AI referrals land on your homepage

AI referral clicks now land mostly on your homepage rather than on the deep page that earned the citation, because assistants have started linking the brand instead of the source URL.

That is an abrupt change, not a gradual drift, and it breaks an assumption sitting underneath most GEO reporting: that the page which gets cited is the page that gets the visit.

#What changed

OpenAI shipped an update on May 7, 2026 that put clickable brand links inline in ChatGPT answers instead of leaving sources tucked into footnotes. Profound measured the share of ChatGPT responses containing a URL rising from roughly 4.5% to about 20 to 24% on the same day, close to a fivefold jump, with OpenAI referrals across the brand sites it monitors nearly doubling overnight.

The composition of that traffic moved at the same time. Similarweb's read of the change put the homepage share of ChatGPT referrals at roughly 25% before the update and about 60% after, with total referrals up 157% overnight and holding at that level since.

Panel data from a different vendor points the same way. SE Ranking, working from 101,574 websites with Google Analytics integrated across January 2025 to May 2026, reported that ChatGPT's worldwide referral share rose from 0.23% in April to 0.32% in May, and that 60% of AI-referred traffic lands on homepages against 17% from organic search. Two independent datasets converging on a homepage share near 60% is about as much corroboration as this field offers.

The volume behind that share kept climbing after May. seoClarity's analysis of referral traffic across ChatGPT, Gemini, Perplexity, Copilot and Claude found total AI referral traffic up 110% month over month in June 2026, ChatGPT up 130% in the same month, and Claude up 424% month over month in July.

#The effect is not uniform

Before you rebuild anything, check whether your vertical actually saw the shift. Profound's breakdown of the same update found B2B software and SaaS gaining more than 200% against baseline, financial services around 60%, and e-commerce and retail essentially flat.

That spread is informative. Brand-level linking helps most where the buying decision is about picking a vendor and least where the user wants a specific product page. If you sell software, this is your change. If you sell items from a catalog, the deep page is still the destination and the homepage numbers below will not move much.

#What it breaks in your reporting

Landing-page reports now understate the pages doing the actual work. A documentation page or comparison post can be the reason a model mentioned you at all, and still show close to zero AI referral traffic of its own, because the click that followed went to the root domain. Judging that page by its own landing-page row will tell you to kill the thing that earned the mention.

The fix is to stop treating citation and arrival as one funnel. They are two, with different units:

• Citation earning happens at the page level. The unit is a URL, and the metric is how often that URL is retrieved, cited, and actually used in an answer.
• Arrival handling happens at the site level, mostly on the homepage. The unit is a session, and the metric is what that session does next.

Reporting on them separately also keeps you honest about attribution generally. A large share of AI clicks still arrives without a usable referrer and gets filed under Direct, which we covered in why your AI traffic still shows as direct. Homepage-heavy landing makes that worse, because a bare root-domain hit with no referrer looks identical to somebody typing your URL from memory.

#Making the homepage do the job

The visitor arriving from an answer is not a cold visitor. They have already read a description of you, written by a model, that you did not control. They arrived to check it. The homepage has to confirm or correct that description quickly.

Put the definitional answer above the fold. One sentence saying what the product is and who it is for, in the same vocabulary the cited page uses. A mismatch between the model's summary and your own headline reads as a mistake to the visitor.

Name the paths instead of hiding them. Somebody who arrived from a comparison answer wants pricing or a competitor page. Somebody who arrived from a how-to answer wants documentation. Both should be one click and one obvious label away, not buried in a mega-menu.

Mark up the homepage as the organization. Google's documentation recommends placing Organization structured data on your home page, or a single page that describes your organization, and notes there are no required properties, so you add the ones that apply. Name, alternateName, description, url, logo and sameAs are the ones that do entity work. This is the page engines will resolve your brand entity against, and it is now also the page they send people to.

Keep the homepage claims consistent with the cited pages. Entity consistency was already a retrieval argument. It is now a conversion argument too, because the visitor is holding a model-written summary in their head and comparing it against what they see.

#What not to do

Do not rewrite deep pages to push readers toward the homepage. The deep page is what gets retrieved. Weakening it to redirect attention costs you the mention that produced the click in the first place.

Do not treat the homepage share as permanent. It moved overnight once because one vendor changed one linking behavior, and it can move back the same way. The e-commerce result from the same update, where nothing meaningful changed, is a reminder that this is a product decision inside ChatGPT rather than a law of AI search.

Do not read a homepage-heavy referral mix as proof that your content strategy stopped mattering. The retrieval layer did not change. Models still read pages, still rank candidates, and still decide which document to draw an answer from. What changed is only where the reader is put down afterwards. Optimize the page for being read by the model, and the homepage for being landed on by the human.