When someone asks ChatGPT, Claude or Perplexity to recommend a product, a cleaning spray, a vitamin supplement, a bottle of wine, what actually shapes the answer? Is it your brand campaign? Your influencer partnerships? Your website copy?
We wanted to know, so we tested it properly. We looked at 125 everyday UK products across five categories and interrogated AI systems to find out what they actually lean on when making a recommendation. The answer was clear, consistent, and a little uncomfortable for anyone who's spent years focusing on traditional marketing.
Retailer reviews are the single most powerful trust signal in AI-led product discovery. Not editorial coverage. Not brand messaging. Not expert endorsements. Reviews (the ones left by ordinary people who actually bought your product and used it) are what AI relies on most when it decides what to recommend.
125 - UK products analysed across 5 categories
#1 - ranking signal: retailer reviews
72% - UK consumers adopted AI chatbots for shopping by 2025
Why We Did This Research
The way people discover products is changing. For years, the path to purchase ran through search engines, brand websites, and comparison pages. Increasingly, it's running through AI assistants. Someone types a question into ChatGPT and gets a shortlist. Someone asks Perplexity to compare two supplements and gets a recommendation with reasoning behind it. Someone uses Claude as a shopping adviser and trusts what it says.
If that shift is real, and the data says it is, with AI chatbot use for shopping growing 72% between 2023 and 2025 in the UK, then the question of what influences AI recommendations becomes one of the most commercially important questions a brand can ask.
This research, conducted in partnership with Brand Allies, set out to answer it. Not theoretically. Practically, using real products that real people buy.
How We Ran the Study
We chose 125 mainstream UK retail products across five everyday categories: household cleaning, alcohol, no and low alcohol, cosmetics, and health supplements. These are the kinds of products you'd find on the shelves at Boots, Sainsbury's, or Amazon, not niche launches or experimental products, but things people genuinely buy every week.
For each product, we recorded what kind of information was available to AI systems: whether retailer reviews existed and in what volume, how recent those reviews were, what Amazon UK review coverage looked like, how much UK editorial coverage the product had, and whether any institutional or authority sources referenced it.
Then we interrogated AI systems, asking them to recommend products, justify their reasoning, and explain what shifted when we changed the available signals. By challenging AI on its own logic and watching substitution patterns, what it replaced one brand with when evidence was thin, we could build a clear picture of the hierarchy it uses.
The Hierarchy AI Uses to Make Decisions
The research surfaced a consistent, stable order of importance across all five categories. Some signals were almost always decisive. Others barely moved the needle at all.
- Retailer Reviews had an impact score of 100 and is the primary trust anchor across all categories
- Editorial Reviews had an impact score of 72 and is a strong factor when reviews are thin then deprioritised when reviews are dense
- Community Discusion had an impact score of 45 which included forums, Reddit, and social. All useful for corroboration
- Brand Longevity / history had an impact score of 38, and provides context but rarely drives decisions alone
- Authority or institutional references had an impact score of 25, however this has a high weight in health categories and conditional elsewhere
- Brand Messaging has an impact score of 8 and has almost no evaluative weight, so it explains elements of the product, service or brand but doesn't necessarily recommend it
The gap between the top signal (retailer reviews, scoring 100) and the bottom signal (brand messaging, scoring 8) is not a small difference. It's almost total. And yet most brand marketing budgets are stacked towards the signals that matter least to AI.
Why Retailer Reviews Dominate Everything Else
This is the part that catches people off guard. Retailer reviews aren't dominating AI recommendations because they're more eloquent than editorial copy, or more trustworthy than expert opinion. They dominate because of their structure.
When AI has to make a recommendation, it's essentially trying to reduce uncertainty. What's the most reliable answer I can give? What's least likely to be wrong? Retailer reviews tick a very specific set of boxes that help AI answer that question with confidence.
They're tightly connected to actual purchase and use, not what someone hoped a product would do, but what it did. They aggregate outcomes across lots of different people, not just one expert's experience. They tend to be low on persuasive intent, most people leaving a review aren't trying to sell you anything. And they're consistent across different contexts and user types, which makes them a stable signal rather than a noisy one.
"Marketing activity that does not translate into dense, stable feedback has diminishing impact on AI recommendations."
When review signals are strong and recent, AI largely ignores the other signals. It has what it needs. Editorial content gets used to add colour. Brand messaging might explain what a product does. But the recommendation itself comes from the reviews.
What Happens When Reviews Are Thin or Missing
This is where it gets genuinely alarming for some brands. When retailer review signals are weak or absent, AI doesn't pause. It doesn't hedge. It doesn't say "I'm not sure about this one." It reaches for whatever it can find and substitutes with confidence, often replacing a brand entirely with a better-evidenced competitor, without signalling to the user that it's made a substitution.
A product with very few retailer reviews doesn't get a cautious, balanced response. It gets marginalised, replaced, or ignored and the person asking the question never knows it happened. From their perspective, AI gave them a great answer. From your brand's perspective, your customer just got sent to a competitor.
The substitution problem: Absence is riskier than negativity. A product with some mixed reviews is anchored in AI's understanding of that category. A product with no reviews simply doesn't register, and AI fills the gap with something else.
The Role of Authority Signals (It's Not What You'd Expect)
Expert endorsements, clinical studies, and regulatory approval do matter to AI, but they matter differently depending on the category, and almost never on their own.
In health supplements, institutional references carry significant weight. When outcomes are harder to verify from customer reviews alone, when the question is whether something is safe, not just whether it tastes nice, AI reaches for authority references to provide legitimacy. But even here, authority references work in conjunction with post-purchase feedback. They rarely carry a recommendation by themselves.
In household cleaning or alcohol, authority signals have far less impact. There's no safety ambiguity for most people choosing a surface cleaner. Reviews are plentiful and directly useful. Authority references get largely skipped.
The implication for brands investing heavily in expert endorsement programmes: they may be building authority signals that AI barely uses in your category.
Brand Messaging: The Signal That Barely Works
If there's one finding from this research that brands need to sit with, it's this: what you say about yourself has almost no evaluative weight in AI-led product discovery. Brand-owned content, your website copy, your social captions, your brand story, supports explanation. It helps AI describe what your product is or does. But it doesn't support recommendation. AI doesn't trust brands to objectively assess themselves, and it's right not to.
This is a significant shift from traditional search engine optimisation, where well-written brand content could drive meaningful visibility. In AI-mediated discovery, the signals that drive recommendations all sit outside the brand's own voice. Reviews written by customers. Articles written by journalists. Discussions had by real users. These are the things AI uses to form a view, and none of them are things you can write yourself.
What This Means Across the Five Categories We Studied
The broad pattern held across all five categories, but with some useful nuances worth knowing.
In household cleaning, review volume was the dominant signal with almost nothing else competing. In alcohol and no/low alcohol, editorial coverage from drinks publications provided meaningful corroboration alongside retailer reviews. In cosmetics, community discussion, forums, Reddit, YouTube commentary, carried more weight than in other categories, likely because product performance is so subjective and experiential. In health supplements, the combination of retailer reviews and institutional references was decisive, with neither working as well alone.
Across all five, the brands that dominated AI recommendations were the ones with dense, recent, multi-platform review coverage. Every time.
Six Things This Research Changes for Brands
Trust is built after the sale, not before it. The signals AI trusts include most reviews and they are created by customers who have already bought your product. Pre-purchase marketing shapes awareness but doesn't directly build the evidence AI uses to recommend. That means post-purchase experience is marketing, whether you treat it that way or not.
Review health is strategic infrastructure. The volume, recency, and consistency of your retailer reviews determines whether AI anchors its recommendations to you, or reaches for a competitor instead. This is not a hygiene task. It belongs in your commercial strategy.
Absence is genuinely dangerous. A product with few or no retailer reviews isn't treated with caution by AI. It gets replaced with confidence. Your customers get a confident recommendation for someone else's product.
Authority signals need pairing. In most categories, expert endorsements and clinical credibility only strengthen recommendations when they sit alongside real customer feedback. Authority alone rarely carries sufficient weight.
Brand messaging has limited reach. It supports explanation, not recommendation. What you say about yourself shapes how AI describes your product, but not whether it recommends it.
Competitive advantage is shifting. As AI-mediated discovery grows, advantage moves away from who tells the most compelling brand story and towards who has generated the most reliable, consistent evidence of real-world performance. That's a fundamentally different kind of marketing.
Frequently Asked Questions
Do AI platforms actually look at retailer reviews like Amazon or Tesco?
AI systems are trained on large amounts of publicly available internet content, which includes retailer review pages. The density, recency, and consistency of those reviews are key factors in how well AI can form a confident recommendation. Products with hundreds of recent positive reviews across multiple retailer platforms are far more likely to be recommended than products with sparse or outdated coverage.
Does this apply to B2B brands, or just consumer products?
This research focused on FMCG consumer products, but the underlying logic, that AI looks for independent, consistent evidence of real-world outcomes, applies across categories. For B2B, the equivalent signals would be platforms like G2, Trustpilot, and Clutch, as well as case study coverage and industry editorial. The principle is the same: evidence that sits outside the brand's own voice.
Our brand has great editorial coverage. Why might we still be invisible in AI search?
Editorial coverage is the second most influential signal we found, but it's consistently deprioritised when retailer review coverage is strong for your competitors. If the brands you're competing with have dense, recent reviews and you don't, AI will lean on their reviews and your editorial may not be enough to compensate. Building review coverage is the more direct lever to pull.
Is this research relevant to Google AI Overviews as well as ChatGPT and Claude?
The signals AI models use to assess product credibility are fairly consistent across platforms, because they're all trying to solve the same problem: how to recommend something reliably. Google's AI Overviews draw on similar structural signals. The specific weighting may vary by platform, but the broad hierarchy, reviews first, editorial second, brand content last, holds across the major systems.
How quickly can a brand improve its AI visibility?
Brands that invest deliberately in building the right evidence base can see meaningful shifts in AI visibility over a 6 to 12 month period. The key variables are review volume, recency, and cross-platform consistency. Starting now matters, the brands that are already well-represented accumulate a compounding advantage every week.
The Bigger Picture
Product discovery is moving into conversational environments. When a UK consumer opens an AI assistant and asks what to buy, the answer they get is shaped by signals that most marketing teams haven't historically tracked or managed. The brands that understand this early (and start building the right kind of digital evidence deliberately) are the ones that will be recommended. The ones that don't will increasingly find their customers being sent somewhere else, by an AI that's completely confident it's giving good advice.
The signals that drive AI recommendations aren't glamorous. Reviews aren't exciting to talk about in a board meeting. But they are the infrastructure that determines whether your brand shows up or disappears in the fastest-growing discovery channel in UK retail.
