AI shopping is moving out of the “interesting future trend” category and into something merchants can measure.
John Lewis told Reuters that AI-agent product searches now account for about 2.5% of its product searches, up from 0.3% a year earlier. Adobe separately reported that AI-referred traffic to U.S. retail sites rose 62% year over year in July 2026, with those visitors converting 60% better than non-AI traffic.
Those numbers do not mean AI is replacing Google, marketplaces or direct traffic. They do mean merchants now have enough evidence to stop treating AI discovery as a theoretical conversation.
The practical question is becoming: if an AI sends someone to your site, can your site actually explain the business well enough to deserve the recommendation?
The signal is stronger because the evidence is coming from different places
John Lewis is one retailer. Adobe’s dataset covers a much broader slice of U.S. retail traffic. Neither source should be universalized, but together they point in the same direction.
Reuters reported that John Lewis saw AI-agent product searches rise from 0.3% to 2.5% in a year. That is still a minority of total product-search activity, but it is large enough to become operationally visible.
Adobe’s July data is stronger evidence of what happens after the click. AI-referred retail visitors converted 60% better, showed 14% higher engagement, spent 59% more time on site and were 33% less likely to bounce than non-AI visitors. Adobe also said AI-referred shoppers added items to carts at a 28% higher rate.
That is not proof that AI traffic will outperform for every merchant. It does suggest these visitors are often arriving after doing more research upstream.
AI may be compressing the research phase
A customer who used to open ten tabs may now ask an AI assistant to compare the options first.
By the time that customer reaches a retailer, they may already know the product category, the tradeoffs, the rough price range and the questions they need answered.
That could explain why AI-referred traffic looks unusually engaged. The site visit is happening later in the decision process.
If that is true, merchants should not judge AI solely by referral volume. A small channel that consistently sends better-prepared visitors can matter before it becomes a large share of traffic.
The uncomfortable part: many websites are still hard for machines to read
Adobe also measured what it calls AI visibility across retail websites. In July, the average homepage visibility score across the broader U.S. retail sample was 61%, meaning substantial portions of many homepages were not fully readable by machines.
That does not mean every website needs a giant “GEO project.” It means the same content problems that frustrate customers can also make a site harder for AI systems to understand.
Important information may be trapped inside images, vague marketing copy, JavaScript-heavy interfaces or inconsistent product data. Product names, pricing, availability, specifications, policies and service-area details may be technically present but difficult to extract with confidence.
An AI system cannot confidently recommend what it cannot confidently understand.
Start with the boring fixes
Before paying for a new category of optimization, I would fix the fundamentals.
Make product and service pages explicit. Use real headings. Put critical facts in readable text. Keep product attributes and availability accurate. Make shipping, returns, guarantees and contact information easy to find. Use structured data where it genuinely matches the page content. Make sure canonical URLs and internal links are sensible.
Then look at the content people use to make decisions. Do your pages explain who the product is for, how options differ, what the tradeoffs are and why someone should trust the business?
These are good website practices regardless of whether the visitor is a person, Googlebot or an AI agent.
Measure AI traffic separately now
Another simple move is to stop hiding AI referrals inside generic referral traffic.
Track visits from ChatGPT, Perplexity, Gemini and other identifiable AI sources separately. Watch conversion rate, average order value, engagement, assisted conversions and the pages those visitors enter through.
The totals may still be small. That is fine. The goal is to establish a baseline while the channel is early enough that changes are still easy to see.
Do not make strategic decisions from a dozen sessions. But do not wait until AI referral traffic is a major percentage of revenue before deciding to measure it either.
Do not confuse machine readability with writing for machines
There is a bad version of this strategy where every product page becomes a pile of robotic keyword blocks designed to impress an LLM.
That would recreate the worst years of SEO with a new acronym.
The better goal is to make the business unusually clear. Accurate facts, useful comparisons, original expertise, good reviews and content that answers real customer questions are valuable precisely because both people and machines can use them.
The best AI-readable site is usually a site that stopped making customers guess.
This is measurable now
AI shopping is still early. Direct AI traffic remains small compared with established channels, and one retailer’s search share should not be turned into an industry-wide forecast.
But we now have merchant-level usage data, broad retail referral data, conversion differences and evidence of machine-readability gaps.
That is enough to change the operating posture.
You do not need to rebuild your ecommerce strategy around AI this week.
You should make sure your website is understandable enough that, when AI does send someone, the business is ready for the visit.
Sources
Ready to scale your brand?
Let's talk about what growth looks like for your business.
Book a Free Strategy Call