It’s Sunday evening, and a parent asks ChatGPT: “Help me find school shoes for a nine-year-old boy who puts his shoes through a lot and needs a wide fit.” They should cost under £40 and must arrive by Monday.” Within seconds, the AI assistant offers a shortlist. Now, if you asked a friend the same question, you’d be sent to the shop other parents swear by at the school gate, and AI assistants learn a great deal from exactly that kind of talk. A retailer with hundreds of school shoes might be left off the list entirely.
Retailers have always been able to type a search into Google and see where they rank. But AI shortlists are put together inside private conversations, so as AI assistants increasingly recommend just a handful of options, a retailer may never know it’s been overlooked.
What makes AI recommend a retailer?
Think about what the parent really needs. They aren’t simply shopping for school shoes. They need shoes that survive the playground, fit properly, arrive tomorrow and stay within budget. Researchers at the Ehrenberg-Bass Institute call these buying situations “category entry points”: the circumstances that make someone think of a brand¹.
In a search box, the parent would have typed “kids school shoes” and done the rest of the thinking themselves. In a conversation, they describe the whole situation, which makes a prompt to an AI a category entry point written down: the situation that used to stay in a shopper’s head is now spelt out for the AI to match. A retailer can describe its shoes as hard-wearing, but parents find that out for themselves over a term of playground football and tell each other. That’s the kind of evidence an AI assistant learns from.
Research reinforces the importance of what other people say. An Ahrefs analysis of 75,000 brands found that brands mentioned more often online also tended to appear more often in Google’s AI Overviews². That’s correlation, not proof of what causes a recommendation, but it suggests that publishing more pages about school shoes is unlikely to be the whole answer.
Muck Rack found that 84% of the sources AI cites come from somewhere other than the brand, including journalism, academic research, government publications and encyclopedias³. Not all of that is customers talking, but very little of it is something a retailer can honestly write for itself.
Three reasons retailers miss the shortlist
When a retailer isn’t being recommended, there are three possible explanations:
-
The reputation gap.
Customers don’t associate the retailer with that particular situation, perhaps because the experience hasn’t earned their recommendation. Better marketing copy can’t compensate for shoes that fall apart or an unreliable returns process.
-
The evidence gap.
The retailer may genuinely sell excellent wide-fit, hard-wearing shoes, but that information is difficult for an AI assistant to find or trust. Inconsistent product descriptions, reviews that never mention what the shoes were bought for or unreliable stock and delivery information can all get in the way. Reputation gets a retailer onto the list, but the hard requirements in the prompt decide whether it stays there: An AI assistant that can’t confirm wide-fit stock or Monday delivery will simply leave it off.
-
The AI itself.
AI recommendations are far less predictable than traditional search rankings. A study of 158,000 ChatGPT answers found that the searches it runs to build an answer rarely match what Google or Bing would show for the same question⁴, and SparkToro found that asking the same question again almost never returns the same list⁵ (although the same brands often recur). Prompt wording and what the assistant remembers about each person add to the variation, so the useful measure becomes how often a retailer appears across the situations customers describe – something which visibility tools that test invented prompts can’t show.
Of the three, the reputation gap matters most over time, because it’s the slowest to change, and nothing technical can stand in for it, though the evidence gap is often where the quickest gains are.
Why gaming AI recommendations won’t last
Whenever a new way of being found appears, someone tries to game it. In the early days of SEO, it was keywords hidden in white text on a white page, and today it might be fake reviews or paid forum posts. AI isn’t immune either, as Harvard researchers showed when carefully crafted text added to a product page pushed an AI into recommending that product above others⁶.
The stronger case against gaming is that it only changes the part of the evidence a retailer controls. Almost everything else an AI assistant draws on comes from what customers actually experienced, from reviews and forum threads to the rules people give their own AI assistants and the stock and delivery promises an agent can check for itself. A trick on a product page is outnumbered by all of that, so any gain is fragile at best. At worst, it backfires, because every parent who arrives on the strength of a trick and finds that the shoes don’t live up to expectation goes on to write the review (or tell their AI assistant “not that shop”) that keeps the retailer off the next shortlist.
There’s an ethical line here too, because earning a recommendation by giving people a better experience is very different from manipulating one.
Where retailers should focus
Improving a retailer’s chances takes more than publishing better content. A review of 45 studies of generative engine optimisation found little evidence that tweaking website content helps a brand get found in the first place⁷. Retailers need to understand what customers expect, deliver an experience worth recommending and make sure AI assistants can find reliable evidence of it.
Start with the situations customers bring to a retailer, in their own words. Conducting research with customers and observing how they shop, including with AI, reveals what a retailer otherwise wouldn’t get to see. Combined with reviews, customer service contacts and conversations with store staff, it shows which situations the retailer has earned a reputation in, which it hasn’t, and which sit close enough to the ones it owns to be worth going after.
Then look across the service. Does the fitting advice work? Are returns straightforward? Can the retailer deliver when promised? These are the moments parents end up talking about, and those conversations are what an AI’s recommendations are built on.
Finally, make the evidence accessible. Describe products consistently, invite reviews that capture what customers bought them for, and make price, availability and delivery information accurate and easy for an AI assistant to check.
The challenge is that these responsibilities sit across different departments. Marketing might manage reviews, merchandising owns product information, and operations is responsible for stock and delivery. But customers, and increasingly AI assistants, judge the experience as a whole. Great service with weak evidence makes it difficult for an AI assistant to see the retailer. Strong evidence with poor service makes the retailer easy to see but difficult to recommend.
Fixing it takes customer research, service design, product, data and engineering working on the same situations together, rather than passing the problem from one team to the next. It’s a practical programme of work, not a guessing game about algorithms.
The shortlist is the new shop window
Reputation has always been built on experience and word of mouth, in the situations people associate with a brand. What’s changed is that an AI assistant now reads all of it, checks what it can and hands each customer a shortlist inside a conversation no retailer sees. Retailers left off the list won’t see the lost sales, only customers who never arrived.
If you’d like to understand why your retail brand is (or isn’t) making those shortlists, we can help you investigate the customer experience, data and technology behind the answer.
Sources
- Ehrenberg-Bass Institute: Category Entry Points Dissected: How They Really Contribute to Growth
- Ahrefs: An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied)
- Muck Rack: Generative Pulse, What Is AI Reading? (May 2026)
- Chen and Koudas: What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization
- SparkToro: New Research: AIs Are Highly Inconsistent When Recommending Brands or Products
- Kumar and Lakkaraju: Manipulating Large Language Models to Increase Product Visibility
- Martinez: Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023–2026)
About the author
Russell Beech is a Principal Consultant and service designer at Equal Experts. He helps organisations understand why their services perform the way they do, tracing what customers experience back to the people, processes, incentives and technology behind it. As AI agents take on more of that work, his focus is designing how people and AI share it, so that investment in AI turns into customers who choose, stay and recommend.
Connect with Russell on LinkedIn