Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
The Visitor Arrives Mid-Sentence
Someone who clicks through from an assistant has just been told what the category is, which options exist, and why you might suit them. They land on your page holding a specific claim and a specific doubt. The page's job is to resolve the doubt, and almost every page instead restarts the conversation from the beginning.
Picture the actual sequence. A person asks an assistant for help choosing something. The assistant produces a summary naming three or four options, yours among them, with a sentence about what each is good for. The person clicks your citation to check whether the sentence is true. They are not asking "what is this category". They are asking "is this the thing that was just described to me, and does it fit my situation".
Now picture what they hit: a hero headline defining the category, a three-column section explaining why the category matters, and a contact form at the bottom. Every element of that page was designed for a visitor at the start of their thinking, because that is the visitor search traffic used to be. The mismatch is not subtle, and it is the reason a traffic source with unusually good intent can still produce unremarkable results.
What the Behaviour Data Actually Shows
Adobe analysed more than a trillion visits to US retail sites and reported that AI-referred traffic converted 42% better than non-AI traffic in March 2026, with those visitors spending 48% more time on product pages, viewing 13% more pages per visit, and showing a 12% higher engagement rate. The pattern across all four numbers is consistent: deliberate verification, not casual browsing.
Larger multiples circulate widely, including a frequently repeated figure of roughly four to five times better conversion, with reported ranges stretching from marginal in low-consideration retail to extreme in business software. The honest reading of that spread is that the direction is well supported and the magnitude is entirely dependent on your category. Anyone quoting a single multiplier for your business is quoting someone else's.
The behavioural numbers are more actionable than the conversion ones anyway. Longer sessions and more pages per visit tell you these visitors are willing to read, compare, and dig. That is permission to put substance on the page rather than a distillation of it. The instinct to simplify everything for a distracted skimmer is calibrated to a different audience than the one arriving here.
The Five Things an Assistant Cannot Carry
An assistant's summary is compressed, potentially stale, and necessarily generic about your specifics. Those gaps are exactly what your page should fill, in the order a verifying reader looks for them.
- Confirmation that you are what was described.
Within the first screen, in plain language that matches how the assistant would have characterised you. If a visitor cannot match the page to the sentence that sent them within a few seconds, they assume a mis-citation and leave. This is a headline problem, and the fix is usually removing cleverness rather than adding it.
- Current price, or the honest reason there is none.
Pricing is the most commonly stale fact in any AI summary and the most commonly checked fact on arrival. A visible number resolves the visit. "Contact us" on a page reached by someone in comparison mode is an invitation to go back and pick a competitor who answered.
- Fit and non-fit, stated explicitly.
Who this is for, and who it is not for. Assistants generalise, so the visitor genuinely does not know whether their constraint is handled. A short section naming the cases where you are the wrong answer converts better than a longer one claiming you suit everybody, because it is checkable.
- Proof that does not come from you.
A named customer, a third-party review, a verifiable number. The visitor has just received a recommendation from a source they consider neutral and is now reading a source they know is not. Self-description does not move that gap; corroboration does, which is also what tends to drive citation in the first place.
- A next step sized for a decision.
Not a newsletter, not a downloadable guide about the category they have already had explained. Something that advances the choice: a real trial, a live check against their own data, a calculator with their numbers in it. Interactive next steps consistently outperform static ones with this audience, because the audience arrived wanting to test a claim.
Auditing a Page Against the Arriving Reader
Take your five highest-value pages and read each one as though an assistant just told you what it does. Count how many screens pass before the page tells you something the summary did not already contain. If the answer is more than one, the page is written for the wrong visitor.
| Page element | Built for search-era visitors | Built for assistant-referred visitors |
|---|---|---|
| Headline | Defines the category and its importance | Confirms what this specific thing is and who it suits |
| First section | Why the problem matters | What is true about this option, specifically |
| Pricing | A separate page, or on request | Visible, current, and dated |
| Proof | Logo bar | Named, checkable, third-party |
| Primary action | Learn more, subscribe, download | Test the claim against your own situation |
Two cautions on execution. Do not build a separate page for this audience: serving different content based on who is asking is cloaking, and assistants cite the canonical URL that ordinary search knows about anyway. And do not treat your analytics AI segment as a precise audience, because a large share of these sessions arrive with no referrer and sit in Direct, as covered in why your AI traffic says Direct. You are improving pages for a better-informed reader in general, which is a change that pays regardless of how the visitor got there.
The Compounding Part
Pages restructured this way are also easier for assistants to quote correctly. Explicit fit statements, visible pricing, and question-shaped headings with direct answers underneath are the same properties that make a page safely extractable. Writing for the arriving visitor improves the odds of arrival.
That loop is the strongest argument for doing this work before chasing more visibility. A page that gets cited but cannot convert the resulting visitor turns attention into nothing, and the fix costs an afternoon per page rather than a quarter of content production. The visibility side of the same system is in the AI search visibility playbook, and the measurement of whether assistants are naming you at all is in share of model.
Start with the pages where a decision actually happens, not the blog. For most businesses that is a short list: the service or product pages, the pricing page, and any comparison pages you maintain. The structural work for those is in service page SEO and comparison and alternatives pages.
Questions People Ask About Converting AI Traffic
- Do visitors from AI assistants convert better?
The evidence points that way, with a range wide enough to demand you measure your own. Adobe, analysing more than a trillion visits to US retail sites, reported AI-referred traffic converting 42% better than non-AI traffic in March 2026. Other published figures run considerably higher, and the variance across industries is large enough that adopting any single multiplier as a forecast would be false precision.
- Why do AI-referred visitors behave differently?
Because the education already happened somewhere else. The assistant defined the category, narrowed the options, and named you as one of them, so the visitor arrives holding a shortlist rather than a question. Adobe's retail data showed these visitors spending 48% more time on product pages and viewing 13% more pages per visit, which is the behaviour of someone verifying a decision rather than starting one.
- What should a landing page do differently for AI traffic?
Stop re-explaining the category and start confirming the claim that sent them. The assistant made an assertion about you, and the visitor is checking it. The page needs to substantiate that assertion quickly with specifics an assistant could not carry: current pricing, real availability, the constraints you do and do not fit, and proof a third party can verify.
- Can I build separate pages for AI traffic?
You should not, and it would not work anyway. Serving different content based on the visitor or crawler is cloaking, and assistants generally cite the canonical URL that ordinary search knows about. The productive version is making the pages you already have work for a better-informed reader, which improves them for everyone rather than creating a parallel site to maintain.
- How do I know which visits came from an assistant?
Imperfectly, and you should build for the imperfection. A meaningful share of assistant referrals arrive with no referrer header and land in Direct, so any AI segment you build is a floor rather than a count. Set up the channel grouping properly, then treat the resulting segment as directional evidence rather than a precise audience you can target.
