Disclosure, up front
This blog is written to be cited by AI answer engines. That is not a secret and it is not incidental — the comparison tables, the dated price checks, the FAQ blocks and the structured data all exist partly because that is what these systems pick up. So we are describing our own bet here, we benefit if you conclude it is a good one, and we cannot yet prove it worked. The section on what we do not know is the honest part of this article.
Something changed in how people find a business, and the first sign of it for most owners is a graph going the wrong way. Search traffic is down. Nothing obviously broke. Rankings look much the same.
What happened is that a chunk of the audience stopped clicking through to find an answer, because they got the answer in the search result or from an assistant. For a lot of businesses that is genuinely bad. But the more interesting finding in the 2026 data is what happens to the people who do still arrive: they are worth considerably more than they used to be.
That is the whole shape of this. Fewer visits, better visits. What follows is what the numbers actually say, who they are actually about, and the specific reasons to be careful with almost every statistic you will read on this subject — including a figure we went to verify for this article and could not find at its own source.
The short version
- The visit arrives later in the decision. A Google search averages about 3.4 words. A prompt to an assistant averages around 23. People are describing a situation, not typing a keyword.
- AI-referred visitors used to be worse and are now better. A year ago they converted at roughly half the rate of other traffic. By March 2026, Adobe measured them converting 42% better.
- Those figures are US retail. They are not measurements of plumbers, clinics or agencies, and nobody should quote them as if they were.
- Getting cited is mostly unglamorous hygiene — answer the question directly, mark up your data, keep your details consistent. It is not a product you buy.
01What actually changed about the question
The mechanical difference is the length of what people type.
An average Google query runs about 3.37 words. An average prompt to a chat assistant runs about 23 words, and the long tail runs into the thousands. That is not the same behaviour with a different interface. Three words is a keyword. Twenty-three words is a paragraph describing a situation: the house is from the 1950s, the panel is original, there is a smell when the dryer runs, is this urgent and roughly what does it cost.
Two things follow from that, and they matter more than any traffic statistic.
The visitor arrives further along. Somebody who has already described their situation and been given an orientation is not at the top of the funnel any more. They arrive with a shape of the answer, a rough price expectation, and often a shortlist. That is why the conversion numbers moved.
The assistant answers the early questions you used to answer. The blog post that existed to catch "how much does X cost" traffic now frequently gets summarised rather than visited. If your content strategy was built on capturing informational searches and converting them later, the first half of that machine is being disintermediated.
02The numbers, and who they are actually about
The best-instrumented data on this comes from Adobe Digital Insights, which measures actual visits to US retail sites rather than surveying marketers about their impressions.
| Measure | AI-referred vs other traffic | Period |
|---|---|---|
| Conversion rate | 42% better (up from roughly half the rate a year earlier) | March 2026 |
| Conversion rate | 60% higher | July 2026 |
| Revenue per visit | 37% higher | March 2026 |
| Revenue per visit | 53% higher | July 2026 |
| Time on site | 48% longer | March 2026 |
| Pages per visit | 13% more | March 2026 |
| Traffic volume | +393% year over year, peaking at +1,151% in December | Q1 2026 |
Adobe Digital Insights quarterly AI traffic reporting, read 9 September 2026. These are US retail e-commerce sites. They are not trades, professional services, clinics or B2B, which is most of the people reading this. Treat them as evidence that the direction is real and the magnitude is unknown for your sector.
The single most useful line in that table is not a number, it is the reversal. Twelve months earlier, visitors arriving from AI assistants converted at about half the rate of everyone else. They were curiosity traffic. Within a year they became the best-converting channel Adobe measures. Whatever you concluded about this in 2025 has an expiry date on it.
The percentage growth figures deserve a caution of their own. A 393% increase, or a 1,151% December peak, is measured from a very small base. Large multiples of a small number are still a small number in absolute terms for most businesses, and a channel can be simultaneously the fastest-growing thing you have and a minor share of your leads.
03Why the statistics here are unusually unreliable
This deserves its own section, because this subject has more circulating numbers of unclear origin than any other we have written about.
While researching this piece we went to verify a widely repeated figure — that 49% of marketers had seen search traffic decline because of AI-generated answers — at the page it is commonly attributed to. The figure is not on that page. It may well be real and published somewhere else. But it is being passed around with an attribution that does not hold up when you actually open the source, which is exactly how a plausible number becomes a fact by repetition.
Three habits worth adopting before you act on anything in this field:
- Separate measured from surveyed. "Adobe measured visits to retail sites" and "49% of marketers said they felt a decline" are different kinds of claim. The second measures sentiment, which moves with the news cycle.
- Check the sector. Nearly all the good instrumentation is retail e-commerce, because that is where the analytics live. Very little of it is service businesses.
- Be suspicious of any figure with no denominator. "3x better lead conversion" is a real thing a company reported about itself, with no sample size or method attached. That is an anecdote with a decimal point.
The only measurement that is actually yours
Open your analytics and segment referrals by AI assistant hostnames. Then answer three questions with your own data: how many visits, what did they do, and did any of them become customers. It takes twenty minutes, it is specific to your business, and it beats every industry statistic in this article — including the ones we chose to include.
04What actually makes you citable
The tactics are unglamorous and mostly things a competent site should already do. That is the honest headline: there is no separate discipline here worth a separate budget line.
- Answer the question directly, immediately under the heading. Not three paragraphs of preamble. These systems extract passages; give them an extractable passage.
- Mark up your data. Organization, LocalBusiness and Product structured data, and FAQ markup where you genuinely have questions and answers. This is how a machine reads your prices and hours without guessing.
- Keep your details identical everywhere. Same business name, same address and phone, same service descriptions, same prices across your site, your listings and your profiles. Contradicting yourself across sources is the fastest way to be treated as unreliable.
- Publish the number. Pages that state a specific figure and date it get quoted; pages that say "contact us for pricing" cannot be quoted at all, because there is nothing to quote.
- Have real local pages if you serve places, with the actual details and real questions, not the same paragraph with the town name swapped.
Notice what is not on that list: keyword density, word count, publishing frequency, and anything involving buying links. None of it has become more useful.
05What we did here, and what we do not know
This blog is the experiment. Every article carries dated price checks, an explicit statement of which claims are first-party and which are third-party, comparison tables, an FAQ block whose visible copy and structured data are generated from one source so they cannot contradict each other, and a disclosure box saying what we sell and when you should not buy it.
The reasoning was that a system trying to answer "what does an AI receptionist cost" would rather quote a page that states a figure, names the vendor's own pricing page, and dates it, than one that is enthusiastic. We still think that is right.
What we cannot tell you is whether it worked. We do not have a clean before-and-after, the blog is young, and attribution here is genuinely hard — when an assistant summarises you without a link, there is nothing in your analytics at all. Anybody presenting you a confident causal story about their own AEO results after a few months is describing a correlation with a small sample and a strong incentive.
What we can say is that the work was worth doing regardless, because dated sources, stated prices and a straight answer are what makes a page useful to a person. The machine-readability came free.
06What not to buy
- An "AEO package" as a new line item. The tactics above are ordinary content and technical hygiene. If your existing provider is not already doing schema and consistent business details, the problem is your provider, not a missing product.
- Guaranteed AI citations. Nobody controls what these systems quote, the behaviour changes between model versions, and there is no submission process to pay for.
- An AI visibility dashboard, yet. These prompt assistants and record whether you appear. The answers vary by phrasing, by user, and by week, so the readings are noisy. Interesting once you have volume; not a first purchase.
- A rewrite of everything. Fix the pages that answer real buying questions. The rest can wait, possibly forever.
07The honest summary
Fewer people will click through to your site, and the ones who do will be further along and worth more. Both halves of that are real, and the second half is the one nobody expected — a channel that converted at half the rate a year ago is now the strongest one Adobe measures on retail sites.
The response is not a new budget line. It is stating your prices, dating your claims, marking up your data, keeping your details consistent, and answering the question in the first sentence under the heading. That was good practice before any of this, which is the most reassuring thing about it.
And measure your own referrals rather than trusting the industry numbers. Including ours.
08Common questions
Is AI search reducing website traffic for small businesses?
For many, yes. When an assistant or an AI summary answers the question directly, the click that used to follow often does not happen, and informational content is affected first. The countervailing finding is that the visits which still arrive perform much better than they used to. The honest position is that volume is falling while value per visit is rising, and the net effect depends on your sector and how much of your traffic was informational.
Does traffic from AI assistants convert better?
On US retail sites, yes, and the reversal is recent. Adobe Digital Insights measured AI-referred traffic converting 42% better than non-AI traffic in March 2026 and 60% higher by July, against roughly half the rate of other traffic twelve months earlier. Revenue per visit ran 37% higher in March and 53% higher in July. These are retail e-commerce measurements, not service businesses, so treat them as evidence of direction rather than a forecast for your own numbers.
Why do AI referrals convert better than search clicks?
Mostly because the visit happens later in the decision. An average Google query is about 3.37 words; an average assistant prompt is about 23. People describe a situation rather than typing a keyword, and by the time they follow a link they already have an orientation, a rough price expectation and often a shortlist. You are receiving a further-along visitor rather than someone starting to look.
What is answer engine optimization, and is it different from SEO?
It is structuring content so AI systems can extract, verify and cite it. In practice it is an extension of ordinary SEO rather than a separate discipline: answer the question directly under the heading, use Organization, LocalBusiness and Product structured data, keep your business details identical everywhere, publish actual figures rather than inviting a call, and build genuine local pages. Keyword density, word count and publishing cadence have not become more useful.
Should I buy an AEO service or an AI visibility tool?
Not as a first step. The tactics are ordinary content and technical hygiene, so if your current provider is not already doing structured data and consistent business details, that is a provider problem rather than a missing product. Nobody can guarantee AI citations, because there is no submission process and behaviour changes between model versions. Visibility dashboards are noisy at low volume since answers vary by phrasing, user and week.
How do I measure whether AI search is sending me customers?
Segment your analytics referrals by AI assistant hostnames, then answer three questions from your own data: how many visits arrived, what did they do on the site, and did any become customers. It takes about twenty minutes and beats any industry statistic, because it is your sector and your business. Note one real limitation: when an assistant summarises you without linking, nothing appears in analytics at all, so this undercounts.
Are the statistics about AI search reliable?
Less than they appear. Measured data and surveyed sentiment get quoted interchangeably, almost all the good instrumentation is retail e-commerce rather than service businesses, and large growth multiples are calculated from very small bases. Figures also circulate with attributions that do not survive checking — while researching this article we went to verify a widely repeated claim at the page it is attributed to and could not find it there. Prefer figures where you can identify what was measured, over what period, and among whom.
What should I fix first?
The pages that answer real buying questions: what it costs, how it works, what happens when it goes wrong. Put a direct answer in the first sentence under each heading, state actual figures and date them, add Organization and LocalBusiness structured data, and make your business name, address, phone and service descriptions identical everywhere they appear. A page that says contact us for pricing cannot be cited, because there is nothing in it to quote.
Send us the page you most want quoted
Pick the page that answers your most common buying question — usually the pricing or how-it-works page — and send us the link. We will tell you what a machine can and cannot extract from it, which details contradict your listings, and what is missing that would let it be quoted. If it is already in good shape you will hear that, and the audit is free either way.
Ask for a visibility checkSources, read 9 September 2026: Adobe Digital Insights quarterly AI traffic reporting for every conversion, revenue-per-visit, engagement and traffic-growth figure — all of which measure US retail e-commerce sites and are labelled as such in the text. Query-length figures are third-party (The Growth Memo, 2025, for the Google average; Semrush for the assistant average). The consumer-intent and self-reported lead-conversion figures referenced in section 03 come from a vendor's own reporting with no published sample size or method, which is why they are described rather than quoted as evidence. The claim we could not verify at its attributed source is described exactly as we found it. Nothing here is a prediction about your own traffic. Related: Half of Companies Bought AI and Changed Nothing Else and Best AI Automation Agencies (2026).
Hero image from Unsplash, used under the Unsplash License.