How Do You Design a Page for Someone Who Arrived From an AI Answer?

How Do You Design a Page for Someone Who Arrived From an AI Answer?

Design it as a single visit, not as a step in a journey. Someone who clicked through from an assistant has already read a summary of you, has a specific question left over, and will probably not look at a second page. The page has to confirm, specify and convert on its own.

This is a genuinely new brief. Most marketing pages assume a visitor who will browse, compare and come back. That assumption is now wrong for a growing slice of arrivals.

The good news is that the data on how these visitors behave is specific enough to design against, rather than something we have to guess at.

What Do We Actually Know About These Visitors?

Three things, all measured. They are fewer than the hype suggests, they explore less than search visitors, and they convert better when they act at all.

On volume, Similarweb's worldwide analysis covering June 2025 to May 2026 recorded an average of 770.7 million AI platform referral visits per month, growing 117.4 percent year over year, while noting that AI referrals stay a low single-digit percentage of most sites' total traffic.

On behaviour, Ahrefs studied 81,947 sites in May and June 2025 and found AI visitors viewed 4 pages per visit against 5.2 for search visitors, and bounced 67.8 percent of the time against 63.7 percent. On conversion, Ahrefs' own funnel data published in June 2025 showed AI search at 0.5 percent of visitors and 12.1 percent of signups.

Put those together and the visitor profile is clear. Rare, decisive, impatient, and unlikely to give you a second page to make your case.

Why Does the Single-Page Visit Change the Design Brief?

Because almost every convention in B2B web design assumes a sequence. The homepage sets up the category, the feature page explains the mechanism, the customer story builds belief, the pricing page handles objections, and the demo form closes.

An AI-referred visitor enters that sequence in the middle, at whichever page got cited, with the earlier steps replaced by something an assistant said about you. They are not going to walk backwards through your funnel to fill the gaps.

So the unit of design stops being the site and becomes the page. Every page that could be cited needs to work as an entry point, which in practice means most pages that answer a question.

What Has to Be Above the Fold Now?

Confirmation, not seduction. The first thing this visitor is doing is checking whether the summary they were given was accurate. If the top of your page does not plainly restate what the product is and who it is for, they cannot make that check and they leave.

That means the hero has to do a different job than it used to. A clever line that only makes sense once you know the category is now a liability, because this visitor may not know the category yet. State the thing, then be clever underneath it.

We are not arguing for bland. We are arguing that the first sentence should be checkable against what an assistant told them. Our broader thinking on that part of the page is in what belongs above the fold.

What Should Replace the Persuasion Sequence?

A compressed version of it, on the page. The elements you would normally spread across four pages need a presence on the one page a visitor might see: what it does, proof it works, what it costs or how pricing is shaped, and what happens next.

That does not mean a five-thousand-word page with everything on it. It means each of those four things gets a short, honest, self-contained treatment with a link to the fuller version for anyone who wants it. Give the summary on the page and the depth behind a link, rather than the other way around.

This is where design for skimming earns its keep, because a visitor checking four things quickly reads in exactly the pattern we described in designing for people who will not read.

Which Details Does an AI Answer Never Carry?

The specific, changeable, disqualifying ones. That is your competitive advantage on the page, because it is the information the assistant either did not have or was not confident enough to state.

Concretely: exact pricing structure and what triggers a tier change, seat and usage limits, which integrations exist today rather than on a roadmap, data residency and security posture, contract length, and what the onboarding actually involves. These are the questions that survive a summary.

Nielsen Norman Group's qualitative study of AI search behaviour, published in August 2025 by Kate Moran, Maria Rosala and Josh Brown, found participants moving between AI and traditional search within the same task to verify what they had been told. That is a small study, nine participants doing their own real research tasks, so treat it as a description of behaviour rather than a measurement of it. But it matches what the click data implies: the visit that reaches you is often a verification visit.

Design for verification and you are designing for the actual job.

How Do You Handle the Trust Problem?

Assume they do not fully believe the summary, including the flattering parts. An assistant's description of you is confident, unattributed, and possibly out of date. A thoughtful buyer discounts it accordingly.

What resolves that is specificity that could not have been generated. Real numbers with context, a named customer, a screenshot of the actual interface, a dated changelog entry, a documented limitation. Anything that is checkable reads as real in a way that adjectives no longer do.

Being explicit about what you are not for is unusually effective here. It is the one move an AI summary almost never makes, and it converts the visitor from checking your claims to checking their own fit.

Does This Break the Page for Everyone Else?

No, and that is the strongest argument for doing it. A page that states plainly what something is, answers the obvious questions, and makes the next step easy is a better page for a visitor from Google, from a newsletter, or from a colleague's link.

The design changes we are describing are mostly removals. Less throat-clearing before the substance, fewer claims that need context to parse, fewer paths that depend on having read another page first. None of that costs you anything with a visitor who did arrive through the front door.

The one genuine tension is with pages built as chapters of a deliberate narrative. If a page only makes sense as part three, decide whether it should be citable at all, or whether it needs a short orientation at the top. We would usually add the orientation.

How Do You Tell If the Page Is Working?

Segment the traffic before you judge anything. AI referrals need their own channel in your analytics, reported separately from organic and direct, or their behaviour will be averaged into a much larger group and become invisible.

Then judge them on their own benchmark rather than your site average. Ahrefs found these visitors bounce more and browse less as a rule, so a higher bounce rate on this segment is normal rather than a failure signal. What matters is whether the ones who do act, act at a rate that beats your organic baseline.

One caution on measurement. Similarweb notes that some AI visits arrive without referrer data and land in direct traffic, so your labelled AI segment is a floor rather than a full count. Do not tune a page against a number you know is incomplete without saying so out loud.

What Changes as This Traffic Grows?

The volume, and probably the quality per visit. Growth of 117.4 percent a year means this stops being a rounding error. Ahrefs' own author noted that AI search users click links about 75 percent less than traditional organic searchers and warned that current click-through rates are probably the highest they will ever be as the novelty wears off.

That combination points one way for design. More arrivals, each slightly less exceptional, each still likely to be a single-page visit. The page that works for them is the page that answers the question completely and does not depend on a second click.

If you want a look at whether your most-cited pages actually survive a single-page visit, that is a review we enjoy doing. Come and find us at phoenix.studio.