Semrush has published one of the most detailed analyses of ChatGPT referral traffic to date — over 1 billion lines of U.S. clickstream data, tracked across 17 months, covering a 200 million-user panel. The headline number is impressive: referral traffic from ChatGPT grew 206% between January 2025 and January 2026. The number that matters more is the one buried beneath it.
The concentration problem
Of all ChatGPT referral traffic, 21.6% goes to a single destination: Google. The next nine domains bring the total to just over 30%. Which means the top 10 destinations on the internet capture nearly a third of all traffic that ChatGPT sends anywhere — and the remaining 160,000-plus domains receiving referrals share the rest in a long tail of minimal, often unmeasurable impact.
This is not a growth story. It is a concentration story. ChatGPT referral traffic is growing fast, but it is accreting overwhelmingly to brands that are already authoritative, already cited, and already embedded in how the model represents categories and topics. The brands that were invisible to ChatGPT in January 2025 are, in most cases, still invisible in January 2026 — just invisible against a larger absolute number.
The study also documents something structurally significant: the number of domains receiving any referrals at all peaked at around 260,000 in 2025 before settling near 170,000. The ecosystem of ChatGPT referral recipients shrank by roughly a third — while total referral volume grew by 206%. More traffic going to fewer destinations. The concentration is not a temporary pattern. It is the direction of travel.
Why citation compounds
The mechanism behind this concentration is not arbitrary. ChatGPT defaults to pre-trained knowledge for most queries, using web search only in specific cases — when users explicitly request sources, or when questions involve recent events. This means that for the vast majority of queries, what ChatGPT says about a brand is determined by what it learned during training, not by what it finds in real time.
Pre-trained knowledge reflects the cumulative weight of authoritative sources over time. Brands that have been consistently cited, accurately described, and structurally coherent across those sources appear in training data with sufficient density to be reliably retrieved. Brands that have not — regardless of their actual market position, product quality, or SEO rankings — are absent or weakly represented.
And because training is cumulative, the gap compounds. A brand that establishes citability now builds a position in training data that subsequent model updates will reinforce. A brand that waits for the channel to mature before investing will find itself competing against the accumulated citation authority of brands that started earlier — not just their current output.
The conversion signal changes the stakes
A separate analysis by Visibility Labs, covering 94 ecommerce brands across 12 months of GA4 data, adds a dimension that makes the concentration problem more urgent. ChatGPT referral traffic converted at 1.81% — 31% higher than non-branded organic search at 1.39%. Revenue per session was $3.65 for ChatGPT referrals versus $3.30 for organic.
The interpretation from Visibility Labs is that users refine their intent inside ChatGPT before clicking. By the time they reach a product or service page, they have already done the evaluative work that traditional search requires them to do on the site itself. They arrive closer to a decision.
If this pattern holds across categories — and there is no obvious reason it would not — then a ChatGPT citation is not just a visibility event. It is a pre-qualified referral. The user has been told about your brand, has absorbed some representation of what you offer, and has chosen to click through. The traffic is smaller in volume and higher in intent. Which means the cost of not being cited is not just lost impressions — it is lost conversions that would have been more valuable than average.
What the data says about citability
The Semrush study identifies something the referral numbers alone do not show: ChatGPT uses web search selectively, not by default. The search-triggered query rate has been declining. This means that an increasing share of ChatGPT responses about brands, products, and services is being generated from pre-trained knowledge — and an increasing share of referral traffic is going to brands that are represented in that knowledge, not brands that happen to rank well in live web search.
This is the structural argument for citability as a distinct discipline from SEO. A brand can rank in position one for its primary category keywords and still be invisible in ChatGPT if its entity definition is inconsistent, its topic authority is weak in pre-trained data, or its cross-platform representation is fragmented. Conversely, a brand with strong citability — consistent, structured, authoritative — can earn ChatGPT referrals even for queries where it does not rank at all in traditional search.
The Citability Score framework measures this readiness across 56 parameters: identity and structured data, reputation and external network, content quality, semantic consistency, and technical performance. Most brands that have not been audited score between 30 and 45 out of 100. The brands in the top decile of ChatGPT referral traffic almost certainly score significantly higher — not because they optimised for ChatGPT specifically, but because the signals that produce strong citability are the same signals that produce strong training data representation.
The window before normalisation
206% growth in referral traffic over 12 months is a channel expanding rapidly. But the concentration data tells us that this growth is not being distributed evenly — it is being captured by brands that are already in the model's mental map of their categories. For every brand in the top 10 destinations receiving a disproportionate share of referrals, there are tens of thousands of brands in the long tail receiving nothing measurable.
The window to establish citability before this channel normalises is real, and it is narrowing. As more brands invest in structured content, entity consistency, and authoritative cross-platform presence, the baseline for what counts as citable will rise. The brands investing now are setting that baseline. The ones waiting for the channel to prove itself before committing will be chasing a moving target set by earlier movers.
ChatGPT referral traffic grew 206% in a year. For most brands, it grew zero percent — because zero times any multiplier is still zero. The compounding is not in the growth rate. It is in the citation position you hold when the growth arrives.
Source: Search Engine Land — One in five ChatGPT clicks go to Google: Study. Data from Semrush analysis of 1 billion+ lines of U.S. clickstream data, October 2024 – February 2026.