Is A/B testing prices legal?

Federal law

The Robinson-Patman Act bans some price discrimination, but it covers “commodities”, physical goods sold between businesses, when it hurts competition.⁠^[1] Legal commentary generally treats software and services as outside it.

The FTC’s rules on deceptive pricing do apply. A crossed-out “former price” that was never really charged is a false bargain.⁠^[2] A test price shown as a plain price, with no fake discount, avoids this.

The FTC has been studying “surveillance pricing”: prices set from someone’s location, browsing or demographics.⁠^[3] In August 2026 it proposed a policy statement that would require clear disclosure when a price is personalised from personal data.⁠^[4] It’s a proposal, and it’s about personal data, not random tests.

State laws

New York’s Algorithmic Pricing Disclosure Act took effect in November 2025. When a price is “set by an algorithm that uses personal data”, the seller must show “THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA”.⁠^[5]⁠^[6] A random split doesn’t set the price from personal data. But the law counts data linked to a device as personal data and there’s no guidance on random tests yet, so if you sell to New York consumers, get advice.

In 2026, Maryland, Connecticut and New Jersey passed bans on surveillance pricing aimed at groceries, retail and delivery.⁠^[7] New York’s broader One Fair Price Act and California’s AB 2564 were still pending at the time of writing.

Since 2022, EU sellers must tell consumers when “the price was personalised on the basis of automated decision-making”.⁠^[8] The Commission’s guidance defines personalised pricing as customisation from profiling: “Traders gather data about individual consumers using automated decision-making to profile consumers’ behaviour.”⁠^[9] A random split profiles nobody. As we read it, it falls outside the rule, though the guidance doesn’t mention A/B tests by name.

The Unfair Commercial Practices Directive “does not, as such, prohibit traders from price discrimination as long as they adequately inform the consumer about the total price”.⁠^[10] Show the full price, taxes included where required, before checkout.

The Geo-blocking Regulation bans different prices based on a buyer’s nationality or country of residence.⁠^[11] A random split isn’t based on either. Testing different prices per country is a different matter.

The Commission is preparing a Digital Fairness Act with “unfair personalisation” in scope; the proposal hadn’t been published at the time of writing.⁠^[12]

Cookies

A test needs to remember which price a visitor saw. France’s CNIL exempts some cookies from consent, such as those needed for a shopping cart, but its guidelines don’t mention A/B tests.⁠^[13] Opinions differ, so list the cookie in your privacy and cookie policy and set it according to your consent setup. PricingFast uses one random ID, pf_vid, and no names, emails or card details.

Most consumer rules protect people buying for personal use. The EU’s consumer directives and New York’s law both define the consumer that way, so purchases by companies are mostly outside them.

Two cautions: the FTC Act isn’t limited to consumers, and many SaaS pricing pages sell to freelancers and solo founders who may count as consumers. Act as if your buyers are consumers and you won’t go wrong.

In 2000, Amazon randomly varied DVD prices for a few days. Buyers noticed, and Amazon refunded about 6,900 customers an average of $3.10, calling it “a random price test, and even that was a mistake”.⁠^[14] In 2025, Instacart ended its price tests after reports that shoppers saw different prices for the same items, even though it said the tests were random.⁠^[15]

Both were consumer retail, where shoppers compare prices item by item. SaaS buyers see a plan price once, but the lesson holds: be ready to explain a test, and to honour the lower price if asked.

• Charge exactly the price you show. No surprise at checkout.
• No fake discounts. Never cross out a price that wasn’t really charged.
• Assign prices at random. Never use location, device, browsing or other personal data to pick one.
• Test on new visitors only. Leave existing subscriptions alone.
• List the test cookie in your privacy and cookie policy.
• If a buyer asks, explain the test and offer the lower price.