Is price testing legal? A/B testing prices in the US, EU and UK (2026)

Random A/B price tests are generally legal for software in the US, EU and UK. Here are the disclosure rules, 2026 state laws and trust practices that apply.

Short answer

Yes, in general. Showing randomly assigned prices to different visitors is legal for software in the US, EU and UK, as long as every visitor sees an honest, complete price and can buy at it. The rules get stricter when personal data sets the price: the EU's disclosure duty, GDPR, and new US state laws such as New York's.

  • Random assignment is not personalized pricing: nothing about the visitor decides the price, which keeps a test outside most personalization rules.
  • Since May 2022, EU law has required traders to tell consumers when a price was personalized through automated decision-making.
  • New York has required a notice on prices set by an algorithm using personal data since November 10, 2025; Connecticut's similar rule starts July 1, 2027.
  • The UK's first consumer-law fine under its new powers, £4.2 million in April 2026, was for a £3 fee missing from an upfront price.
  • Amazon's 2000 DVD price test was random and still ended in refunds to 6,896 customers, so plan the trust side before you launch.

We found no law in the US, the EU or the UK that bans showing two visitors two different, randomly chosen prices for the same software plan. Regulators have aimed elsewhere, mainly at hidden fees and at prices built from personal data. The UK's first consumer-law fine under its new powers, in April 2026, was £4.2 million for the AA, over a £3 booking fee that was missing from the upfront price.

So the useful questions are narrower. Which kind of price test are you running, and which honesty rules apply to every price in it? This article explains the rules as they stand on November 17, 2026, and it is not legal advice.

What's the difference between price testing, dynamic pricing and personalized pricing?

The difference is what decides the price. In a random price test, chance decides: each new visitor gets one of a few prices and keeps it. In dynamic pricing, market conditions decide, and everyone shopping at that moment sees the same number. In personalized pricing, data about the person decides.

Random price testDynamic pricingPersonalized pricing
What sets the priceA random drawDemand, time, stockData about the person: location, device, history, predicted willingness to pay
Two people at the same momentCan differ, by chanceSame priceCan differ, by design
Same person on the next visitSame priceMay changeMay change
Typical exampleHalf of new visitors see $29, half see $39Airline seats, event ticketsA higher price for a visitor on an expensive phone
Extra legal hooksConsumer protection basicsSector rules, such as ticketingDisclosure laws, GDPR, US state laws

The law cares about the input. A test stays a test while nothing about the visitor (country, device, traffic source, purchase history) feeds the choice of price. Let any of those in and you move toward personalized pricing, where the extra rules below start to apply.

A multi-armed bandit doesn't change this. It shifts how many visitors see each price based on total results, and each visitor's price is still a random draw.

Yes. A 2022 study for the European Parliament concluded that even personalized pricing "is, in principle, allowed under EU law" as long as traders respect anti-discrimination law. Random tests are the milder case.

Three sets of rules apply today, and a fourth is on the way. Two of them protect consumers, whom the Consumer Rights Directive defines as people acting "for purposes which are outside" their trade or profession. If every customer buys for a business, those consumer rules mostly fall away. GDPR does not.

The personalized-price notice

Since 28 May 2022, Article 6(1)(ea) of the Consumer Rights Directive has required traders to tell consumers, before a distance contract, "where applicable, that the price was personalised on the basis of automated decision-making". The Omnibus Directive (EU) 2019/2161 added it.

Recital 45 of the Omnibus Directive describes the target: prices personalized for consumers or groups of consumers through "profiling of consumer behaviour allowing traders to assess the consumer's purchasing power". The same recital says the duty does not cover dynamic or real-time pricing that follows market demand without personalization.

A random test fits that description poorly. The price comes from a random number, not from anyone's behavior or purchasing power, so there is no personalization for the notice to report. Two caveats stop this from being a guarantee. The directive never defines "automated decision-making", as the Parliament study points out, and we found no court ruling on price A/B tests. And once a visitor attribute feeds the choice, the recital's description starts to fit, so disclose.

Honest prices in every variant

The Unfair Commercial Practices Directive treats deceptive information about "the price or the manner in which the price is calculated" as misleading (Article 6(1)(d)), and an invitation to purchase must give "the price inclusive of taxes" (Article 7(4)(c)). Its Annex I bans some practices in all circumstances, including bait advertising, bait and switch, and false claims that a price is available only for a very limited time.

In a price test, that means every variant must be a real, complete price someone can pay: tax included for consumers, no "was €49" anchor that never existed, no countdown timer that quietly resets.

A random price doesn't evaluate anyone, so it isn't "profiling" in the GDPR sense, which means using personal data "to evaluate certain personal aspects relating to a natural person". Article 22's right not to be subject to decisions "based solely on automated processing" is aimed at the personalized end, and even there the Parliament study calls its application to pricing "highly controversial".

GDPR still covers the identifier that keeps a visitor on one price, because personal data includes "an online identifier". You need a lawful basis for it and a line in your privacy notice.

The cookie itself falls under Article 5(3) of the ePrivacy Directive, which requires consent for storing information on a device unless it is "strictly necessary" for a service the user explicitly asked for. Whether a price-assignment cookie qualifies is a question for your counsel. If the answer is consent, show visitors who decline your standard price and leave them out of the test.

What would the Digital Fairness Act change?

On 11 November 2026 the European Commission proposed the Digital Fairness Act. Its questions and answers say the act would turn the Unfair Commercial Practices Directive into a regulation that applies directly in every member state, would "effectively ban drip pricing", and would stop companies "presenting higher personalised prices to minors or vulnerable consumers".

It is still a proposal that the European Parliament and the Council have to agree on. Random tests aren't its target, but its rules on full prices and price comparisons would apply to every variant you show.

Yes. We found no federal law that bans charging different customers different prices for software, and the FTC's guide to price discrimination says "price discriminations are generally lawful". The federal rules that matter are about deception, and about competition when physical goods are sold to businesses.

Section 5 of the FTC Act declares "unfair or deceptive acts or practices in or affecting commerce" unlawful. In a price test, the deceptive versions are easy to name: advertising a price people can't get, charging more at checkout than the page showed, or labeling a test price as a discount from a price nobody paid.

The Robinson-Patman Act sounds relevant and rarely is. It targets sellers who charge competing buyers different prices for the same commodity. The same FTC guide notes that it "applies to commodities, but not to services" and requires "the reasonable possibility of injury to competition". Your subscribers don't compete with each other for resale. If you sell physical goods to resellers, that is a conversation for a lawyer.

The FTC has also shown interest in pricing built from personal data. In January 2025 its staff published initial findings on intermediaries whose tools set individualized prices based on "location, demographics, browsing patterns and shopping history". A random test uses none of those. Separately, the FTC's fees rule, in force since May 12, 2025, requires upfront total prices but covers only live-event tickets and short-term lodging.

Which state laws target algorithmic pricing?

The new rules are in the states, and they target personal data.

State and lawStatus on November 17, 2026What it covers
New York, Algorithmic Pricing Disclosure Act (General Business Law §349-a)In effect since November 10, 2025A price set by an algorithm that uses personal data must carry the notice "THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA". Up to $1,000 per violation, after a cease-and-desist letter and time to cure.
New York, One Fair Price Act (A9349B)Passed both houses on June 4, 2026; the bill page lists no signatureWould ban pricing set in whole or in part by an algorithm that uses personal data to offer different prices to different customers. Starts 180 days after it becomes law.
Connecticut, Public Act 26-130, section 11Signed June 4, 2026; effective July 1, 2027Bans surveillance pricing by retail sellers of goods and by food delivery services. Anyone else who uses a consumer's personal data to set a price that isn't a discount must show "THIS PRICE WAS INCREASED USING YOUR PERSONAL DATA".
Maryland, HB 895 (Chapter 154)In effect since October 1, 2026Food retailers and third-party delivery services.
New Jersey, Fair Price Protection ActSigned July 23, 2026Retailers' prices for groceries and other necessities.

For a software founder, New York's disclosure law is the one to read closely. It defines personal data as "any data that identifies or could reasonably be linked, directly or indirectly, with a specific consumer or device", and it covers "dynamic pricing set by an algorithm that uses personal data".

A random test doesn't use personal data to decide a price. Most tests do key the assignment to a cookie ID tied to a device, though, and we found no ruling on whether that counts. The law protects people buying "for personal, family or household use", so business purchases fall outside it. If you sell to New York consumers, ask counsel. If you ever add a visitor attribute to the pricing decision, show the notice.

Yes. The UK's consumer rules now sit in the Digital Markets, Competition and Consumers Act 2024, whose unfair-practices rules took effect on 6 April 2025 and let the Competition and Markets Authority fine companies directly, up to 10% of global turnover. Nothing in the act targets random price tests. It is strict about what every price must show.

An invitation to purchase must give "the total price of the product" (section 230), and that total "includes any fees, taxes, charges or other payments that the consumer will necessarily incur". If your checkout adds VAT on top for UK consumers, the number on your pricing page is not the total price.

Schedule 20 of the act lists practices that are unfair in all circumstances, including bait advertising, bait and switch, and false claims that a product or its terms are available only for a limited time. The CMA's price transparency guidance adds that "the total price should normally include any unavoidable or 'mandatory' charges", and that hiding fees until later in the purchase, known as drip pricing, is illegal.

Enforcement has started. In November 2025 the CMA said it had reviewed more than 400 businesses in 19 sectors and opened eight investigations into fees, time-limited offers and pre-ticked extras. The AA fine followed in April 2026, together with refunds of more than £760,000.

Personal data works much as in the EU. The UK GDPR covers the identifier behind a sticky price, and the UK cookie rule, regulation 6 of PECR, was rewritten in February 2026. Its new exceptions include storage for statistics "with a view to making improvements to the service", provided users are told and can object. Whether a price-assignment cookie fits that is, again, a question for counsel.

What went wrong with Amazon's price test in 2000?

Customers noticed and assumed the worst. In September 2000 Amazon ran a five-day test on 68 DVD titles with random discounts between 20% and 40%. Buyers who compared notes suspected that returning customers were being charged more.

Amazon said the discounts were random and had nothing to do with customer demographics. It refunded 6,896 customers an average of $3.10 and promised that in any future test, everyone would pay the lowest price in it.

The refunds came to about $21,400 (6,896 × $3.10). What the test lacked was a plan for the day customers compared notes, and that plan is cheap to write in advance.

How do you run a price test that customers won't resent?

Write the rules before the first visitor sees a test price. These prevent most complaints:

  1. Charge the price you showed. Create the checkout from the same server-side quote that rendered the page, never from a number the browser sends back.
  2. Keep each visitor on one price: the same on reload, on the next visit and, once they sign in, on another device.
  3. Never raise a price on someone partway through a purchase. When a test ends, a returning visitor who saw a lower test price last week will see the winner today. Decide now to honor any price a visitor was shown, for example for 30 days.
  4. Show complete prices in every variant. For consumers, that means taxes and mandatory fees included, with the same billing terms in every arm.
  5. Leave out fake anchors and fake urgency. No "was $49" unless $49 was a real price, and no countdown timers.
  6. Test on new visitors only. Current subscribers keep the price they signed up for, a practice known as grandfathering.
  7. Keep ads and emails on the standard price. A campaign that promises $29 has to deliver $29 to everyone who clicks it.
  8. Keep personal data out of the choice. Location, device or history in the pricing decision can put you under the disclosure rules above.
  9. Decide the refund rule now. If a customer paid more than the lowest test price and asks, refund the difference.
  10. End the test when it has an answer, and move everyone to one price.

What should support say when a customer finds out a friend paid less?

The truth, within a day, with the difference refunded before they have to ask twice. This reply covers most cases:

Support reply: price test
Hi {first name},

Thanks for asking, and sorry for the confusion.

We're testing prices for {plan} at the moment. New visitors are shown
one of a few prices, chosen at random, so two people can see different
prices on the same day. It has nothing to do with who you are or how
you use the site.

You paid {your price}. The lowest price in the test is {lowest price},
so we've refunded the {difference} difference to your original payment
method. Nothing else about your plan changes.

If anything else looks off, just reply to this email.

{your name}

Keep a count of these replies. How many people compare prices tells you something about your market, and a sudden rise is a sign that a variant sits too far from what people expect.

How does Leverprice's design fit these rules?

Mostly through what it leaves out. Leverprice assigns each visitor one price from your range at random and keeps it with an HMAC-signed first-party cookie. A user ID, hashed on your server, puts a signed-in user in the same bucket on every device. Random assignment uses no personal data to choose the test price, and Leverprice does not offer personalized pricing based on personal data. If you turn on regional prices, the visitor's country, read from a header your CDN sets, also sets the price level, so check how the laws above treat location data before you do.

Prices are computed on your server by open-source SDKs and checked against the minimum and maximum you set in code. Your checkout code asks for the same quote, so the price on the page is the price at checkout. Crawlers, link previews and monitors always get your standard price and are never enrolled; the reasons are in does A/B testing prices hurt SEO.

Revenue is verified through your payment provider's webhooks, and refunds come off the revenue of the price they belong to, so refunding a difference doesn't distort the result. When a winner is clear, you lock it in with one click and it is live for everyone in about a second. More on how Leverprice runs a price test.

None of this is a compliance guarantee. Your notices, cookie consent, refund policy and terms are yours to set with counsel. The design keeps the test random and stable, which is the side of the line these laws are least concerned with.

Frequently asked questions

Do I have to tell customers I'm testing prices?

We found no US, EU or UK law that requires announcing a random price test. Disclosure duties start when personal data or automated profiling sets the price: the EU's Article 6(1)(ea), New York's notice since November 2025 and Connecticut's from July 2027.

Is a multi-armed bandit treated differently from an A/B test?

Not by any law we found. A bandit changes how many visitors see each price based on total results, and each visitor's price is still a random draw that ignores who they are. Here is how bandits work for pricing.

Does New York's law apply if my company isn't based in New York?

It applies to any entity "domiciled or doing business in New York state", and it protects consumers buying for personal, family or household use. Selling online to New York consumers is the case to raise with counsel.

Possibly. The ePrivacy Directive requires consent unless a cookie is strictly necessary for a service the user asked for, and a price-assignment cookie sits in a gray zone. If consent is required, give visitors who decline your standard price and keep them out of the test.

Should existing customers be part of a price test?

No. Test on new visitors and keep current subscribers on the price they signed up for. A surprise price at renewal turns a test into a support ticket.