Let’s start with something that’s easy to miss: Amazon’s recommendation engine is genuinely extraordinary.
The engineering behind it is world-class. The data it runs on is vast. The machine learning models that power it have been trained, refined, and optimised over two decades by some of the best technical minds in the industry. By almost any measure, it is one of the most sophisticated AI systems applied to retail anywhere on the planet.
The problem isn’t the quality of the AI. The problem is what it’s been built to optimise for. And that problem affects every search you run on Amazon, every recommendation you see, every result that appears at the top of the page.
What Amazon’s AI Actually Optimises For
Amazon’s recommendation system doesn’t primarily optimise for your satisfaction. It optimises for Amazon’s revenue.
These are related but different objectives — and the gap between them is where shoppers consistently lose out. A system optimised for your satisfaction would surface the product that best matches your need, regardless of margin, regardless of whether it’s sold by Amazon directly or a third-party seller, regardless of who has paid for placement.
That is not how the system works. Amazon’s algorithm is known to favour products sold directly by Amazon — often the same or similar products to third-party listings, but carrying higher margin for the platform. It weights heavily toward products enrolled in Amazon’s own fulfilment network. And it reserves prominent placement for sellers who have paid for it through Amazon’s advertising platform.
None of this is secret. Amazon has disclosed elements of how its algorithm works in various regulatory contexts. The point isn’t that Amazon is doing something hidden — it’s that the system is working exactly as designed, and what it’s designed for isn’t your best outcome.
💡 The Conflict of Interest
Amazon is simultaneously a marketplace that hosts third-party sellers and a retailer that competes with those same sellers. Its AI recommendation system sits in the middle of that conflict — and when Amazon’s own products compete with a third-party listing, the algorithm doesn’t operate as a neutral referee. The platform has a financial stake in the outcome.
The Data Problem Nobody Talks About
Amazon’s AI is trained on what you’ve bought before. This makes it good at predicting what you’ll buy again. It makes it significantly less good at helping you discover something new.
Purchase history is a backward-looking signal. It tells you what someone has wanted in the past. It doesn’t tell you what they need now, what they’d prefer if they knew it existed, or what they’d choose if they were presented with a genuinely complete set of options rather than a curated selection weighted toward Amazon’s commercial interests.
The result is a recommendation experience that feels personalised but is actually quite narrow. You see variations on things you’ve already bought. You see sponsored alternatives to things you’ve searched for. You rarely see something genuinely unexpected that turns out to be exactly what you needed — because the system isn’t built to find that. It’s built to close the transaction quickly and profitably.
The Sponsored Layer
Layered on top of all of this is the sponsored placement system — which we’ve written about separately, but which is impossible to separate from any honest discussion of Amazon’s AI.
The products that appear at the top of Amazon search results are not the products the AI has determined are the best match for your search. They are the products whose sellers have paid the most for that position. The AI-ranked organic results begin several rows down — by which point many shoppers have already made a choice.
When Amazon talks about its AI-powered search and recommendations, it’s describing a system that operates underneath a paid placement layer that fundamentally compromises the results. The AI might be trying to show you the best product. But the best product is rarely what you see first.
“The most sophisticated AI in retail is also one of the most constrained — by the commercial interests of the platform it serves. That’s not a technical problem. It’s a values problem.”— Justin Hodnett, Founder, ShopWithMore
What Shopper-First AI Actually Looks Like
An AI that genuinely works for the shopper starts from a different question. Not “what product will generate the most revenue from this search?” but “what does this person actually need?”
Those questions produce different answers. A shopper-first AI doesn’t weight results by margin or fulfilment network membership. It doesn’t reserve top positions for paid placements. It doesn’t surface what you bought last month as a proxy for what you want today.
It understands intent — the actual goal behind the search — and works to meet it as efficiently as possible. It’s transparent about what it is and isn’t showing you. And it has no commercial conflict of interest sitting between the recommendation and the result.
This is what ShopWithMore is building toward. V1 is our foundation — fast, lean, affiliate-driven, with no sponsored layer and no paid placement. V2 adds the conversational AI layer that makes discovery genuinely intent-driven. The AI we’re building works for the shopper because there is no structural incentive for it to work for anyone else.
Amazon’s AI is impressive. It’s also compromised by design. The next generation of ecommerce AI doesn’t have to make that trade — and we’re not making it.
See what unbiased product discovery looks like at shopwithmore.co.uk.