ChatGPT drives 92% of AI referral traffic: what brands must prioritize now
ChatGPT commands 92.4% of standalone AI referral traffic in 2026. The Previsible data changes how brands should prioritize their AI visibility strategy.
B
Stephan Charles
12 min read
July 24, 2026
Share
By Stephan Charles | Last fact-checked: 2026-07-24
In July 2026, Previsible published its 2026 State of AI Discovery Report, analyzing 6.77 million LLM-driven web sessions across 166 websites over 19 months. The headline: ChatGPT alone drives 92.4% of all standalone AI referral traffic. These are the sessions where users query an AI assistant and then click through to a brand's website. BrandCited is an AI brand visibility platform that tracks how often ChatGPT, Perplexity, Claude, Gemini, Copilot, Grok, You.com, Brave, and DeepSeek name your brand in their answers, scored on a 0-100 composite index updated every 30 days. The Previsible findings align with what BrandCited observes across its tracked-brand cohort: ChatGPT visibility gaps are the highest-impact gaps most brands are not fixing first.
Run a free BrandCited scan at brandcited.ai to see your ChatGPT citation rate across 9 AI platforms. Results in 30 seconds.
What does the 2026 AI referral traffic data actually show?#
The 2026 data shows ChatGPT dominates standalone AI referral traffic at 92.4%, while total LLM-driven sessions have grown roughly 10x in 19 months. Previsible's study tracked 6.77 million sessions from November 2024 to May 2026. Monthly LLM-driven sessions across all platforms grew from 65,249 in November 2024 to 644,478 in May 2026, a 9.9x increase in under two years.
Within that growth, the concentration toward ChatGPT is sharp. Search Engine Land reported that ChatGPT's 92.4% share has continued climbing, up from roughly 84% in December 2025. Perplexity's share of U.S. AI referral traffic fell from 19.7% in early 2025 to 6.85%, a 61% drop from its March 2025 peak, according to Previsible's report.
Track your AI visibility now
See exactly how AI engines like ChatGPT, Perplexity, and Gemini perceive your brand.
The BrandCited team covers GEO, AI search optimization, and brand visibility strategy. We publish research, practical guides, and product updates every week.
Spot an error in this article? Report it.
The Gemini story runs opposite to Perplexity. Gemini grew 3.2x over the study period and now sits as the second-largest standalone AI traffic source. Claude, despite a 64-fold growth in monthly sessions (133 to 8,528), still represents a small absolute share.
One important context from the report: Google's AI surfaces (AI Overviews and AI Mode) drive more AI-influenced discovery than all standalone LLM assistants combined. The Previsible data measures sessions that originate from an LLM and end with a click to an external website. AI Overviews clicks happen within Google search results and show up differently in analytics. Both matter; they require different optimization tactics.
Why ChatGPT's two citation modes change which brands get cited#
ChatGPT operates in two distinct modes: standard quick-answer and high-reasoning Thinking mode. The two modes cite different sources. A Semrush and Kevin Indig study of 200 ChatGPT responses across 20 buyer journeys found that the two modes share only 25.6% of cited sources.
The difference in behavior is significant:
Standard mode: 50% of responses include citations, averaging 2.6 citations per response, with 5.5 web sub-queries per prompt
Thinking mode: 68% of responses include citations, averaging 4.5 citations per response, with 24 web sub-queries per prompt
Source type distribution shifts in Thinking mode. Search Engine Land's coverage of the Semrush study shows Reddit's share falling from 15% to 7%, user-generated content dropping from 14.3% to 6%, and official documentation and support pages rising from 12.4% to 17.5%. Government and academic sources grew from 1.9% to 8.8%.
The practical implication: a brand that appears in ChatGPT's standard-mode answers may not appear when a user enables Thinking mode for a high-stakes buying decision. That is the exact moment when citation matters most. Brands need to audit both modes separately. BrandCited's ChatGPT probes now include reasoning-mode variants to surface this gap.
What does ChatGPT actually weigh when deciding which brands to mention?#
ChatGPT weighs two separate input categories depending on whether a query triggers a live web search or a parametric recall response. Roughly 60% of ChatGPT responses come from parametric memory, knowledge encoded during training. The remaining 40% involve live Bing web lookups, according to analysis by Contently.
For the 40% of lookup-driven answers, three factors dominate:
1Bing ranking position: Brands ranking in Bing's top results for the target query appear more often. ChatGPT's web search connects to Bing's index. A brand absent from Bing's results is absent from ChatGPT's browsed answers.
2Schema markup: Article, FAQPage, Organization, and Person schema give ChatGPT extractable, structured signals about who published the content and what it claims. Brands with complete schema markup appear in web-lookup answers at higher rates than brands without it.
3Answer-first content structure: ChatGPT extracts citations from content that places the direct answer in the first sentence of each section. Hedged prose and preamble-heavy content structures get cited at a fraction of the rate of direct, factual openings.
For the 60% parametric answers, a different signal stack matters:
Wikipedia and Wikidata presence: ChatGPT cross-references brand identity against Wikipedia, Wikidata, and Crunchbase when deciding whether to name a brand with confidence. A brand without a Wikipedia entity or Crunchbase profile is harder for the model to attribute claims to.
Cross-domain mention frequency: Brands that appear on many distinct domains in training data get recalled at higher rates. A PR mention on TechCrunch or a feature in a trade publication carries more parametric weight than 100 blog posts on owned channels.
Entity consistency: The brand name, product description, and founder/author identity must appear consistently across all sources. Inconsistent naming ("BrandCited," "Brandcited," "Brand Cited") fragments the parametric entity and reduces citation confidence.
LLM Pulse's ChatGPT citation guide confirms: there is no submission form to ChatGPT and no "Add URL" tool. Citations are tied to specific URLs and content blocks, not brand name frequency within a single page.
Why Perplexity's decline should change your platform priority#
Perplexity's traffic decline from 19.7% to 6.85% of U.S. AI referral share does not make Perplexity irrelevant. It does change where it belongs in a brand's optimization priority queue.
Perplexity still has strong retention in specific segments: developers, researchers, and technical B2B buyers who value source attribution and transparency. FATJOE's Perplexity stats compilation notes Perplexity's monthly active user base remains in the tens of millions. The argument for optimizing for Perplexity is less about overall traffic share and more about category fit.
The sharper strategic shift from the Previsible data is toward Gemini. Gemini's 3.2x growth makes it the second-most important platform for brands to track, and its citation behavior differs from ChatGPT. Gemini's retrieval layer is grounded more heavily in Google's Search index than Bing's, which means Gemini citations respond to Google Search ranking improvements faster than they respond to schema-only fixes.
The practical priority ranking for most brands in 2026:
1Google AI Overviews (Gemini-powered, grounded in Google's index, highest total AI-influenced traffic)
2ChatGPT (92.4% of standalone AI referral traffic, Bing-grounded for browsed answers)
BrandCited's audit runs structured probes against ChatGPT in both standard-mode and reasoning-mode variants to surface four specific gaps that the Previsible data shows most brands are missing.
The first gap is Bing index presence. If ChatGPT's web-lookup mode cannot find your brand on Bing, the 40% of queries that trigger live search will not return your brand. BrandCited's audit flags brands absent from Bing's top 20 for their primary target queries as a critical finding.
The second gap is parametric recall. BrandCited probes ChatGPT with non-browsing queries on category topics to measure whether a brand appears from memory. Brands scoring below 15 on this component typically have thin cross-domain entity coverage: few distinct domains mention them, no Wikipedia entry, or inconsistent naming across sources.
The third gap is schema completeness. BrandCited checks for Article, FAQPage, and Organization schema on primary pages and flags absent or malformed JSON-LD blocks.
The fourth gap is content structure. BrandCited's content audit flags sections that open with preamble rather than a direct answer, since ChatGPT's extraction system favors answer-first structures.
Run a free audit at brandcited.ai to see your ChatGPT score component alongside scores for Perplexity, Gemini, Claude, Copilot, and the others. Results take 30 seconds.
Anthropic (July 23): Claude voice mode now lets users choose between Opus, Sonnet, and Haiku models during voice conversations, defaulting to the fastest version of whatever model was last used in text mode. (TechCrunch)
Perplexity (July 2026): Perplexity Computer is now inside Microsoft 365 apps (Word, Excel, Outlook, Teams, and PowerPoint), bringing AI-assisted workflows to enterprise users without leaving their existing tools. (Releasebot)
Google AI Overviews (ongoing): AI Overviews now appear on 48% of all Google searches, with a position-1 click-through rate of 2.6% when an AI Overview is present, according to tracking data compiled by TheStacc. Zero-click searches have reached 60% overall.
GEO market: At least eight platforms now offer AI answer-engine visibility monitoring as a defined software category, according to Market Scale. GEO has moved from a specialist tactic to a measurable discipline with its own analytics stack.
Six actions, ranked from highest to lowest impact on ChatGPT citation frequency:
1Allow GPTBot in robots.txt and publish llms.txt: Check your robots.txt to confirm GPTBot and OAI-SearchBot are not blocked. Add a /llms.txt file at your domain root listing your key pages and their topics. Evidence for llms.txt's direct impact is limited in 2026, but unblocking GPTBot is confirmed to affect ChatGPT's browse-mode coverage.
1Add Article, FAQPage, and Organization schema to your top 20 pages: Schema is the highest-confidence signal ChatGPT's extraction layer uses in browse mode. Missing FAQPage schema on a page with FAQ content is the single most common correctable gap in BrandCited's audit data.
1Rewrite section openers to lead with the direct answer: Audit every H2 section on your primary pages. If the opening sentence does not answer the question the heading asks, rewrite it so it does. ChatGPT's extraction system lifts the first two to three sentences of a passage more than any other position.
1Build your Wikipedia and Crunchbase entity presence: If your brand does not have a Wikipedia article or a complete Crunchbase profile, parametric ChatGPT responses cannot confidently attribute claims to you by name. A Crunchbase profile takes 20 minutes. A Wikipedia article requires independent notability from existing press coverage.
1Prioritize Bing ranking for your top 10 commercial queries: Since ChatGPT's browse mode pulls from Bing's index, Bing ranking position directly affects whether your brand appears in the 40% of ChatGPT answers that involve live web lookup. Submit your sitemap to Bing Webmaster Tools if you have not already.
1Run separate prompts in ChatGPT's Thinking mode for high-value queries: Standard-mode and Thinking-mode answers come from overlapping but distinct source pools (only 25.6% shared). The buyer using Thinking mode for a purchase decision is the highest-intent user in your funnel. Test whether your brand appears in both modes and diagnose gaps separately.
What share of AI referral traffic does ChatGPT send?
ChatGPT drives 92.4% of all standalone AI referral traffic as of May 2026, according to Previsible's 2026 State of AI Discovery Report covering 6.77 million LLM-driven sessions across 166 websites. Monthly LLM-driven sessions grew from 65,249 in November 2024 to 644,478 in May 2026.
How does ChatGPT's Thinking mode change which brands get cited?
Thinking mode and standard mode share only 25.6% of cited sources, based on a Semrush and Kevin Indig study of 200 responses across 20 buyer journeys. Citation rates rise from 50% to 68% in Thinking mode. Official documentation and support pages gain citation share; Reddit and user-generated content lose it. Brands optimizing only for standard-mode answers may be invisible at the highest-intent points in the buyer journey.
Why has Perplexity's traffic share dropped?
Perplexity's share of U.S. AI referral traffic fell from 19.7% to 6.85%, down 61% from its March 2025 peak, according to Previsible's 2026 data. Gemini's 3.2x growth in the same period contributed to the relative decline. Perplexity retains strong usage among technical and research-focused users, but its share of consumer and commercial brand-discovery queries has shrunk.
Should my brand publish an llms.txt file?
Publish it, but do not expect it to be the primary lever. In 2026, GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot rarely fetch /llms.txt. They crawl HTML pages. The file costs nothing to publish and is future-proof for enterprise RAG and smaller LLM-search systems. The higher-impact fixes are schema markup, answer-first content structure, and Bing index presence.
How does BrandCited measure ChatGPT visibility?
BrandCited runs structured probes against ChatGPT in both standard and reasoning modes every 30 days, tracking citation frequency, citation prominence, and URL attribution per engine. The ChatGPT component of BrandCited's 0-100 composite score uses both browsing-enabled and non-browsing probes to capture parametric recall and live-search answer coverage separately. Run a free scan at brandcited.ai.
Is Google AI Overviews more important than ChatGPT for brand visibility?
Yes, by volume. Google's AI Overviews and AI Mode drive more AI-influenced discovery than all standalone LLM assistants combined, according to Previsible's 2026 report. But the two surfaces require different optimization approaches. AI Overviews are grounded in Google's Search index; ChatGPT's browse mode uses Bing. A brand that wins one does not automatically win the other. BrandCited tracks both.
ChatGPT's 92.4% share of AI referral traffic is the clearest signal that brand AI visibility is not a "spread evenly across all platforms" game. It is a prioritization problem, and right now most brands are solving the wrong half of it first. Fix the ChatGPT gaps, build the Bing and schema foundations that carry over to AI Overviews, and then fill in Gemini, Perplexity, and the others.
Run a free AI visibility audit on your brand at brandcited.ai. You will see your score across 9 AI platforms in 30 seconds, with every gap ranked by impact.
