Table of Contents
TL;DR
- After ChatGPT’s May 7 2026 Branded Link Update, Similarweb measured homepage referrals from ChatGPT rising 354.7% week-on-week, and the homepage share of OpenAI referrals jumped from 26-32% before to about 60% after.
- An AI-sent visitor arrives with the chat answer still on screen. They are not browsing. They are checking whether the brand the AI just named is real.
- Three lines do most of the work on the first screen. A clean one-line definition of what the business does. A short proof line in the buyer’s vocabulary. One specific next step the AI did not already cover.
- The rest of the homepage matters less than usual. The AI has already done the persuasion. The homepage’s job is to confirm the answer the visitor already heard.
- The change does not undo good homepage discipline. It does raise the cost of a vague hero and lower the value of a long brand vignette before the first scroll.
Homepage referrals from ChatGPT rose 354.7% week-on-week in May 2026. That single number is the reason this article exists.
For most of the last twelve years, the homepage was the page nobody read. Buyers entered the site through a blog post, a product page, an ad landing page, or a search-result deep link. The homepage existed for navigation and for trust signalling.
In May 2026, Similarweb measured the homepage share of OpenAI referrals rising from roughly 26-32% before the Branded Link Update to about 60% after, while total ChatGPT referrals rose 157.7% in the same week. The homepage stopped being a navigation page and became an arrival page. Average page views per visit rose from 3.8 to 4.7 across the same panel.
So the homepage is back. It is back as something different from what it was, and it is back for a reader who is not the reader it was last written for.
Why did homepages start carrying AI-sent traffic in May 2026?
The mechanism is simple. Before May 7 2026, ChatGPT mostly cited specific source pages in a footnote pane that few readers opened. The traffic that did flow went to those source pages, not to the brand’s front door.
After May 7 2026, ChatGPT started naming brands by name inside the answer, with the brand’s root domain as the clickable link. The reader who wanted to learn more clicked the brand name. The brand name pointed at the homepage.
Profound described the same shift from the brand side: homepages went from a rounding error in the OpenAI referral mix to roughly one in four clicks within a week. The change was structural, not seasonal.
The brands that benefited were the ones whose names the engine chose to put in the answer. They did not run an ad. They did not pay for the placement.
What they did do, in most cases, was write a homepage that the engine could match against the question it was answering. That match is the thing this article is about.
How is an AI-sent visitor different from a Google-sent one?
A Google-sent visitor arrives sceptical. They saw ten search results and clicked one. They expect to compare it against the other nine, and they will leave fast if they cannot place the page.
An AI-sent visitor arrives already half-convinced. They asked a question, read an answer, and the answer named your brand and said one specific thing about you.
They clicked through to check whether what the chat said is true.
So the homepage is no longer the first impression. The chat answer was the first impression. The homepage is the confirmation step.
That changes what the visitor is looking for in the first three seconds. They are not asking "what is this?" They are asking "is this what the chat just told me it was?"
A page that confirms the chat answer cleanly earns the visit. A page that contradicts the chat answer (a vaguer hero, a different promise, an inconsistent tone) breaks the trust the chat had built and sends the visitor back to the chat to ask again.
The upstream voice work that keeps a homepage and a chat answer consistent is the same work that earns the named mention in the first place. The hero that the chat answer leans on is the one written from a real position and a real voice.
What does the AI visitor expect to see on the first screen?
Three things, in this order. A clean one-line definition of what the business does, in plain words. A short proof line in the buyer’s actual vocabulary. One specific next step that is not "browse our site".
The definition matters because the AI chat already gave the visitor one. The homepage either confirms that definition (good) or replaces it (confusing). A homepage that opens with a brand vignette instead of a definition forces the visitor to do the matching work themselves, and most of them will close the tab.
The proof line matters because the AI cited the brand on the strength of one specific claim. The homepage either grounds that claim (with a named outcome, a real number, or a named customer kind) or leaves it floating. A proof line that reads as generic praise tells the visitor the chat may have overstated the case.
The specific next step matters because the visitor did not come here to navigate. They came here to take one action. A homepage that offers ten next steps offers no next step, and the visitor goes back to the chat for a clearer answer.
Which three lines do most of the work on an AI-visited homepage?
The hero line names who the business is for, what they get, and what they otherwise would have done. Twenty words at most.
The supporting line gives one concrete proof. A real outcome, a real number, a real customer kind, in the buyer’s own words. A second twenty words at most.
The CTA line names one next step that matches the chat answer. If the chat said you help small businesses pick a website host, the CTA is "see the host shortlist" or "ask me which host fits your situation". It is not "learn more".
Together those three lines fit on a phone screen. The visitor sees them in three seconds. The chat answer either holds or it does not.
A hero that decorates instead of defines makes the visitor wait for the supporting line. A supporting line that lists features makes them wait for the CTA. A CTA that says "explore" makes them go back to the chat to ask the same question of a different brand.
The simple structure the hero must carry sits in how to write a simple value proposition for a small business, and the move that swaps feature lists for outcomes sits in how to turn features into benefits.
What does this change about the rest of your homepage?
Less than you might think. The standard homepage anatomy still holds. Hero, proof bar, problem section, solution section, social proof, CTA stack, FAQ block.
What changes is the weight each section carries. The hero carries more weight than before because the AI visitor reads it as a confirmation, not as an introduction. The proof bar carries more weight because the AI visitor needs to verify the chat answer fast.
The middle sections carry less weight than before. The AI visitor has already absorbed the equivalent of a feature-and-benefit pitch from the chat. They scroll fast through the middle of the page because they are looking for one specific thing.
The FAQ at the bottom carries more weight than before, because the AI chat itself reads the FAQ section to build its next answer about the brand. A homepage with a sharp FAQ feeds the chat better than a homepage with a soft FAQ does, and that compounds across the next few weeks of citations.
The change is calibration, not rebuild. Most homepages do not need a redesign. They need the hero rewritten, the proof line sharpened, and the CTA narrowed.
Other questions worth answering
What should an owner check after rewriting the hero for an assistant-routed reader?
Open the chat in another tab. Ask it the question a buyer would type before knowing your brand. Read the reply against your hero.
If they disagree, fix the upstream voice work, not the page. The hero either confirms the reply or it loses the trust the chat had built.
How long does the Branded Link referral pattern hold for a brand?
Per Profound’s Branded Link Update analysis from May 2026, daily OpenAI referrals roughly doubled across monitored brands. The lift held at the new level through the following week. So treat it as a baseline to defend, not a short novelty to ride out.
Does an assistant-routed buyer actually purchase more than someone arriving through paid search?
Per Similarweb’s May 2026 measurement, OpenAI-referral arrivals convert at 7.1%, behind only paid search. Time per session rose from 3.5 to 3.9 minutes. So a small lift in arrivals tends to compound into a much larger lift in revenue, because the conversion floor is unusually high.
Which bottom section quietly feeds the next round of brand citations?
The FAQ block at the bottom. Chat assistants read it to build their next reply about you, so a sharp FAQ becomes the substrate the engine paraphrases when someone asks. A vague FAQ feeds the engine vague answers, and vague answers fail to earn the named mention you want next month.
Should you keep the long brand vignette above the fold?
No. The arriving reader has the chat reply still open and wants the page to confirm it within two short sentences. A vignette delays that confirmation and reads as evasion. Keep the vignette below the proof bar, where a scroller can find it after the trust check has already cleared.
How would you rewrite your homepage for the AI visitor?
Open ChatGPT in a separate tab. Ask it the question you wish a buyer would ask before finding you. Read the answer.
Now open your homepage and read the hero against the chat answer. Does your hero confirm what the chat just said? Could the chat have written your hero from the substrate of your existing pages, or did the chat get the brand wrong?
If the chat got it right, the homepage already does most of the work and the rewrite is small. Sharpen the proof line. Narrow the CTA. Move one buyer outcome up from below the fold.
If the chat got it wrong, the rewrite is upstream of the homepage. Read how to get AI to mention your brand by name before touching the page, because the homepage cannot fix what the chat is reading from elsewhere.
If your homepage receives meaningful AI-sent traffic and you cannot tell whether your hero confirms or contradicts the chat answer, you can contact me here. Send the URL and the chat question you would like the AI to send you. I will run the side-by-side read and name the one line on the page that is doing the most damage.
No charge and no follow-up call. The hero that ships after the diagnostic is the one that holds the chat-built trust through to the next click.