The Funnel Went Dark

First, a quick recap of the half-funnel problem

If you've read us before, you know the setup. If you haven't, here it is in a paragraph.

A new category of companies — GEO, AEO, “AI search optimization,” the acronyms keep multiplying — exists to get your brand mentioned when a shopper asks ChatGPT or Perplexity or Gemini what to buy. That work is real and you should do it. AI is becoming the front door of an enormous number of shopping journeys, and whoever the model surfaces wins; whoever it doesn't, effectively doesn't exist for that customer.

But GEO operates on only the first half of the funnel. It gets you mentioned. It gets you the click. Then its job is done — and the customer lands on a site that knows nothing about the high-context conversation they just had. We called this the half-funnel problem: the intelligence stops at the click, and most brands hand a deeply-qualified customer off to a 2015 product page with a color dropdown and a price slider.

That's still true. But underneath it is a second problem that's quietly more dangerous, and almost nobody is pricing it in.

It's not just that your site can't continue the conversation. It's that you can't even see the conversation that's happening.

The funnel went dark.

The Two Numbers That Don't Fit Together

Start with the good news, because it's genuinely good. Traffic that arrives from an AI assistant converts dramatically better than traffic from anywhere else. McKinsey clocks AI-recommended product traffic converting at 4.4x traditional search; some B2B measurements run far higher. The reason is obvious once you say it out loud: a customer who shows up from ChatGPT already had a long, specific conversation about what they need. They arrive with a shortlist and a budget. There's even behavioral fingerprinting to prove it — Ahrefs found AI-referred visitors view about 50% more pages per session and stay longer. They're not browsing. They're closing.

Now the bad news, which is the same fact wearing a different coat. You can't see any of that conversation.

In the old funnel, you saw everything. The search query, the category page, the filters they touched, what they hovered on, where they bounced, what they abandoned in the cart. The entire consideration phase happened on your property, in your analytics, where you could study it and act on it.

In the new funnel, the behavioral data stream doesn't begin until the add-to-cart moment. Everything before that — the discovery, the comparison, the “actually, what about this other one” — happened inside someone else's chat window. By the time the customer reaches you, the most informative part of the journey is over and you weren't invited.

What “Dark” Actually Means

“Attribution is hard” undersells it. Let's be precise about what's disappearing.

Discovery is invisible. When a customer asks an AI “best running shoe for someone with knee pain training for a half marathon,” that query — the richest first-party intent signal imaginable — never touches your servers. The AI hears it. You don't.

Attribution collapses into “direct.” A huge share of AI-influenced purchases never arrives as a trackable referral at all. The customer reads the recommendation, closes the tab, and types your URL directly a day later. Pew Research, studying 68,879 searches, found click-through drops from 15% to 8% the moment an AI summary appears — and only 1% of users click the inline links inside it. In Google's full AI Mode, 92–94% of queries end with no external click at all. The AI did the selling and got none of the credit — which means you can't tell what's working, which means you can't invest behind it. The industry has a name for this now: the dark funnel. It grows every month.

Personalization breaks. You can't tailor an experience to a journey you never observed. The AI knew this shopper was a heel striker on a deadline with a $150 ceiling. Your site knows none of it and greets them with a homepage.

Retail media goes dark, too. This one's bigger than it looks. Retailers spent a decade building retail-media networks on top of their behavioral data. When the behavior moves into the chat, the data those networks run on goes quiet. It's not a coincidence that Target started testing ads inside ChatGPT this year through its Roundel network — a roughly $2B business it can't afford to let go dark. When the consideration phase leaves your property, the advertising business built on watching that phase has to chase it out the door, on terms the platform sets.

Put it together: the most valuable layer of commerce — the part where the customer reveals what they actually want — is migrating into interfaces you don't own, can't see, and increasingly have to pay to appear in. Accenture projects more than 30% of online commerce, about $3.1 trillion, will run through AI agents by 2030. That's the size of the blind spot.

The Uncomfortable Truth About the Upstream Fix

The instinct is to solve a visibility problem by spending more on visibility. Better GEO. More citations. A dashboard that queries the models and reports your “share of voice.”

All of that is worth doing. But here's the thing nobody selling those dashboards wants to dwell on: the ranking they're tracking barely exists.

In late 2025, SparkToro's Rand Fishkin and Gumshoe's Patrick O'Donnell ran the most rigorous public test of this to date — 2,961 identical brand-recommendation prompts across ChatGPT, Claude, and Google's AI, each prompt fired 60–100 times by 600 volunteers. The result should be taped to the wall of every marketing department:

These models are probabilistic. Even at temperature zero, outputs drift. So “ranking #2 in ChatGPT for your category” isn't a position you hold — it's a coin you flip, and it lands differently for every shopper. Meanwhile, roughly $100M a year is flowing into AI-visibility tracking tools measuring a number that changes every time you look at it.

This isn't an argument against GEO. You still have to be present in the conversation — being mentioned 40% of the time beats being mentioned never. It's an argument about what GEO is: it buys you a probability of being mentioned inside a black box. It cannot tell you what the customer said before the mention, what they were weighing, or why they almost picked the competitor. And you will never get that from OpenAI or Google, because the conversation is the product they're monetizing. Asking the platform for the full transcript of every shopping conversation about your brand is asking it to hand over the asset.

So the upstream layer is, structurally, one you rent and can't fully see into. The visibility you lost upstream has to be regained somewhere you actually own.

There's only one place left that qualifies.

Your Site Is the Last Place You Can See Clearly

Here's the reframe.

The on-site intelligence layer isn't only about conversion. We've made the conversion argument before — a site that can ask the same smart questions the AI just asked will close far more of that high-intent traffic. That's still the headline.

But the dark funnel adds a second benefit that's suddenly urgent. When your site can hold a real conversation with the customer — ask the clarifying questions, narrow to the right products, weigh the tradeoffs that matter to this person — you're not just converting better. You're finally seeing the consideration phase again. On your own property. In your own data.

The shopper who told ChatGPT about the knee pain and the half marathon and the deadline will tell your site the same things — if your site is capable of the same kind of conversation. The intent doesn't vanish because the AI saw it first. It re-expresses itself the moment someone competent asks. A static product page never asks, so the intent evaporates at the door. An intelligent on-site experience asks — and the richest signal in commerce, the one you lost to the chat window, lights back up where you can use it.

That's the part that compounds:

The brands relying entirely on third-party AI for the conversation are accumulating nothing. The brands running their own conversation on-site are accumulating the one asset the platforms are working hardest to keep for themselves.

The New Math

The old split was SEO and CRO. The new split is GEO and on-site intelligence — and the dark funnel changes what each side is really for.

GEO gets you into the conversation you can't see. Necessary, and not enough, because being mentioned in a black box — probabilistically, differently every time — is still being in a black box.

On-site intelligence is how you re-open the box on your own terms. It turns the highest-intent, lowest-visibility traffic you've ever had into the highest-intent, fully-visible traffic you've ever had — the moment it lands on a site that can actually talk to it.

Do only the first and you get more visitors you can't understand and a conversion rate that leaks. Do only the second and you have a great experience nobody's been pointed toward. Do both and the AI does your qualification for free, sends you a customer already deep in intent, and your site catches that intent, converts it, and — for the first time since the funnel went dark — gets to keep it.

The brands that win the next few years won't be the ones with the most AI citations. They'll be the ones who turned the part of the journey everyone else lost to the chat window into the part they understand best.

The funnel went dark for everyone at the same time. The question is who turns the lights back on, and where.

We're building the on-site intelligence layer that lets brands continue the conversation — and finally see it — when the customer arrives. If that resonates, we'd love to talk.