAI put GPT-6 in front of every ChatGPT user on 7 and 8 October; Google launched Gemini 4 Argon on 13 October. A new model is now the background condition of the market, not news in it.

What did change this quarter is the shape of the market around those models. The questions are spreading across assistants. The clicks are not. Answers are turning into interfaces, the interfaces are taking ads, and agents are beginning to buy. This analysis takes each of those in turn, and ends with what we think they add up to. Where a claim is our reading rather than a measurement, it says so.

Three generations in five weeks

OpenAI shipped GPT-6 to Plus, Pro, Business and Enterprise on 7 October and to Free and Go users on 8 October. Paid plans get GPT-6 Sol, the free tiers GPT-6 Luna. It arrived with what OpenAI calls Intelligent UI: an answer can now contain charts, buttons, forms and small working tools rather than text with links. OpenAI says GPT-6 Instant starts answering web-search questions 44% sooner on average than GPT-5.6 Instant. Citations stay behind a Sources button, and OpenAI has not said how sources will appear inside a generated chart or tool.

Google's Gemini 4 Argon reached the Gemini app and AI Mode on 13 October, for Google AI Pro and Ultra subscribers only; free users keep the default model. AI Mode had already moved to Gemini 3.8 Flash on 2 September, so its underlying model changed twice in six weeks. Anthropic released Claude Opus 5.5 on 22 September, describing it as working at the level of Claude Fable 5.1 on most tasks at 40% lower running cost than Opus 5, followed by Sonnet 5.5 on 28 September and Haiku 5.5 on 7 October.

Outside the three, Meta launched Muse on 8 September in the US, a personal agent that browses, fills forms, sends email and buys; xAI released Grok 4.6 around 12 August; DeepSeek's newest family is still V4, released with open weights in April, and we found no major DeepSeek release this quarter.

The practical consequence for anyone measuring visibility: the system that decides which sources an assistant reads can change without any change in the web. A page's citation share can move because a model was swapped, not because anything on the page did.

The questions are spreading out

Three independent datasets agree on direction. Comscore's US panel puts ChatGPT at 50% of AI prompt volume in June 2026, down from 70% in January, while Gemini rose from 17% to 30% and Claude from 2% to 11%. Similarweb's worldwide share of generative-AI website visits for September has ChatGPT at 56.4%, Gemini at 25.0% and Claude at 9.6%, against 71.8% and 13.6% for the first two a year earlier. Similarweb also notes that Gemini's share slipped for four months running, so the shift is not a straight line, and it counts websites only, not apps.

Google's own numbers point the same way at larger scale. The Gemini app passed 1 billion monthly users in August, which Google called the fastest-growing product in its history. On the Q2 earnings call on 22 July, Sundar Pichai said AI Mode had passed 1 billion monthly users and that AI Overviews and AI Mode had been merged into one Search experience. OpenAI's most recent figure is 1.2 billion weekly ChatGPT users, which Sam Altman gave at DevDay on 29 September.

MeasureChatGPTGeminiClaudeSource
US AI prompt volume, June 202650%30%11%Comscore
Gen-AI website visits, September 202656.4%25.0%9.6%Similarweb
AI chatbot referrals to websites, September 202680.48%10.91%0.95%Statcounter
AI referral traffic, August 202695.1%2.4%0.8%BrightEdge

The clicks are not

The bottom two rows of that table tell a different story from the top two. Statcounter's September data puts ChatGPT at 80.48% of chatbot referrals to websites, Gemini at 10.91%, Perplexity at 5.85%, Copilot at 1.79% and Claude at 0.95%. BrightEdge's August data is more extreme: 95.1% of AI referral traffic came from ChatGPT, whose referrals grew 101% between January and August.

Our reading: the assistants gaining share of questions are the ones that keep users on their own page. Gemini answers inside Google; Claude is used heavily for work that never needed a link. So a brand that measures AI through referral traffic is watching the assistant whose share of questions fell by twenty points this year, and seeing almost nothing of the two that gained it.

Referral traffic is small in any case. Across Chartbeat's publisher network, AI referrals grew eightfold in two years and are still about 0.01% of traffic. And the one dependable label on ChatGPT clicks proved fragile this month: after GPT-6, many links arrived with no utm_source=chatgpt.com tag and no referrer, so analytics filed them as direct. Adobe's Cedric Huesler measured ChatGPT visits falling by about 31,000 a day on some 500 pages while direct visits to the same pages rose by about 26,000. On 15 October he reported that OpenAI called it a bug and that tagging was returning; the referrer is still missing on most clicks, and visits already filed as direct stay there. We covered the episode in the 15 October issue.

Try it on your own page

See what PatternScope finds on a page of yours

Everything above is measurable on your own work, and the check takes about a minute. Paste one address and read what comes back.

Open PatternScope

Free · no account · one at a time

Answers become interfaces, and interfaces take ads

Intelligent UI is the visible half of a broader change: answers are becoming places where things happen rather than pages of text that point elsewhere. That changes where a source can appear. A citation in a paragraph is a link; a citation inside a comparison widget is whatever the interface decides to show.

Advertising has moved in at the same time. ChatGPT ads, tested in the US from 9 February, reached 31 European markets on 24 August and seven Asian markets on 23 September; OpenAI says ads now run in more than 60 countries, shown only to Free and Go users. On 31 August OpenAI said ChatGPT Ads had reached a $1 billion annualised revenue run rate in under 200 days. Measured performance is modest: Graphite, using Similarweb panel data, found click-through per ad-bearing exchange of 0.73% in the US and 0.74% in the UK, and 1.91% in Brazil. The PPC side of this is in our separate analysis of paid search.

In less than 200 days after launch, ChatGPT Ads has reached $1 billion in annualized revenue run rate.

Agents start to buy

This quarter agents moved from demonstrations to defaults. Meta's Muse completes purchases with a one-time card through Stripe Link and asks the user before sensitive actions. Chrome's Gemini auto browse reached Android phones for US AI Pro and Ultra subscribers on 18 August, after launching on desktop in January; Google says it will not place an order without approval. On 22 September Google Merchant Center told Shopify merchants that native checkout inside AI Mode and Gemini, built on its Universal Commerce Protocol, was on by default, with an opt-out in the Shopify admin. At DevDay OpenAI announced Dots, always-on agents that act without being asked.

Not every move went one way. In March OpenAI stepped back from Instant Checkout, saying merchants could use their own checkout while it focused on product discovery. For a brand, the question is shifting from "does the model mention us" to "can an agent verify our price, stock and terms and complete the order". That is a data-quality question as much as a marketing one.

What the assistants read

Which sources an assistant cites is not stable, and it is not the same between assistants. Ahrefs compared the domains ChatGPT cited in 2.4 million US answers in March with 3.8 million in September: Reddit fell from 19.1% of answers to 2.3%, Wikipedia from 14.1% to 2.9%, and PubMed rose from 64th to 4th. Only 14.4% of free, logged-out responses ran a web search at all. Profound followed 883,000 pages across seven engines for a year and found the median page loses half its citation share 11 days after its peak, with almost no correlation between engines in how long a page stays cited (0.03 to 0.09).

Our reading: there is no single "AI" to be visible in. There are several engines with different appetites, each changing on its own schedule. A single audit describes one engine on one day. The brands analysis covers what this means for who gets recommended.

Publishers are paid case by case

The courts have narrowed one route to compensation. On 30 September Judge Amit Mehta dismissed the federal antitrust claims of Penske Media and Chegg over AI Overviews, writing that "an expectation is not an agreement"; he left their state unjust-enrichment claims undecided. Payment is moving to bilateral deals and pilots. Google is testing an AI contribution programme in Search Console that pays publishers when their content significantly contributes to answers; The Information reported about 100 participants, with payments under 0.1% of ad revenue for small sites. OpenAI signed Village Media on 17 September and the Times of India group and Indian Express on 23 September, its first deals in India. The suits continue: Seattle Times and Newsday sued OpenAI and Microsoft on 7 September, and USA Today Co sued OpenAI for $250 million on 9 October.

Discovery in the New York Times case against OpenAI and Microsoft has also produced the clearest numbers yet on what answers do to clicks. The publishers' brief of 17 September cites Microsoft's internal data showing Copilot and Bing Chat answers cut clicks to New York Times properties by 87–93%, and the same brief puts the share of ChatGPT users who visit no external site during a search at 87.78%, against 26.91% for Google. Those are the plaintiffs' framing of discovery material, not independent studies, and should be read that way.

Regulators move the dates, and name manipulation

The EU's Digital Omnibus on AI, in force since 27 July, postponed the AI Act's high-risk obligations under Annex III from 2 August 2026 to 2 December 2027, and those for product-embedded systems to 2 August 2028. Systems already on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marking under Article 50. The Commission's powers to enforce the obligations on general-purpose AI providers have applied since 2 August 2026.

In the UK, the CMA proposed on 23 September a search choice screen on Android and Chrome, shown at first use and then yearly, on which AI assistants that meet its criteria could appear; the consultation closed on 9 October and a decision is expected by the end of the year. We found no confirmed change in US federal AI rules this quarter.

Platforms are also writing their own rules against manipulating answers. Anthropic's updated usage policy, effective 12 November, bans using Claude to manipulate the sources AI systems draw answers from by seeding content that misrepresents its origin, authorship or independence. Google's documentation, rewritten on 1–2 October, now names fabricated author profiles as deception. Practices sold as generative engine optimisation that depend on fake independent sources are now explicitly against the rules of the systems they target.

What it adds up to

This section is our reading, not a measurement.

  • Two markets, not one. The market for questions is becoming plural; the market for clicks is still ChatGPT's. Budgets planned from traffic data will over-weight one assistant.
  • Google is keeping the journey on one page. AI Mode at a billion monthly users, AI Overviews merged into it, native checkout on by default: question, answer, follow-up and purchase without leaving Google.
  • Model changes are now a visibility event. AI Mode changed model twice in six weeks. Any month-on-month change in citations needs the model dates beside it.
  • Being cited, being recommended and being advertised are three different results. An assistant can cite your page, recommend a competitor and show a third party's ad in one answer. Each needs its own measurement.
  • Automated traffic already outnumbers people. Cloudflare says automated traffic passed human traffic in May 2026, about a year earlier than it had forecast. Logs that do not separate agents from people are now mostly logs of agents.

What to watch

  • Comscore's next report: whether Gemini and Claude kept gaining share of prompts after GPT-6.
  • How citations appear inside Intelligent UI widgets and Gemini 4 answers, and whether citation share moves after 7 and 13 October.
  • Whether Gemini 4 reaches free users in AI Mode, which would change the model behind most AI Mode answers.
  • The CMA's final decision on AI assistants in the choice screen, and 2 December, when the EU's Article 50 grace period ends.
  • Alphabet's Q3 results, expected at the end of October, for new AI Mode and Gemini usage figures.

To see what three assistants say about a brand today, Brandscope runs the buyer questions around it in one reading.

Questions people ask

Is ChatGPT losing the AI assistant market?

It is losing share of questions, not leadership. Comscore puts ChatGPT at 50% of US AI prompt volume in June 2026, down from 70% in January, and Similarweb at 56.4% of gen-AI website visits in September. It still sends most AI referral clicks: 80.48% in Statcounter's September data and 95.1% in BrightEdge's August data.

Which AI models shipped in autumn 2026?

Claude Opus 5.5 (22 September), Sonnet 5.5 (28 September) and Haiku 5.5 (7 October); GPT-6 in ChatGPT (7–8 October); Gemini 4 Argon in the Gemini app and AI Mode for paid subscribers (13 October). AI Mode moved to Gemini 3.8 Flash on 2 September.

Why does AI referral traffic underestimate AI visibility?

Because the assistants gaining the most questions, Gemini and Claude, send few clicks, and because answers increasingly satisfy the question on the page. Chartbeat measures AI referrals at about 0.01% of publisher traffic. A brand can be named in many answers and see almost none of it in analytics.

Are ads in ChatGPT available worldwide?

In more than 60 countries as of 23 September 2026, for Free and Go users only. They launched in the US on 9 February, reached 31 European markets on 24 August and seven Asian markets on 23 September. OpenAI says the business passed a $1 billion annualised run rate in under 200 days.

Sources

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

The product

Watch it on every page, every day

One reading tells you where a page stands today. The product asks the same questions of the same assistants continuously, so a change is something you are told about rather than something you go looking for.

Request an invitation

Invite-only while we keep the readings honest

Author

Olexandrauthor

15+ years in marketing and SEO. Dozens of products, mostly the US market. Research and measurement.

LinkedIn

Pass it on

LinkedInXRedditHacker News

Read in your language

FrançaisDeutschEspañolУкраїнськаPolski

Opens a browser translation of this English article. The original source stays in English.

All issues