Referral traffic from ChatGPT to brand websites rose roughly 150% in the week after May 7, when OpenAI began surfacing more prominent inline links to brand homepages within its answers. That is the finding from web analytics firm Similarweb, reported by Search Engine Roundtable on May 27. The product change was quiet; the traffic signal was not.
The scale of the shift is worth pausing on. The share of ChatGPT answers carrying any link at all jumped from approximately 0.4% to 6.2% in a single day, a roughly 14x increase by Similarweb’s count. For most of 2024 and into 2025, the dominant practitioner view of ChatGPT was that it functioned as a zero-click answer engine: capable of satisfying user intent without dispatching anyone to an external site. That framing is now outdated for brands that appear in ChatGPT answers, at least while this link-surface behavior holds.
The composition of where that traffic lands deserves scrutiny. Similarweb reports that around 60% of the new referral visits arrive on brand homepages, not product pages, category pages, or editorial content. That pattern tells a specific story: OpenAI is linking to entity roots, not deep URLs. For SEOs, this means the referral is brand-level, not keyword-level. A brand homepage visit is harder to attribute to a specific query or answer surface, and it behaves differently from organic traffic that lands users mid-funnel. The engagement data offers some reassurance: pageviews per visit were up 24% and time on site up 11% compared to the pre-May 7 baseline, suggesting the referral audience is not immediately bouncing.
Sector variation is steep. Similarweb and additional third-party reporting indicate that B2B software and SaaS categories saw daily OpenAI referrals rise more than 200% from their pre-May 7 level. Financial services and fintech saw a 60% lift. The gap suggests that ChatGPT’s link behavior is not uniform across verticals; some categories are generating far more linked answers than others, likely reflecting query-type differences and the availability of brand entities in OpenAI’s training corpus and real-time retrieval layer.
That last point is significant for how SEOs should think about the underlying mechanism. A brand that appears in ChatGPT answers in the first place has likely accumulated enough web presence, structured data, and brand mentions across authoritative sources to surface as a recognized entity in the model’s output. The May 7 change did not alter which brands appear in answers; it altered whether a link accompanies the mention. This means the referral opportunity is gated upstream by entity recognition, not by the same signals that drive a blue-link ranking.
Practitioners who have built GEO (generative engine optimization) programs oriented around brand mentions in LLM training corpora now have a direct traffic signal to validate their work. Brand mentions that previously produced zero referrals may now produce measurable homepage visits. The causal chain from brand entity presence to referral traffic is shorter and more observable than it has ever been on a generative platform.
Instrumentation is the immediate problem. Many analytics stacks are not currently set up to attribute ChatGPT as a named referral source, segment AI-origin traffic in GA4, or connect a referral session back to the ChatGPT query that produced it. OpenAI does not expose query-level data to sites receiving referrals, which means teams must work from what is observable: UTM parameters on homepage URLs if the referral carries them, referrer strings in session logs, and GA4 custom segments filtering for AI-origin sources. Without this infrastructure in place now, the 150% lift Similarweb is measuring will show up in an analytics dashboard as unattributed direct traffic.
The window for clean baseline data is narrowing. The May 7 behavior change is three weeks old. Teams that build measurement infrastructure today can establish a pre-optimization baseline against which future brand-presence and entity-building efforts can be compared. Teams that wait will be trying to reconstruct that baseline from noisy historical data.
Set up a GA4 segment for AI referral sources, tag your brand homepage with a UTM parameter where the referral chain allows it, and pull a week of referrer-string logs to confirm whether ChatGPT traffic is already hitting your site uncredited.
Reported by Barry Schwartz at Search Engine Roundtable on May 27, 2026, citing Similarweb traffic data published on X following OpenAI’s May 7 link-surface product change.
