Marketing × design · February 12, 2026
What Happens to the Funnel When the Agent Does the Browsing
We have spent twenty years designing pages for a person who scans, hesitates, and can be persuaded. A growing share of the traffic now does none of those three things.
I have worked the marketing side of this for long enough to have watched the funnel get renamed about four times without changing much underneath. Somebody searches, they land, they read, they hesitate, they convert. Every technique any of us has written down assumes that shape.
That shape assumes a person with eyes, attention, and doubt. Which is worth saying out loud, because a growing share of the traffic arriving at your pages has none of those.
What an agent does on your page
It doesn't land on one page. It retrieves several, extracts what it can, compares, and returns one answer to somebody who never sees any of them.
So consider what that skips. It doesn't scan, so visual hierarchy does nothing for it. It doesn't hesitate, so scarcity and urgency have nowhere to act. It doesn't feel reassured by a logo wall, because a logo wall is an image and an image isn't an argument. It won't be moved by your hero copy. It's looking for facts it can extract and attribute.
The uncomfortable part is that the middle of the funnel (the part we're proudest of, the part we A/B test) is the part it walks straight past.
Which isn't the same as saying the page stops mattering
I want to be careful here, because there's a fatalistic version of this argument going around and it's wrong.
The page matters more, not less. It now has two readers with completely different needs, and only one of them can be charmed.
An agent needs your claims to be specific, structured, attributable, and consistent with what you say elsewhere. A person needs to feel like this is for them. Those aren't in conflict as often as people assume, and where they do conflict, it's usually because the marketing copy was vague in a way that was never serving the person either.
Most of what makes a page legible to an agent is what makes it honest. The price on the page. The number stated rather than implied. The limitation admitted somewhere findable.
What changes in practice
The specific claim beats the persuasive one
Industry-leading uptime is nothing to an agent. It can't extract it, can't attribute it, and can't compare it. 99.95%, measured monthly, with the last twelve months published, is a fact that survives being read by software and it also happens to be more convincing to the human.
Answer the question on the page that gets retrieved
This is the actual shift behind all the AEO and GEO acronyms, and it's less exotic than the acronyms suggest. Retrieval happens at the chunk level, not the page level. A paragraph that answers a real question, in the words somebody would use to ask it, is what gets pulled, and it needs to make sense lifted out of its page, because that's how it will arrive.
Structure the things you would rather bury
Pricing that only appears after a form is invisible to retrieval. So are specifications inside a PDF, limitations mentioned only on a call, and comparison tables built as images. Every one of those is a decision to be absent from the answer, and most of them were made for reasons that no longer hold.
Assume the form gets filled by something that isn't a person
Multi-step forms, progressive profiling, and anything that depends on hesitation are all built around a psychology that isn't present. Meanwhile every dark pattern any of us ever inherited (the pre-checked box, the buried opt-out) gets read as written, with no charitable interpretation at all.
The measurement problem nobody has solved
Here is where I have to be honest about the limits of what I know.
If an agent reads your page and somebody buys three days later having never visited it themselves, your analytics show a direct session and nothing else. The page did the work. It gets none of the credit, and the model you use to decide what to fund is now wrong.
I don't have a clean answer for this and I'm suspicious of anyone who says they do. What I've started doing is treating unattributed conversions as a signal rather than noise, and watching whether that share grows. It's a crude instrument. It's better than assuming the number is zero.
The takeaway
The funnel didn't disappear. A second one opened up next to it, and it runs on different fuel.
One rewards persuasion, hierarchy, and the feeling that somebody understands your problem. The other rewards specificity, structure, and claims that hold up when they're lifted away from the design that was carrying them.
The good news, and I do mean this, is that the second list is a description of a page we should have been building anyway. It took something that can't be charmed to make the case for it.