Designing for AI search, when the answer isn't a page.
For twenty years the deal was simple: rank a page, get a click, own the reader from there. That deal is being renegotiated without us. A growing share of searches now end in an answer the user never had to click to read — and the brands being quoted inside those answers are not always the ones ranking first.
This is the question we get most often now, usually phrased as some version of "is SEO dead?" It is not, and the panic is mostly misdirected. But the shape of the work has changed, and the parts that changed are not the parts most agencies are talking about.
Here's what we've actually adjusted, and what we've deliberately left alone.
What changed: the click is no longer the unit
The old funnel had one job for a page: earn the click. Everything downstream — the conversion, the trust, the enquiry — happened on your property, under your design control.
AI answers break that model in a specific way. The reader now meets a summary of you, assembled by a machine, before they meet you. Sometimes they arrive afterwards. Often they don't arrive at all, having got what they needed.
That sounds like a loss, and for thin content it is. A page that existed only to capture a query and funnel a click has lost its entire reason to exist. But for brands with genuine expertise, something more useful happens: the summary becomes a referral with a citation attached. The traffic that does arrive is further down the decision than it used to be.
What gets quoted, and what gets skipped
We've been tracking which of our own pages and our clients' pages surface inside AI answers. The pattern is not subtle, and it is not really about SEO at all.
Quoted pages tend to share four traits:
- They answer one question. Not a topic — a question. Pages built around "everything about X" get summarized into nothing.
- They commit to a position. Hedged writing has nothing extractable in it. "It depends on your needs" is unquotable by design.
- They show their work. Numbers, named methods, specific trade-offs. Anything a model can attribute back to a source.
- They come from a recognizable entity. The same signal that governs traditional ranking, doing more work than ever.
Skipped pages are usually the mirror image: broad, hedged, unsourced, and published by an organization the machine cannot confidently identify.
A model can only quote you if there is something in the page worth quoting. Most SEO content was engineered to avoid saying anything that specific. — Egzecute SEO note
That last trait is why we keep pushing entity work ahead of content volume. A brand with a coherent, verifiable identity gets cited. A brand the machine can't resolve gets paraphrased without attribution, if it's used at all — which is the same problem we wrote about for brand-name search, arriving through a different door.
What this means for design
Here's the part that gets left out of most write-ups on this, and the part that matters most to us: the design implications are real, and they run in the opposite direction from what people expect.
The intuitive response to "machines read my site now" is to design for machines — strip things back, prioritize parseability, treat visual craft as overhead. That is exactly backwards.
If a share of your audience gets your substance from a summary, then the people who do click through are self-selected: they wanted more than the answer. They came to evaluate you. What they land on is no longer competing with nine other search results — it's competing with the impression the summary already gave them.
| Pre-AI-answer search | Post-AI-answer search | |
|---|---|---|
| What the click is worth | One visitor among many results | A visitor who chose you specifically |
| What the page must do first | Convince them to stay | Confirm they were right to come |
| Cost of a generic template | Some lost conversion | Contradicts the expertise they just read |
| Role of visual craft | Differentiator | Evidence |
| Content that wins | Comprehensive | Specific, attributable |
That fourth row is the one we'd underline. When a prospect arrives already half-convinced by something a machine quoted from you, the site's job is corroboration. A generic template quietly undercuts the claim. This is not an argument for decoration — it is an argument that craft has become a form of proof.
What we haven't changed
Roughly as important, and much less discussed. We have not:
- Started writing for models. Every "optimize for LLMs" tactic we've tested either does nothing or is a rewording of "write clearly and be specific," which was already the job.
- Abandoned long-form. Depth still wins, because depth is what produces the specific, quotable claim in the first place.
- Added an FAQ section to everything. Schema-stuffed FAQ blocks were a 2019 tactic. Answering real questions in the prose works better and reads like a human wrote it.
- Chased volume. Publishing more mediocre pages was a bad strategy when humans were the only readers. It's worse now.
Take any page you've published this year and try to extract one sentence from it that a competitor could not also honestly publish. If you can't find one, the page has nothing a model can cite and nothing a reader can remember. That is a content problem, not a technical one — and no amount of schema markup will fix it.
Where this actually lands
The uncomfortable summary is that AI search has made the fundamentals matter more, not less, while making the shortcuts stop working entirely. Brands with real expertise, a coherent identity, and the willingness to say something specific are being cited more than they were. Brands that were coasting on keyword coverage have lost the mechanism that was carrying them.
That's an unglamorous conclusion, and it's the one the data keeps supporting. The work is: be a legible entity, know something worth quoting, say it plainly, and build a site that holds up when someone arrives to check whether you meant it.
None of that is new. It's just that the alternatives have stopped working.
