76 Billion AI Sessions Reveal an Overlooked Brand Competition

Key findings

Graphite published The State of Gen AI on November 9, 2026. Its data presents a striking contrast: AI usage and AI referral traffic are growing quickly, yet AI referrals still account for less than 1% of average website traffic in its analysis.

The six largest generative AI products generate about 76 billion sessions per month worldwide, up 37% year over year. ChatGPT accounts for 81.2% of those sessions. Claude's sessions rose 1,901% year over year, while worldwide AI referral traffic grew 92%.

Google AI Overviews also appear in about 64% of US Google searches in Graphite's data. Together, these figures show that AI has become a large environment for information and task completion. Yet a website analytics dashboard may still make its contribution to customer acquisition look small.

The central question: as the way people find information changes, are companies still using old traffic metrics to judge a new form of brand competition? Graphite's report does not directly answer that question, but it provides important clues.

01 | AI is no longer niche, but usage is not website traffic

By September 2026, ChatGPT, Gemini, Claude, Grok, Perplexity, and DeepSeek together reached about 76 billion monthly sessions worldwide, up 37% from a year earlier. Growth was not confined to one product: Claude grew 1,901% year over year, Gemini 182%, and Grok 123%. ChatGPT remained the clear leader, but competitors expanded quickly.

Graphite also found that 89% of worldwide AI sessions occur on mobile devices, including apps and mobile web. AI is increasingly part of ordinary use, not only a tool opened occasionally by specialists.

For a business, though, 76 billion sessions do not mean 76 billion commercial needs or purchase decisions. People use AI to write emails, code, study, and organize information. They also use it to research vendors and compare solutions. Those tasks have very different commercial value.

A B2B SaaS company may care most about solution comparisons. A professional services firm may focus on how buyers assess expertise and credibility. A developer tools company may focus on technical selection and integration questions. The starting point for GEO is to identify the AI use cases that affect the business, rather than pursue all AI traffic.

02 | AI referrals rose 92%. Why is the impact hard to see?

Graphite found that worldwide AI referral traffic increased 92% year over year. Even so, AI referrals still represented less than 1% of average website traffic in its analysis.

Referral data records a click from an AI platform to a website. It does not record every occasion when AI helped someone learn about a market or shortlist a vendor.

The reverse is also possible: the buyer already knew the brand and used AI only to confirm a preference. Traditional referral data cannot reliably distinguish these paths. A low referral share therefore does not prove a low effect on awareness or research. Equally, a rise in branded search, direct visits, or conversions cannot simply be attributed to AI.

Graphite calls this the AI attribution gap. In a case study with automation platform n8n, it compared AI referred conversions with users' self reported sources and found that the former captured only a small fraction of relevant conversions. Graphite cautioned that one case study cannot establish a universal market wide ratio.

For companies evaluating GEO, the practical lesson is to investigate possible effects missed by referral analytics, then test them with additional evidence rather than estimate them without data.

03 | AI is entering existing search behavior

Graphite reports that AI Overviews appeared in about 23% of US Google searches in September 2025 and about 64% in August 2026. AI's reach therefore extends beyond standalone chat products into a search entry point people already use.

Previously, a user often selected a webpage from search results to get an answer. When an AI Overview appears, some information is assembled on the results page. Which sources users click, and whether they keep searching, may change.

Presence is not the same as a zero click rate. These figures do not establish how much organic traffic any particular industry lost. They do show that a company's content can help answer a question even when a user does not leave the search page.

A strong industry study may still generate rankings, organic visits, and conversions. It may also become a source for an AI answer that a user reads without clicking. Search traffic remains valuable, but clicks no longer necessarily capture every way content contributes to discovery. SEO and GEO should therefore be studied together: how does the same content support search discovery, AI understanding, and business outcomes?

04 | ChatGPT leads, but market share alone is not a GEO strategy

ChatGPT accounts for 81.2% of worldwide sessions across the six standalone AI products Graphite studied. In Graphite's global AI referral data, however, ChatGPT accounts for about 60%, followed by Gemini at 14% and Claude at 10%.

The two data sets have different scopes and cannot be used directly to calculate each platform's referral efficiency. They do, however, show why total platform usage cannot stand in for the commercial value of a platform to a particular company. User tasks, answer formats, citations, and external link behavior vary.

AI is also available through established search entry points. Graphite found about 491 million US web visits to Google AI Mode in August 2026, compared with about 360 million US sessions for the standalone Gemini product. These are different metrics, but the comparison illustrates the importance of looking beyond chat apps.

Companies should ask how their brand performs when target customers put the same business question to different AI platforms. A project management SaaS company, for example, might examine recommendations for remote teams, deployments above 100 employees, Jira integrations, and alternatives to Asana or monday.com. These prompts express different intent. A high aggregate mention count can hide absence from the scenarios most relevant to the product.

Compare the brand's competitive position across platforms within high value use cases. That tells a business more than a visibility score detached from customer needs.

05 | GEO needs competitive diagnosis

Growing AI use, integrated search, and incomplete referral attribution can tempt companies to treat GEO as another content production task: publish more blog posts, generate FAQs in bulk, or expand a prompt list. Those activities may help, but only if they address the reason a brand is missing or misrepresented.

The brand is not recognized as a candidate

A software company may have established customers in a niche yet never appear in relevant vendor selection answers. The first question is whether its positioning, use cases, and evidence are clear enough for AI systems to identify it as relevant.

The brand appears with incorrect information

An answer may cite outdated pricing, describe an existing feature as unavailable, or misstate an integration. Correcting these errors can matter more than increasing exposure because the inaccuracies can change a buyer's decision.

Competitors have stronger evidence

AI may find independent reviews, customer cases, technical documentation, and third party discussion for competitors but little reliable material about the brand. The problem may be the availability of verifiable evidence, rather than keyword coverage.

The brand appears in broad prompts but misses core needs

A company may occasionally feature in “best marketing software” answers while missing the specific scenarios that matter most to its buyers. More total mentions may not improve business outcomes.

Each situation calls for different work: clearer site content, better product information, updated documentation, genuine customer cases, independent evidence, or a sharper competitive position. Mature GEO work finds the gaps in high value AI questions and tests whether the changes actually improve performance.

06 | How should a company measure GEO?

If AI referrals cannot capture the full effect, measurement should examine three connected layers.

1. Performance on important questions

Build a set of real, business relevant questions from customer interviews, sales records, site search, support requests, and existing search demand. Track whether the brand is mentioned, described accurately, and placed in the right category. Note where competitors perform better and whether citations are dependable. This shows whether the brand can participate in the relevant AI information market.

2. Whether changes improve competitive performance

Respond to the specific problem: improve documentation, correct old information, publish authentic customer cases, clarify content structure, or strengthen independent evidence. Recheck the same question set. Because AI answers vary, a single appearance is not proof of success. Measure repeatedly and, where possible, compare with questions or pages that did not receive the same treatment.

3. Whether the change creates business value

Observe AI referred visits, branded searches, direct visits, qualified leads, and sales opportunities. B2B teams can ask people booking a demo or speaking with sales how they found the company. Each signal has limits: advertising and PR can affect branded search, self reports have recall bias, and AI referrals may not represent incremental customers. Phased rollouts, control groups, and other experiments can help test added value.

This approach avoids both dismissing GEO because referral traffic is small and claiming revenue impact merely because AI mentions increased. A credible result should explain what changed, how competitive performance moved, and which business outcomes the evidence supports.

07 | AI growth is fast, but its path is unstable

Graphite also describes a changing growth pattern. Sessions across major AI products climbed quickly from early 2024 to mid 2025, grew only about 2.3% in the second half of 2025, and accelerated again in 2026. Individual products followed different paths: Claude surged, Gemini grew and briefly fell back, Grok declined from a peak, and Perplexity contracted more noticeably.

A GEO plan tied to one platform's current traffic or one fixed answer pattern may therefore age quickly. Companies need ongoing observation: which platforms matter to their customers, which business questions are changing, where brand and competitor performance diverge, and whether new evidence improves those results. This does not mean chasing daily algorithm changes. It means treating AI discovery as a changing competitive environment.

Conclusion | GEO is about more than AI traffic

Graphite's report captures two truths at once. AI has massive use: about 76 billion monthly sessions across six products, 92% year over year growth in worldwide AI referrals, and AI Overviews across a large share of US Google searches. Yet AI referrals remain a small share of average website traffic.

These facts are compatible. AI may help a person learn about a market or compare solutions without producing a direct website click. That possibility is not itself proof of business value. Companies still need data and experiments to establish what AI changes for their own customers.

GEO should therefore address a more basic competitive question: when people use AI to understand a market and assess solutions, is the brand recognized accurately and competitive in the needs that matter? The capability worth building is a repeatable way to find gaps, improve the quality of brand information, and verify business results.