ChatGPT started linking out. The biggest winner was your homepage.
On 7 May 2026, ChatGPT began showing clickable links instead of only naming websites. Referral traffic nearly doubled. But the gain did not land where most teams expected, and the number everyone is quoting is not the one that describes a typical site.
Key takeaways
- On 7 May 2026, ChatGPT began displaying clickable source links rather than only mentioning site names.
- Across 52 sites, referral sessions rose 97.9%, from 183,497 to 363,089, in matched six-week windows either side of the change.
- Homepages grew 1,316%, from 7,990 to 113,183 sessions, lifting their share of ChatGPT referral traffic from 4.4% to 31.2%.
- Blog sessions fell 28% over the same period, from 3,495 to 2,509.
- The median site grew 54%, not 98%. One site took 64% of the net increase, and 14 of the 52 declined.
- The strategic read: AI search rewards being nameable, not being comprehensive. Your homepage is now an AI landing page.
What changed on 7 May
Before the change, ChatGPT would frequently tell you about a business without giving you a way to reach it. The model named the source in prose, and that was the end of it. You could be the answer to someone's question and receive nothing at all: no click, no session, no line in your analytics.
On 7 May 2026 that behaviour changed. ChatGPT began rendering clickable links to the sources behind its answers. The mention became a route.
This is a bigger shift than it sounds. It changed the economic value of being cited. Before, a citation was a branding event you could not measure. After, it is a traffic source with a referrer.
What the data shows
The clearest measurement so far comes from an analysis by Botpresso covering 52 websites, 9 B2B and 43 B2C, comparing two matched six-week windows: 26 March to 6 May against 7 May to 17 June 2026.
| Page type | Before | After | Change |
|---|---|---|---|
| All sessions | 183,497 | 363,089 | +97.9% |
| Homepages | 7,990 | 113,183 | +1,316% |
| Landing pages | 130,104 | 208,255 | +60% |
| Blog pages | 3,495 | 2,509 | -28% |
Two figures in that table deserve more attention than they have had. Homepages went from a rounding error to nearly a third of all ChatGPT referral traffic, moving from 4.4% share to 31.2%. And blog traffic went backwards while everything else grew.
Why the homepage won
This is the part worth understanding properly, because it reveals how AI search differs from Google at a structural level.
Google matches a query to a page. Someone searches for a specific thing, and the best-matching document wins. That is why deep content works in traditional SEO: a long, specific article can outrank a homepage because it answers the exact question asked.
ChatGPT does something different. It answers a question by naming a brand. When a model says "you could look at Acme for this", the natural link target is not a blog post about the topic. It is Acme. The link goes to the entity being recommended, which in practice means the homepage.
In other words, Google returns documents and AI returns recommendations. Documents link deep. Recommendations link to the front door.
The consequence is that your homepage now receives cold traffic from people who have just been recommended you by a machine they trust. That is a completely different visitor from the one most homepages were designed for, who usually arrived already knowing the brand or via a specific campaign.
The number everyone is quoting is the wrong one
The 97.9% figure has travelled widely. It is accurate, and it is also misleading if you read it as "AI referral traffic roughly doubled for everyone".
The same study reports a median growth of 54%. The gap between the mean and the median tells you the distribution is heavily skewed, and the site-level numbers confirm it: the single biggest winner accounted for 64% of the entire net increase, and the top five accounted for 91%.
Of the 52 sites, 36 gained, 2 were flat, and 14 declined. More than a quarter of the sample got less ChatGPT traffic after the update than before it.
So this was not a rising tide. It was a redistribution, and a concentrated one. Presence in the set of sources a model draws on is winner takes most, which is precisely why knowing whether you are in that set matters more than knowing your average position in anything.
What this means for content
The instinctive response to AI search has been to publish more: more articles, more depth, more coverage, on the theory that more surface area means more chances to be cited. The blog figure complicates that theory. Blog sessions fell 28% in the same window that total sessions nearly doubled.
Read that carefully, because the wrong conclusion is easy to reach. It does not mean blog content is worthless for AI visibility. Editorial content is a large part of how a model comes to understand what you do and decides you are worth naming at all. What the data suggests is that blog content is doing its work upstream, shaping the model's understanding, rather than serving as the click destination.
The practical translation: being nameable beats being comprehensive. A model has to be able to say, in one clause, what you are and who you are for. If it cannot, it will name a competitor who is easier to describe, no matter how much you have published.
What to do about it
Four things, in the order they pay off.
1. Treat your homepage as an AI landing page. Assume the visitor arrived with no context beyond a machine recommending you. Can they tell within one screen what you do, who it is for, where you operate, and what to do next? Most homepages assume prior knowledge that this visitor does not have.
2. Make your brand nameable. One consistent description of what you are, used everywhere. Ambiguity is the enemy: if your positioning takes three sentences to explain, you will lose the recommendation to a competitor whose takes one.
3. Check whether you are linked or only mentioned. These are now different outcomes with different value. Ask the questions your buyers ask and look at whether your name appears as plain text or as a link, and on which engines.
4. Measure it continuously. AI answers vary between runs, so a single check is a snapshot rather than a trend. Our guide to tracking brand mentions in ChatGPT covers doing this by hand, and how AI chooses which sources to cite explains what moves the outcome.
Questions and answers
What exactly did ChatGPT change on 7 May 2026?
It began displaying clickable links to the sources behind its answers, rather than only naming websites in the text. The practical effect is that a citation can now produce a visit, where previously it often produced only an unmeasurable brand impression.
Did every site gain traffic from the change?
No. In the 52-site sample, 36 gained, 2 were flat and 14 declined. The aggregate rise of 97.9% is heavily concentrated: one site accounted for 64% of the net increase and the top five for 91%. The median site grew 54%, which is a more representative figure than the aggregate.
Why did homepages benefit so much more than blog pages?
Because AI answers name brands rather than returning documents. When a model recommends a business, the natural link target is the business itself rather than an article about the topic. Google matches queries to pages, so deep content wins there. AI makes recommendations, so the entity wins.
Should I stop publishing blog content for AI search?
No. Editorial content is a large part of how a model learns what you do and whether you are worth naming. The data suggests it works upstream, shaping understanding, rather than acting as the click destination. Keep publishing, but do not expect blog pages to be where AI referral traffic lands.
Sources
- Botpresso, "98% Growth in ChatGPT Referral Traffic After the May 2026 Links Update." Analysis of 52 websites across matched six-week windows. 2026. botpresso.com
- SE Ranking, "Referral Traffic from ChatGPT Hit an All-Time High in May 2026." 2026. seranking.com
- SearchScore, "SearchScore methodology and scoring model." 2026. searchscore.io/methodology
Is your homepage ready for AI traffic?
A free SearchScore audit grades 250+ signals across AI search, SEO and conversion, including whether your homepage tells an AI engine clearly enough what you are and who you are for.
Run a free audit →