For most of retail history, a price tag did two jobs. The obvious one was telling you what something cost. The less obvious one was a quiet social contract: whatever you were being asked to pay, the person behind you in the queue was being asked to pay too. You could hate the price. But you hated it together. (Shared suffering is the foundation of most functioning societies, and apparently also of grocery pricing.)

AI is now loosening that second job, often without touching the number at all. That's the bit most of us in ecommerce are underestimating. The headlines keep asking whether retailers are charging different people different prices. The bigger near-term problem is that shoppers increasingly suspect they might be, and they have absolutely no way to check.

Dynamic pricing is the familiar one: prices moving with demand, inventory and competitors. Anyone who has run an Amazon repricer knows the drill. It doesn't automatically mean prices go up, either. One web data company tracking US price changes in 2025 found that roughly half were reductions. (A statistic that will never, ever trend on TikTok.)

Electronic shelf labels bring this offline. Walmart had them in around 2,300 stores by March and plans to roll them across its entire US estate within a year. But most chains still update them on a fixed weekly cadence, and researchers studying one grocery chain after installation found almost no evidence of surge pricing. (The robots, it turns out, mostly work Tuesdays.)

Personalised pricing, which regulators increasingly call surveillance pricing, is the one driving the anger. This is using what's known about you (location, device, browsing, purchase history, inferred income) to estimate what you're willing to pay. An FTC study published in January 2025 found that intermediaries selling these tools worked with at least 250 clients, and documented signals as granular as mouse movements and abandoned basket items being used to shape offers. (Somewhere out there is a dashboard tracking exactly how long you hovered over that air fryer before chickening out.)

AI sits underneath all three, making it far cheaper to weigh hundreds of variables at once.

When Consumer Reports and Groundwork Collaborative had more than 400 shoppers fill identical Instacart baskets at the same moment, around three-quarters of products showed different prices, with gaps of up to 23%. Instacart said these were randomised tests assigning shoppers to groups rather than pricing on personal data, and it ended them. A similar exercise with Uber and Lyft found a median gap of 42% on identical routes; both said base fares aren't personalised.

Every one of those explanations was plausible. And every one of them did very little to make anyone feel better.

Behavioural economics worked this out back in the 1980s. People don't judge a price as fair on the number alone; they judge it on the reason they believe sits behind it. A retailer passing on its own rising costs? Fine. A retailer charging more because it's detected you need something urgently? Exploitation, even if the final price is identical.

AI pricing makes the motive invisible. And when people can't see why a decision was made, they reliably fill the gap with the least charitable explanation available. Which is how social media ended up convinced that Walmart's new shelf labels scan faces, and how Walmart's leadership ended up spending much of October insisting it prices "the product, not the person." The labels, the company has confirmed, contain no cameras, no microphones and collect no customer data. (There's something quietly poignant about one of the largest companies on earth issuing public statements clarifying that its price stickers cannot see you.)

The real damage is to the reference price, the internal anchor shoppers use to judge a deal. Once someone believes the number on screen was generated just for them, a discount stops reading as a discount. So shoppers wait, the same way ride-hailing users learned to wait out surges and department store regulars learned never to pay without a coupon. (Congratulations to every retailer who spent a decade training customers never to pay full price. The algorithm has inherited your life's work.)

Then there's the twist nobody wants on a slide. A University of Chicago study from 2023 found personalised pricing raised expected profits by 19% while leaving more than 60% of consumers better off, mostly because targeted discounts pulled in shoppers who'd otherwise have walked. Follow that through and the shoppers most protective of their data see only list prices, which in a personalised system tend to be the highest ones. Privacy starts carrying a price tag of its own, and it lands on exactly the customers paying the closest attention. (Read the terms and conditions, they said. Be an informed consumer, they said.)

Assistants like Rufus and Walmart's Sparky work best when shoppers tell them things: what they need, what they're worried about, how soon they need it, what they're prepared to spend. Every one of those disclosures is also, in principle, a willingness-to-pay signal. Walmart has explicitly said that telling Sparky you need something urgently won't be used to raise your price. Which is reassuring. The fact that it needed saying is the more interesting detail.

Conversational commerce runs on candour. If shoppers start suspecting that what they tell an assistant could shape what they pay, the rational response is to say less, or to say it strategically. (Picture millions of people haggling with a chatbot like it's a market trader. "The hiking boots? No rush. Just browsing. Honestly couldn't care less if I get them this year.") That would degrade recommendation quality at precisely the moment brands are rebuilding their listings around how these assistants retrieve and rank products. The trust question and the effectiveness question may turn out to be the same question.

The FTC study also flagged something subtler. In one example, a shopper identified as a new parent was shown pricier baby thermometers in search results. Nobody was charged a different price for the same product. The selection changed instead. As discovery moves into assistants that decide which handful of products to surface, personalising the assortment may prove both more powerful and much harder to detect than personalising the price.

Prices have always moved, and loyalty schemes have tailored effective prices for years. The genuine fault lines are transparency, consent and verifiability. Target says it doesn't use identity or location to raise prices, while its Circle loyalty programme tailors promotions. Personalised promotion, not personalised price. Whether consumers and regulators accept that distinction is wide open, because from the shopper's side of the screen both can produce the same outcome: two people paying different amounts for the same thing at the same moment.

The second fault line is competition. The FTC's case against Amazon includes allegations about Project Nessie, a tool the agency says identified products where Amazon could raise prices confident competitors would follow, generating more than $1 billion in excess profit. Amazon says that grossly mischaracterises a tool built to stop price-matching spirals, which it discontinued before the lawsuit. Whatever the outcome, the broader concern stands: when lots of sellers run similar algorithms on similar data, prices can drift upwards together without anyone agreeing to anything. (No collusion required. Just a lot of spreadsheets independently reaching the same conclusion at the same time.)

Regulators, meanwhile, are moving at their customary pace and in their customary number of directions. The FTC warned in August that undisclosed personal pricing could be deceptive, while admitting it can't ban it. New York now requires disclosure, and legislators in at least 30 states considered pricing bills this year. (Fifty states, fifty opinions, one exhausted compliance team.)

For brands selling first-party to Amazon, Walmart or Target, the most direct exposure isn't reputational. It's margin.

The retailer owns the shelf price, and as pricing automates, so does price matching. A competitor's promotion on one site can now be matched across several others within hours, and shelf labels extend that speed into physical stores. Retailers that match down rarely absorb the cost quietly; Amazon's vendor model has long relied on suppliers funding shortfalls through margin support, promotional funding or chargebacks. Faster matching means more of those conversations, with more at stake in each. (Anyone who has sat through a margin support negotiation will appreciate that "more often" is not the selling point it might sound like.)

Channel pricing gaps get exposed too. A targeted discount on your own DTC site that a crawler picks up could cost you the Amazon Buy Box, on the marketplace where most of your volume lives.

Then there's promotional spend. Retailers increasingly let algorithms decide which shoppers get which brand-funded offers, which can reach new buyers or hand discounts to loyal customers who'd have paid full price anyway. Brands typically can't see which. Paying for promotions you can't audit is, in effect, outsourcing part of your margin decisions to someone else's model.

And if a retailer gets accused of personalising prices, the viral screenshot will show a product. Someone's branded product. (Your logo. Their algorithm. Everyone's outrage.)

Brands running their own sites need to decide, explicitly, whether they personalise prices, promotions, assortment or none of the above, because regulators and journalists will ask. Post-Instacart, even routine undisclosed price testing carries reputational risk.

For brands selling through retailers, the harder problem is information. How often your products are matched down, and who your promotions actually reach, sits inside the retailer's systems. Whether brands can win real visibility there, or the asymmetry simply widens, could become one of retail's defining commercial tensions.

A stable, shared price was always partly a trust mechanism, and AI is wearing it down whether or not any individual company personalises anything. Some brands will decide the efficiency is worth it. Others may find a visibly consistent price becomes a differentiator in its own right. Either way, the price on the screen isn't being read the way it used to be, whether your pricing has changed or not.

🤖 Top GPTs - Gain access to the leading OpenAI GPT models without needing a Plus subscription.

📧 Inbox Zero - Open-sourced AI email cleanup management

📂 Dewstack AI - Effortlessly craft and manage AI-powered docs that elevate your content and empower your users with instant answers.

📝 Superlist - Create to-do lists, capture thoughts or detailed notes, assign tasks to teammates, and more

😍It was great, all I needed

🙂Good, not great

🙄It sucked