Search/No. 25/6 min read
Six in ten AI referrals land on your homepage. That’s the page you never instrumented.
AI assistants send buyers to your front door, not your landing pages. If the homepage captures no source, your highest-intent traffic arrives untracked and converts worst.
Your highest-intent traffic is landing on the least instrumented page you own. AI assistants don’t send people to the blog post that answered the question. They send people to the brand’s front door, and roughly six in ten ChatGPT referral visits now land on a homepage rather than a specific page, according to Similarweb’s 2026 AI search data. Your homepage has no UTM, no campaign association, and usually a “Contact us” form with no source field on it.
So the visit that took the longest to earn shows up as Direct, converts on your worst form, and gets attributed to nothing.
AI traffic doesn’t behave like search traffic, and your stack assumes it does
A visitor from organic search arrives mid-question. They clicked a result because that page promised an answer, so they land deep, on a post or a product page you built and tagged on purpose.
A visitor from an AI assistant arrives after the question is already settled. The assistant did the comparing. It named you. The user clicked your brand name inside the answer, which means the click goes to the domain, not to the page that earned the mention. Similarweb’s data shows this shifted hard in May 2026, when ChatGPT started surfacing brand names as clickable callouts inside answers. Homepage referrals went from a minority of ChatGPT referral traffic to roughly six in ten, and stayed there. That’s a change in shape, not a spike.
Two things break at once when that happens.
First, the referrer. A large share of AI-assisted sessions arrive with no referrer header at all, because the click came out of an in-app browser or a native app rather than a web page. No referrer means no source, and no source means the session gets filed under Direct, next to bookmarks and typed URLs. The channel that just produced your most qualified visitor of the week is indistinguishable from someone typing your domain from memory.
Second, the landing page. Your homepage was built to introduce the company. It was not built to capture a source, qualify a visitor, or route anything. Nobody wrote a UTM for it, because you can’t tag a link you didn’t place.
The fix is three moves, and none of them are a new platform
1. Build the AI assistants channel in GA4 yourself
GA4 will not do this for you by default. The AI referrals it can see get lumped into Referrals, mixed in with everything else. You have to define the channel.
Google documents the pattern in its custom channel groups guide, including a worked “AI assistants” example: create a channel, match the assistant domains with a regex, then reorder the channel list so AI assistants sits above Referrals. That ordering step is the one teams skip, and it’s the one that matters. Traffic is assigned to the first channel whose definition it matches, so if Referrals comes first, your new channel stays empty and you conclude AI isn’t sending you anything.
Do this before you argue about whether AI traffic is worth chasing. You can’t have that argument without the number.
2. Stamp the source on the record before the CRM overwrites it
Channel reporting in GA4 tells you about sessions. It tells you nothing about pipeline. To connect an AI referral to a deal, the source has to survive onto the contact record and stay there.
HubSpot gives you most of the frame already. Original Traffic Source (hs_analytics_source) holds the first known source for a contact and doesn’t change, while Latest Traffic Source updates on every subsequent visit. HubSpot also classifies assistant traffic natively now: its traffic source documentation lists AI Referrals as a source value, with the platform domain (chatgpt.com, claude.ai) in drill-down 1.
Two gaps are left, and they’re the expensive ones. That native classification depends on a referrer surviving the click, so the no-referrer sessions still land in Direct traffic with nothing but an entrance URL. And if your reporting reads Latest Source instead of Original, an AI-discovered buyer who returns through a branded search two weeks later gets recorded as organic, and the assistant that created the demand disappears from the record.
The build, concretely:
- Capture the landing page and document referrer in session storage on first hit, before any internal navigation clears them.
- Write both into hidden fields on every form, including the generic homepage contact form. That form is a primary acquisition path now, so treat it like one.
- Map those hidden fields into a read-only first-touch property on the contact, then roll it to the deal so it survives to closed-won.
- Leave the native source properties alone as your system-of-record baseline, and report against your first-touch field when the native value says Direct.
This is the same discipline behind fixing a broken UTM taxonomy. The difference is that with AI referrals you can’t tag the link at the source, so the capture has to happen on arrival instead. Same problem, one layer further down.
3. Treat the homepage as an acquisition page, because it is one now
If six in ten AI referrals hit the homepage, the homepage is a landing page whether you designed it as one or not. Most B2B homepages fail that job the same three ways: the primary action is a demo request aimed at someone who is still deciding, the proof lives three clicks deep, and the form asks for nine fields and captures no source.
Fix the page to match the visitor. Someone arriving from an AI answer already knows roughly what you do, because a machine just explained it to them. What they need is verification and a low-friction next step. Put the specifics on the page: real numbers, named systems, actual proof. That’s the same structural principle behind why AEO isn’t a new channel, it’s the structure your pages should have had already. The pages that get cited and the pages that convert AI referrals reward the same thing: being concrete in public.
What it costs you to leave this alone
The reporting failure compounds quietly. Your Direct bucket grows, and everyone treats it as noise. AI-sourced pipeline gets credited to branded search or to nothing at all. Budget keeps flowing toward the channels that can prove themselves, which are the ones with clean tagging, not the ones producing the best buyers. And when someone finally asks what AI search is worth to the business, the honest answer is that you can’t tell, which gets rounded down to zero.
Fixed, the picture is boring in the best way. You can see an AI assistants channel in GA4. You can filter contacts by a first-touch source that didn’t get overwritten. You can put a pipeline number next to it. Then the question of whether to invest in being cited becomes a math problem instead of a debate.
This sits at the seam between two things we build. The capture layer and the page itself are Websites & Search work. The property structure that carries the source from contact to deal to closed-won is Marketing Operations & CRM work. Doing one without the other gives you a tracked visit you can’t tie to revenue, or a revenue report with a hole where the source should be.
Where to start
Open GA4 and look at your Direct traffic trend over the last twelve months. If Direct is growing faster than your brand is, some of that is AI referral traffic arriving without a referrer, and you’re one custom channel group away from seeing it.
Then open your homepage form and check whether it captures a source at all. Most don’t. That’s a one-afternoon fix that changes what you can prove for the next two years.
Not sure how much of your Direct bucket is actually AI discovery? That’s one of the things we map in the free 30-minute audit. We look at how traffic arrives, what your forms capture, and where the source dies between the website and the CRM, then hand you the prioritized list of fixes. Whether we work together or not.