Something quietly significant happened in retail this year. People stopped browsing for products and started asking for them. Instead of typing “best waterproof walking boots” into Google and clicking through ten tabs, a growing number of shoppers now ask ChatGPT, Perplexity or their assistant of choice to find the right product, compare the options and hand back a shortlist.
That shift has consequences for every business with something to sell online. If an AI agent is doing the shopping, the only businesses in the running are the ones the agent can actually see.
Most SME websites were built for two audiences: human visitors and Google’s crawler. There is now a third audience, and it behaves differently from both.
The numbers say this is already happening
None of this is speculative. Q1 2026 data showed AI referral traffic to US retail sites grew 393% year on year. Traffic from AI platforms is projected to drive around $20.9 billion in retail spending in 2026, nearly four times the 2025 figure.
Growth curves like that usually come with a catch: the traffic is low quality, or it does not convert. Not this time. The same research found that AI-referred shoppers outperform traditional traffic on conversion rate, revenue per visit and engagement.
That makes intuitive sense once you think about how these visitors arrive. Someone who lands on your product page from an AI assistant has already described what they want, compared alternatives and been handed your page as the answer. The agent did the browsing. The human turns up ready to buy.
AI referrals are not just growing fast. They are some of the highest-intent visitors a website can receive. For a small retailer or service business, a channel that converts better than search is not a curiosity. It is a channel worth deliberately competing for.
Amazon just proved that visibility is a choice
Here is the most instructive story in the whole dataset. Amazon, the largest retailer on the internet, receives under 3% of ChatGPT shopping referrals. Walmart, Etsy, Target and eBay each absorb 10-20% or more.
The reason has nothing to do with preference. Amazon blocks AI crawlers in its robots.txt file. The agents cannot read Amazon’s product pages, so they recommend retailers whose pages they can read. The traffic, and the spending behind it, flows to whoever left the door open.
Amazon opted out of AI shopping referrals with a few lines in a text file. Every other retailer inherited its share.
Amazon can afford that choice. It has its own AI ambitions, its own app, and more direct traffic than any competitor. Your business almost certainly cannot afford the same choice, and here is the uncomfortable part: plenty of SME websites are making Amazon’s choice by accident.
Security plugins, CDN bot protection, and copy-pasted robots.txt templates routinely block AI crawlers without anyone deciding to. The site owner never opted out of anything. They just never checked. If your robots.txt is quietly turning away the crawlers that feed ChatGPT and its rivals, you are invisible to the fastest-growing referral channel in retail, and no amount of good product photography will fix it.
Structured data is the new price of entry
So what makes an AI assistant cite one page rather than another? The clearest signal in the research is structured data.
Analysis of pages cited by ChatGPT found that 71% of them carry schema markup. Pages with structured data earn roughly 3.1 times more AI citations than pages without it.
Structured data, usually written as JSON-LD, is a machine-readable layer that sits in your page’s code and states the facts plainly: this is a product, here is the price, here is availability, here is the brand, here are the review scores. Humans never see it. Crawlers rely on it.
This used to be an SEO nicety, the sort of thing that earned you a star rating in Google results if you got around to it. For AI visibility it is no longer a nicety. It is closer to the price of entry. An agent assembling a shortlist under time pressure will favour the pages it can parse with confidence over the pages it has to guess at.
The good news for smaller businesses is that this is a genuinely level playing field. Schema markup does not care about your domain authority or your ad budget. A well-structured product page from a ten-person retailer is as legible to an AI agent as one from a national chain, and often more legible than the chain’s bloated template.
What to do about it
None of this requires a big-company budget. It requires an afternoon of checking and some disciplined tidying.
Check your robots.txt first. It takes two minutes: visit yourdomain.com/robots.txt and look at what is being blocked. If AI crawlers such as GPTBot are disallowed and you never made that decision deliberately, change it. Check your CDN and security plugin settings for bot-blocking rules while you are there. This is the single cheapest fix in digital marketing right now.
Add structured data to your key pages. Product schema for anything you sell, with real prices, availability and review data. Service and FAQ schema if you are a service business. LocalBusiness schema if location matters. Most modern platforms support this through plugins or templates. Given that pages with structured data earn roughly 3.1 times more AI citations, this is the highest-leverage technical task on the list.
Make your pages factually dense. AI agents extract facts, and they cannot extract facts you have not stated. Real prices rather than “contact us”. Actual dimensions, materials, lead times, delivery costs and stock status. Vague brand copy gives an agent nothing to work with. A plainly written page full of specifics gives it everything.
Treat service pages like product pages. The same mechanics apply well beyond e-commerce. An assistant asked to “find an accountant near Reading who works with contractors” needs the same things a shopping agent needs: clear statements of what you do, who you serve, where you operate and what it costs.
Then verify it worked. Ask ChatGPT and Perplexity the questions your customers would ask, and see whether you appear. It is a crude test, but it is the test that matters.
The window is open, for now
Every channel like this follows the same arc. Early on, visibility is cheap because most competitors have not noticed. Later, everyone piles in and the advantage evaporates.
AI shopping referrals are early. Most SMEs have not checked their robots.txt, have no structured data worth the name, and are competing for AI citations without knowing the contest exists. The businesses that fix the basics this year will be the ones the agents recommend by default next year. The work is unglamorous. The window will not stay open forever.
Flux Dynamics helps SMEs make their websites legible to AI agents and search engines alike, as part of fractional CTO engagements. Tell us about your project and we will give you an honest read on where you stand.