Being recommended by AI now depends on product feeds, product pages, PR and measurement working together. Here is what each team needs to do, based on two days of expert talks at BrightonSEO.
A UK pharmacy retailer with 232,000 organic visits a month, 1,300 referring domains and a content score of 8.8 out of 10 had an AI visibility score of 0.0.
That retailer, Weldricks, was one of the case studies Adam Preston of Summit shared at BrightonSEO in October. By the usual SEO measures it was in good shape. When AI assistants were asked for recommendations in its category, it didn't appear at all.
Our CSO, Lucy Davis, spent 8 and 9 October at BrightonSEO, attending 21 talks on AI search, measurement, ecommerce and digital PR as she builds out our AEO (answer engine optimisation) consultancy. Her main conclusion: "SEO used to be the responsibility of one person or a team, and now AEO is a cross-functional strategy that touches all teams in different ways."
This guide breaks that down into five areas, who owns each one, and what to do first.
Summit studied 250 retail websites across five industries. The team scored each site on nine technical and content signals, including schema, crawlability, headings and alt text, then compared those scores with how often each site appeared in AI answers. The on-site signals barely correlated with visibility. Brand demand, meaning how often people search for the brand by name, correlated at 0.74. And when a brand appeared in an answer to a prompt that didn't name it, 91% of the time the information came from someone else's website rather than its own.
Jean Dejsuvan, SEO Director at MG OMD, gave brands a simple test on day one: if a brand's website disappeared from the internet, would AI still have enough evidence to recommend it? For most brands the answer is no, and fixing that sits with brand and leadership as much as with SEO.
She also gave a rule for positioning. If a selling point isn't strong enough to go on the homepage, it isn't strong enough to be the thing AI remembers. Brands that describe themselves differently on each platform leave AI to guess which version is true. The ones that repeat one clear message everywhere tend to get that message repeated back.
What to do first: agree the one thing the brand wants to be known for, and check that the website, social profiles and marketplace listings all say it the same way.
Malte Landwehr, whose team analysed more than a million products across 250 million ChatGPT conversations, showed that ChatGPT Shopping runs on two different systems. In the US, OpenAI uses its own feed-based index. In Europe, it still pulls products from Google Shopping. For UK and European brands today, then, a strong presence in ChatGPT shopping results largely follows from a strong Google Shopping presence, which puts AEO inside the performance marketing team's existing feed work.
The US system gives a preview of what will matter when it reaches Europe. Every product description is cut off after 50 tokens, roughly 40 words, so the most important terms need to come first. Malte also recommended sending every size and variant as its own item in the feed.
Stijn Bergmans of Follo said most feeds he audits are around 60% complete. They include the fields Google requires and leave optional ones, such as product highlights and Q&As, empty. His slide summed up why that matters: "Required gets you listed. Optional gets you recommended."
What to do first: audit the Google Shopping feed for empty optional fields, and rewrite product descriptions so the key terms sit in the first 40 words.
Greta Koivikko of NOVOS argued that Google and AI platforms now do most of the research and comparison for the shopper. By the time someone lands on a brand's site, it is often straight onto a product page, so that page has to explain who the product is for and carry the brand on its own.
Her first point was technical. Most AI crawlers cannot read content that loads with JavaScript after the page opens. She showed a Uniqlo size guide that isn't in the page's source code at all. An AI assistant asked what size to buy can't see it, so it answers from reviews and third-party sites instead. The fix is to render important content on the server so it is part of the page from the start.
From there, she recommended adding sections that answer real-life questions, such as "perfect for", "when to use" and "do not use with". Storing product attributes as structured data (Shopify metafields, in her examples) means the same information can flow into both the product page and the product feed. That matters because Google Shopping cross-checks a product page against the feed and retailer listings, and mismatched details count against the brand.
Adam Preston added a quick win for category pages. In one example, John Lewis was cited by AI for "best modern dining chairs" with a page that had very little copy, because its filter pages could be crawled and indexed. Competitors' filter pages could not.
What to do first: check whether size guides, reviews and product details are visible in each product page's source code, and make key filter and collection pages indexable.
Amy Gibson of Digitaloft showed that editorial "best of" roundups, in titles such as The Guardian, Which? and Expert Reviews, are among the sources AI answers rely on most for product questions. Product PR has often been dismissed by SEO teams because the links tend to be affiliate or nofollow. Amy's view is that those roundups are now a direct route to being recommended.
Her clearest example was sleep brand Emma, whose weighted blanket was named best overall in The Guardian's roundup. That one article generated more than £10,000 in affiliate revenue. Over 12 months, Emma's product PR produced 159 pieces of coverage and a 167% increase in AI referral traffic.
Her advice was to build media lists from AI answers themselves, by finding the sources AI cites for a brand's key prompts and pitching the journalists behind them. She also stressed that a brand can't be recommended as the best if it is the only one saying so. What the brand claims and what independent sources say have to match.
What to do first: list the publications AI cites for the brand's top product categories, and check whether the brand appears in them.
A theme across the first day was that mention rate, the headline number on most AI visibility dashboards, is close to a vanity metric. Several speakers argued that how a brand is mentioned, and whether it is actually recommended rather than just cited as a source, matters far more.
Standard analytics also miss most of the effect. Stijn Bergmans said LLMs account for between 0 and 1% of tracked traffic and conversions across his agency's clients, yet when he asks clients how they found the agency, roughly half say it came up in an AI conversation. People research in an AI assistant, then search for the brand by name or type in the URL later. His answer is to track those follow-up actions, branded search and direct visits in particular, and to ask customers where they heard about the brand.
What to do first: add branded search and direct traffic to AI search reporting, and ask new customers where they heard about the brand.
The BrightonSEO exhibition hall was full of AI visibility tools, but a dashboard on its own doesn't tell a brand which of these five areas to fix first. A brand like Weldricks needs fame more than technical fixes. A well-known brand with thin product data needs the feed and product pages fixed before anything else. Jean Dejsuvan suggested asking "what is the biggest thing holding us back right now?" rather than "what should we do next?" Answering that well takes someone who understands the brand's margins and customers.
Timing matters too. OpenAI's own shopping system is US-only for now, and so is Google's checkout inside AI Mode and Gemini. Given the pace of AI rollouts so far, Lucy thinks it is only a matter of time before Google's checkout reaches the UK and Europe.
AI search touches brand, performance marketing, engineering, PR and analytics at once. As Tim Richardson, our CEO, has argued in Offcuts, brands can't switch it off, and work that belongs to every team usually ends up falling between them.
That is why Lucy's second takeaway matters: "having an agency partner who gets all of it, not just some of it, is a competitive advantage." Most agencies are built around one discipline. We connect engineering, design, performance marketing and AI into a single commercial operation. For AI search, that puts two of these five areas squarely in our work: the product feed, through performance marketing, and product pages machines can read, through engineering and design. Brand, PR and measurement shape how far that work goes, so we plan it alongside the teams who own them.
Ranking well on Google still matters. It no longer guarantees a place in the answer.
