AI referrals land on your homepage. That’s the problem.

Three separate datasets, three different methodologies, one uncomfortable finding: AI systems are citing your deep pages as evidence and then sending the resulting traffic to your homepage. The machine reads deep. It sends shallow.

Slobodan Manic laid the numbers out yesterday and they're worth staring at. Similarweb's 2026 Generative AI Landscape report found 65% of ChatGPT citations point to URLs two or three folders deep. And yet 58.8% of ChatGPT referral traffic lands on homepages. Previsible's analysis of 6.77 million AI-referred sessions across 166 websites adds another wrinkle: 28.8% of ChatGPT referrals land on internal search results pages. Ahrefs reported over 80% of its AI referral traffic arriving at the homepage, product pages, and free tools rather than the editorial library that actually earned the citations.

Three panels. Three methodologies. Same split. The direction is consistent enough that we can stop pretending it's a measurement artefact.

And the depressing part is that this isn't a new problem. It's the oldest mistake in conversion optimisation, wearing a new referrer string.

The ad-to-collection-page mistake, one level up

Anyone who's spent time in paid search knows this failure mode. You run an ad for a specific product. The click lands on a collection page or the homepage. The visitor clicked on a promise and arrived at a haystack. Some do the work of finding the thing they clicked for. Most don't. The campaign bleeds money at the landing step.

Paid search built an entire discipline around not doing this. Message match. Quality Score. Ad-to-landing-page relevance as a ranking factor. Two decades of practitioners hammering the point that specific intent needs specific landing pages.

AI referrals are breaking that rule at scale, and nobody in the SEO industry seems to have noticed yet because we're all still arguing about whether AI Overviews are killing traffic in aggregate. Meanwhile the traffic that does arrive is arriving on the wrong page.

Why this is worse than the paid search version

The AI-referred visitor is not a cold click. Similarweb's downstream analysis found users who received a brand recommendation from ChatGPT were 2.5 times more likely to visit that brand's site in the following seven days. They're not browsing. They're not comparing. The chatbot already did the comparing. They arrived pre-convinced.

The visitor arrives sold. The page tries to sell them.

Which means the homepage — designed as a brochure for cold visitors who need convincing — is being served to warm visitors who need a checkout button. The hero banner is doing the wrong job. The mega menu is asking someone who was just told "this is the answer" to start their search over.

The visitor arrives sold. The page tries to sell them.

That mismatch is worse than the paid search version because the intent gap is bigger. A paid click has intent from a keyword. An AI referral has intent from a full conversational exchange in which your product was named as the recommendation. If the paid search version bled money, the AI version haemorrhages it.

The internal search results problem is even worse

The Previsible finding on internal site search deserves its own paragraph. Nearly 29% of ChatGPT referrals landing on internal search pages means the AI is generating URLs with query strings, and users are being dumped onto search result pages generated on the fly — often for queries that don't match your product catalogue cleanly.

That's not a landing page. That's a purgatory. It's what happens when an AI hallucinates a URL structure that sort of looks right and the site helpfully returns "0 results found" or, worse, a list of vaguely related products the visitor didn't ask for.

In my experience auditing sites for AI-referral behaviour over the last six months, this is the single most fixable problem on the list. Most sites have never looked at their internal search referrer logs because they've never had to. Now they have to. Not for SEO reasons — for revenue reasons.

Why this is happening structurally

ChatGPT's linking behaviour changed this spring when it started surfacing prominent brand links in answers alongside the source citations. There are now two link types in the same response: the source (which is deep) and the brand link (which is the homepage). The Similarweb data shows the homepage share of referrals roughly doubled after that change.

So the citation link and the referral link are increasingly not the same link. Your deep page earned the citation. Your homepage got the visit. And most analytics setups don't distinguish between them cleanly because both arrive with the same referrer string.

That's the loop. And we built it — by treating the homepage as the default destination for every ambiguous link, by not investing in landing pages for AI-mediated traffic, by assuming AI referrals would behave like search referrals.

What this actually means

If you're running a site that's getting any meaningful AI referral traffic, three things become urgent.

First, look at where AI traffic actually lands versus where it's cited. If the citations are pointing to deep evidence pages and the visits are arriving on the homepage, you have a landing page problem, not a citation problem. Fixing citations won't fix the conversion gap. Fixing landing pages will.

Second, the homepage needs a rethink for the visitor who arrives pre-convinced. That doesn't mean stripping the hero. It means adding a path for someone who was just told you are the answer and wants to take the next step immediately. A visible, above-the-fold action for warm visitors. Not a mega menu.

Third, the internal search results issue is a bug, not a feature. If AI systems are generating URLs to your search endpoint, those pages need to convert. Or you need to redirect them to canonical landing pages. Or you need to configure your search UI to handle these arrivals differently. This is not a philosophical question — it's a hygiene fix.

The measurement layer needs to catch up too. Most GA4 and platform-analytics setups still treat AI referrals as a single bucket. Splitting citation-source landings from brand-name landings is the first step to seeing the actual pattern. Until you can see it, you can't fix it.

The uncomfortable read

The reason this matters isn't that AI referrals are a huge percentage of traffic yet — they aren't, for most sites. It's that the intent quality of AI referrals is unusually high, and the conversion gap between what those visitors want and what your homepage offers is unusually wide. That gap represents the biggest single conversion optimisation opportunity most sites have right now, and almost nobody is working on it.

Paid search learned this lesson in 2003. Twenty-three years later, the industry is being asked to learn it again, at a different level of abstraction, before the traffic gets big enough for the losses to become obvious in aggregate reporting. The teams that fix their landing page architecture for AI referrals now will look extremely smart in eighteen months. The ones still debating whether AI Overviews are good or bad for SEO will be wondering where the conversions went.

The machine reads deep and sends shallow. The fix isn't more content. It's a homepage that treats a pre-convinced visitor like a pre-convinced visitor.