On 7 May 2026, ChatGPT changed how it points readers to brands: it moved from footnote-style citation chips to inline, clickable brand links inside the answer, and it started including a brand URL far more often. The measurable result was immediate. Referral traffic to brand websites, and to homepages in particular, jumped sharply the same day and stayed elevated. As of mid-2026, two independent datasets agree on the shape of the shift.
This piece collects the evidence. Every figure attributed and dated, and explains what the change does and does not mean. It is the data companion to the practical playbook on optimising your homepage for AI referrals.
What changed on 7 May 2026?
ChatGPT switched from showing sources as footnote-style chips to embedding inline hyperlinks on brand names and terms within the answer text, and it began surfacing a clickable brand URL in a far larger share of answers. Practitioners have called it the "branded-link update."
The clearest single measure comes from Profound (an AI-visibility analytics vendor), which reported that the share of ChatGPT answers containing a brand URL rose from about 4.5% to roughly 20-24% on 7 May 2026: about a 5× increase in one day. The effect was structural, not a blip: the elevated rate held in the weeks after. Because this is one vendor's tracking of the sites it monitors, treat it as directional and dated rather than an audited industry constant.
How much did referral traffic change?
A lot, and two datasets measured with different methods point the same way. The table below keeps each figure tied to its source.
| Metric | Figure | Source |
|---|---|---|
| Answers containing a brand URL | ~4.5% → ~20-24% on 7 May (~5×) | Profound (2026) |
| OpenAI referrals to monitored brand sites | ~60-65% lift overnight, held | Profound (2026) |
| Total ChatGPT referrals, week-on-week | +157.7% (30 Apr-6 May vs 7-13 May) | Similarweb desktop panel (2026) |
| Homepage referrals, week-on-week | +354.7% (same window) | Similarweb desktop panel (2026) |
| Page views per ChatGPT visit | 3.8 → 4.7 (+24%) | Similarweb desktop panel (2026) |
| Time on site per ChatGPT visit | 3.5 → 3.9 min (+11%) | Similarweb desktop panel (2026) |
Source: Profound (server-side referral tracking across a basket of monitored brand sites) and Similarweb (desktop clickstream panel, 30 Apr-20 May 2026). Both are single-source datasets with undisclosed or panel-limited samples, so the direction (a large, sustained jump in ChatGPT referrals, concentrated on homepages) is firmer than any single percentage. The fact that two independent methodologies agree on that direction is what makes the finding credible.
Why did homepage traffic surge specifically?
Because the inline link ChatGPT now adds most often is a brand's homepage, not a deep product or article page. When an answer names a company and makes that name clickable, the natural target is the brand's root URL.
The homepage share of ChatGPT referrals shifted dramatically:
- Similarweb: homepages went from roughly 26-32% to about 60% of ChatGPT referrals immediately after 7 May.
- Profound: on its own site, homepage share of OpenAI referrals moved from about 28% to 62%; industry-wide across its monitored basket, homepages went from about 4% to 24% of OpenAI referral traffic.
The two homepage figures from Profound measure different scopes (one site vs a broad basket), which is why they differ. Both show the same direction. The practical read: the homepage has become the page AI most often sends a clicker to, so it now carries a discovery job it did not have a year ago.
The branded-link shift turned ChatGPT into a referral channel that lands on your front door. Your homepage is now the page an AI answer hands a curious buyer.
Did every industry benefit equally?
No. Profound's breakdown showed the lift concentrated in some sectors and absent in others:
| Sector | Daily OpenAI referral change vs pre-7-May baseline |
|---|---|
| B2B Software / SaaS | Up more than 200% |
| Financial Services / Fintech | ~60% lift |
| E-commerce / Retail | Essentially flat |
Source: Profound (2026), single-vendor. The pattern is consistent with how people use ChatGPT differently by category: research- and consideration-heavy sectors (SaaS, fintech) generate answers that name and link a brand, while transactional retail queries more often route through product surfaces and shopping features than through an inline brand link. Do not assume the SaaS-scale lift transfers to a retail catalogue.
Does more referral traffic mean more recommendations?
No, and conflating the two is the trap. The branded-link change affects how often a clickable link appears and where it lands, not whether your brand is the one named as the answer. This is the difference between being cited versus recommended in AI search: you can pick up more referral clicks while a competitor is still the brand the engine actually recommends.
It also does not mean ChatGPT is the whole picture. Each engine is its own citation ecosystem. Being linked more by ChatGPT says nothing about Gemini, Claude, or Perplexity, which is why one brand gets cited and another doesn't varies by engine. And a sizeable share of AI-driven visits still arrive with no referrer, landing in analytics as Direct, so even this surge is undercounted unless you unmask AI traffic in GA4.
What should you do about it?
Three durable moves, none of which depend on the exact 7 May numbers holding:
- Treat your homepage as an AI landing page. It is now the page ChatGPT most often links, so it must state who you are and what you do in plain, extractable text, and route visitors onward. The step-by-step is in how to optimise your homepage for AI referrals.
- Measure the channel honestly. Build an AI-referral view in GA4 and track the share that lands on the homepage. Remember the scoreboard has moved from clicks to citations. Referral clicks are one signal, answer-level presence is the other.
- Earn the recommendation, not just the link. Strengthen the entity signals and corroboration that make an engine name you, and follow the playbook on how to get recommended in ChatGPT. A link you didn't earn the recommendation behind converts poorly.
The honest caveat: these are mid-2026 figures from two vendors, and AI answer formats change fast. OpenAI could adjust link behaviour again. Anchor on the durable shift (the homepage is now an AI referral destination) and date the volatile specifics.
Knowing whether this change is sending you traffic, and whether those answers actually name your brand. Is exactly what Buffy Intel tracks: your presence, citations, and AI-referred traffic across every major engine, measured over time rather than spot-checked.