Designing a Page for Visitors Who Arrive From an AI Answer

How is a visitor from an AI answer different?

They arrive already briefed. An assistant has given them a summary and they clicked because they wanted something the summary could not give them: detail, proof, a specific situation, or a reason to trust the source. Repeating the summary at them wastes the only advantage you have.

This is a design problem more than a writing problem. The structure that serves someone arriving cold from a search result serves this reader badly, because it spends the first screen establishing context they already have.

Here is how I now think about page structure when a meaningful share of arrivals come this way.

What does the AI summary already tell them?

The general shape of the answer, drawn from several sources at once. Google's guidance on generative AI features describes retrieval augmented generation as a technique that retrieves relevant, up to date pages from the Search index and reviews information from those pages to generate a response, showing prominent clickable links to the pages supporting it.

Read that mechanism carefully, because it defines your visitor. They have seen a composite answer assembled from you and several competitors, and then chosen your link specifically. That choice is a signal about what they wanted more of.

The documentation also describes query fan-out, a set of concurrent related queries the model generates to gather more information, giving the example of a query about fixing a weedy lawn producing follow up queries about herbicides and prevention. So the visitor may have arrived through a question adjacent to the one your page answers.

That combination explains the two things this reader wants. Depth on the specific thing they clicked for, and evidence that you actually know it rather than having summarised someone else.

Why does the top of your page need to change?

Because the standard opening is now redundant. Most pages begin by explaining what the topic is and why it matters, which is exactly what the assistant already did. A reader who gets that again concludes, within a few seconds, that your page adds nothing.

Attention is unforgiving here. Nielsen Norman Group's eyetracking research found that users spend about 57 percent of their page viewing time above the fold, and 74 percent within the first two screenfuls. Spending that budget restating a summary is the most expensive mistake available.

So lead with the part the summary could not carry. A specific number with a source, a named tradeoff, a situation where the general advice fails, or a first hand account of what happened when you did it. Something that could only come from a person who has done the work.

This does not mean removing the definition. It means putting it where a reader who needs it can find it, rather than in the place where a reader who does not need it has to walk past it. I dealt with the ordering logic more generally in how far down a long page attention actually goes.

Where do these visitors actually land?

On whichever page contained the passage that was cited, which is frequently not your homepage and frequently not the top of the page they land on. Deep arrival is the norm, and pages that only make sense from the top are pages these visitors bounce off.

Design each section to stand alone. A section that opens by referring to what was just explained is fine for a linear reader and useless for someone who arrived in the middle of it. Name the subject again, briefly, at the start of every section.

Give them an orientation cue too. A visitor who lands mid page should be able to tell, without scrolling up, what this page as a whole is about and who wrote it. That is a small design decision with a large effect on whether they stay.

What should each section do for a partly informed reader?

Answer one question completely and independently. Retrieval systems pull passages rather than pages, which means every section is a potential entry point and a potential citation. The structural discipline that makes a section quotable is the same one that makes it usable to a human who lands on it.

Keep the answer and the evidence adjacent. A claim in one paragraph and its source three paragraphs later reads as unsupported to a skimmer and is hard to extract cleanly for a machine. Put the number and the attribution in the same sentence wherever you can.

Then add the thing a summary structurally cannot include, which is a judgement. Summaries are consensus machines. They flatten disagreement. A section that says plainly which option you would pick and why is the one thing a composite answer will never contain. The mechanics of how passages get selected are in why AI search quotes one paragraph.

How do you earn the click that follows the answer?

By being the obvious next step rather than the destination. This visitor is mid task. They wanted information, they have most of it, and the useful thing you can offer is either the depth that finishes the job or the person who can do it for them.

Match the offer to how far along they are. Somebody who arrived from a definitional question is not ready for a sales conversation, and asking for one reads as tone deaf. Somebody who arrived from a comparison question might be very ready indeed.

Keep the path short and visible without being intrusive. One clear invitation, placed where the reader has just been given something useful, converts better than three placed where they have not.

What should you stop doing on these pages?

Stop front loading atmosphere. A large decorative image, a mood setting headline and a scroll cue are pure cost for a reader who arrived with a specific question and a summary already in their head. They were never free, and this visitor makes the bill visible faster than any other kind.

Stop hiding the substance behind interaction. Accordions, tabs and content that appears on scroll all put a gate between the reader and the thing they came for. They also make your content harder for a retrieval system to see in the first place, so you lose twice.

And stop writing openings that congratulate the reader for being interested in the topic. Those paragraphs were always weak. For a visitor who arrived from an assistant, they are actively insulting, because the assistant was more efficient than you are being.

How do you tell if this traffic is arriving at all?

Look at your entry pages and your referrers rather than at your homepage. Traffic from assistants tends to be low in volume and unusually engaged, so averages hide it. Sort by pages that receive few visits with long engagement and you often find it.

Watch which pages get cited too, by asking assistants the questions your pages answer and seeing what comes back. That is a crude method and it is the one that tells you whether your content is in the answer at all, which is the prior question to any design change.

Then check the shape of what these visitors do. If they land, read one section and leave, that may be a success rather than a failure. You answered the question. Judge these pages on whether the right people take the next step, not on time on page.

What should you do next?

Take your three most cited or most likely to be cited pages and read the first two paragraphs of each as if you already knew the summary. If those paragraphs tell you nothing new, rewrite them with your most specific material first and move the general explanation lower.

Then make every section independent. Name the subject at the start, keep the claim and the evidence together, and state a judgement somewhere on the page that a consensus summary could never produce. The skimmable structure that supports all of this is in designing skimmable headings.

If you want a read on whether your pages are set up for this kind of visitor, send me one and tell me the question it is meant to answer. I will tell you what a briefed reader would find in the first ten seconds.