The search bar is losing its monopoly on how people find things to buy. Instead of typing "best running shoes," scanning a page of links, and clicking through review sites and product pages, shoppers increasingly just ask an AI assistant — "what are the best trail running shoes under £150 for flat feet?" — and get a short, reasoned recommendation back. Around half of consumers already use AI search, and among those who do, a striking share now call it their primary source of insight for buying decisions, ahead of traditional search, retailer sites, and review sites. McKinsey projects roughly $750 billion of US consumer spend will flow through AI-powered search by 2028.
For anyone selling products online, that's not a minor channel shift — it's a change in who your customer even talks to first, and, increasingly, in what reads your product information. Because the newest wrinkle is that the shopper isn't always a person anymore. Let's unpack what's actually happening and what to do about it.
From ten blue links to one recommended answer
Traditional search hands you a list and makes you do the work: open tabs, compare specs, read reviews, reconcile contradictions, decide. AI search collapses all of that into a single conversational exchange. You describe your need — including budget, use case, and constraints — and the AI does the comparison across many sources and hands back a curated shortlist, often with reasons attached.
That's a profound change in the shape of discovery. The evaluation that used to happen across a dozen of your carefully optimised pages now happens inside the AI's answer, before the shopper ever clicks. If your product makes the shortlist, you're in the running; if it doesn't, you're invisible — there's no page two to languish on, just an answer you're absent from. Ranking is being replaced by being recommended.
The new shelf There's no page two in an AI answer. You're either in the recommendation, or you don't exist for that shopper.
The shopper might be a machine now
Through 2026, discovery has gone a step further into agentic commerce: AI agents that don't just recommend but research, compare, and increasingly buy on the shopper's behalf. Someone tells an assistant "find me a sustainable sneaker under £150 with good arch support," and the agent works through the options and either presents the best match or completes the purchase. The major AI shopping surfaces — ChatGPT's shopping features, Google's AI Mode, Perplexity, and others — are racing to enable exactly this, with the emerging consensus that discovery happens inside the AI while checkout is handed back to the merchant. For merchants, that raises a new bar — being "agent-ready," with catalogues an agent can query reliably through the commerce protocols now taking shape, complete with accurate identifiers and real-time price and stock. A store an agent can't cleanly read is a store it can't recommend, let alone buy from.
Two things make this more than a novelty. First, these shoppers convert: Adobe reported in early 2026 that AI-referred traffic began converting meaningfully better than non-AI traffic — a reversal from a year earlier — because people arriving from an AI recommendation are already deep into a decision. Second, when an agent is the one comparing products, your listing has to satisfy a machine's need for clean, structured, comparable data. The primary consumer of your product information is no longer only a human browsing; it's often an algorithm parsing.
Why your brand might not show up
Here's the uncomfortable part, and the reason even big brands are getting caught out: AI search largely isn't reading your website. Analysis suggests a brand's own site accounts for only around 5 to 10 percent of the sources AI-powered search draws on. The rest comes from reviews, user-generated content, publishers, forums, and affiliate sites — the wider web's conversation about you.
Only about 5–10% of the sources AI-powered search references are a brand's own site. The majority are third-party: reviews, UGC, publishers, communities, and affiliates.
The implication: a polished website and strong brand are no longer enough. If the web around you is thin — few reviews, little third-party coverage, no structured data to parse — the AI has little to go on, and it recommends someone else.
This is why traditional brand strength is no guarantee of AI visibility, and why a great-looking store can still be absent from the answers where decisions are now made. It also reframes the job: influencing AI recommendations is less about your homepage and more about the entire ecosystem of information about your products — which makes it a core part of modern search visibility work, and sits at the heart of generative engine optimisation and the broader upheaval we track in how AI Overviews are reshaping SEO.
How AI finds and picks your products
So what actually gets a product into the recommendation? A handful of signals, most of which reward substance over tricks.
| Signal | Why it matters to AI |
|---|---|
| Structured data (schema) | Product, Offer, and Review markup with identifiers like GTINs makes your listing machine-readable and comparable — the single biggest lever for being cited and selected |
| Accurate product feeds | Complete, current feeds (price, availability, attributes) are how agents access your catalogue reliably |
| Rich, precise specifications | AI matches on details — size, material, use case — so vague copy loses to specific copy |
| Genuine reviews & ratings | A major trust and relevance signal, and part of the third-party sources AI weights heavily |
| Off-site reputation | Mentions and coverage on the publishers, communities, and affiliates the AI actually reads |
| Buyer-question content | "Best for" and comparison content that matches how people actually ask AI |
The through-line is that structured, accurate, well-reviewed products win. Industry testing consistently finds that listings with complete structured data are cited and selected by AI far more often than those without — the machine can only recommend what it can cleanly read and verify. Getting the fundamentals of your product pages right, and building a site that machines can parse as covered in designing websites for AI search, is now table stakes rather than a nice-to-have.
The playbook: be the answer, not just a result
Pulling it together into what to actually do.
- Make your data machine-readable. Implement complete Product, Offer, and Review schema with proper identifiers, and keep an accurate, complete product feed. This is the foundation everything else sits on.
- Win off-site. Since most of what AI reads isn't your site, invest in genuine reviews and in being covered by the publishers, communities, and creators the AI references — the same shift toward trusted independent voices we explored in why everyday creators outperform celebrity endorsements.
- Answer real buyer questions. Create comparison and "best-for" content that mirrors how people actually query AI, so your expertise is part of the source material.
- Build the brand. AI tends to recommend brands that are well-established across its sources; brand-building and PR now feed discovery directly.
- Be present across surfaces. Discovery is fragmenting across AI assistants, marketplaces, and social — the same multi-surface reality driving social commerce in 2026. Show up where your shoppers ask.
Measure what you can't see the old way
The catch with all of this is that the decision now happens where you can't watch it. When a shopper gets a recommendation inside an AI answer and never visits your site before buying, your familiar analytics go quiet — and yet the influence was real. The sale still happened; you simply didn't get to watch the shop window where the decision was made, which is what makes the old dashboards feel misleadingly quiet. It's telling that only a small minority of brands systematically track their AI search performance at all.
The answer is a new measurement layer: track your visibility and sentiment inside AI answers (are you recommended, and described well?), your share of voice against competitors across the major AI platforms, and the quality of AI-referred traffic when it does arrive. This is the same reckoning with a click-less journey we cover in why zero-click content is forcing marketers to rethink success metrics, and it compounds the broader difficulty explored in why attribution is getting harder. You can't optimise what you don't measure, and the brands that build this visibility tracking now will have a real head start.
The bottom line
Product discovery is moving off the results page and into AI assistants and agents that research, compare, recommend, and increasingly buy — and the primary reader of your product information is now as likely to be a machine as a person. That flips the goal from ranking on page one to being the recommended, machine-selectable answer, and it depends far less on your own website than on the structured data, accurate feeds, genuine reviews, off-site reputation, and brand authority that AI actually references. Feed the machines clean, complete, trustworthy information; earn a strong reputation across the wider web; and measure your presence inside the answers themselves. The shelf shoppers browse is increasingly an AI's recommendation — make sure your products are on it.
Make sure AI recommends your products — not your competitor's.
We optimise your product data, structured markup, and off-site presence so your store gets discovered in the age of AI search.
Explore E-commerce Marketing →Frequently asked questions
What is AI product discovery?
It's when shoppers use AI assistants — like ChatGPT, Gemini, Google's AI Mode, or Perplexity — to find products by describing what they need in natural language, and the AI researches, compares, and recommends specific options rather than returning a page of links. Instead of scanning results and review sites themselves, the shopper gets a curated shortlist or a single recommendation, and increasingly an AI agent can even complete the purchase.
Why doesn't my brand show up in AI search?
Because AI search doesn't rely mainly on your website. Research suggests a brand's own site makes up only around 5 to 10 percent of the sources AI-powered search draws on; the rest comes from reviews, user-generated content, publishers, and affiliates. So a strong brand with a great website can still be missing from AI answers if it lacks structured product data, accurate feeds, and a solid off-site reputation for the AI to reference and trust.
What is agentic commerce?
It's when an autonomous AI agent handles shopping on a person's behalf — you give it a goal like "find running shoes under £150 for flat feet" and it researches options, compares them, and either recommends the best match or completes the purchase. In the current model, discovery happens inside the AI while checkout is handed back to the merchant, connected by emerging commerce protocols. The practical result is that your product data increasingly has to satisfy a machine, not just a human.
How do I optimize products for AI search?
Make your product information machine-readable and trustworthy. Add complete structured data (Product, Offer, and Review schema with identifiers like GTINs), keep an accurate and complete product feed, provide rich and precise specifications, earn genuine reviews and off-site coverage, and publish content that answers real buyer questions such as comparisons and best-for guides. Products with complete structured data are markedly more likely to be cited and selected by AI.
Is traditional SEO still worth doing?
Yes, but it's no longer sufficient alone. Traditional search still drives meaningful traffic, and many of AI search's building blocks — quality content, structured data, reviews, authority — overlap with good SEO. The change is that you now also have to optimise to be surfaced and selected inside AI answers, a discipline often called generative engine optimisation. Think of it as expanding your remit from ranking on a results page to being the recommended answer across both.
