From Sidhant Tamrkar | Product & Market Analysis
FTC Personalized Pricing Draft: Its Clauses, Its Critics, and AI Pricing Tools
On this page
On 19 August 2026 the Federal Trade Commission voted 2-0 to publish a draft policy that does not ban personalized pricing. It does something more useful to a pricing vendor: it says what a lawful personalized price has to disclose. The FTC personalized pricing draft expects 3 disclosures, and each describes information that lives inside the pricing tool, not the retailer's checkout page.
Key takeaways
- The draft regulates silence, not price variation. It says failing to disclose a personalized price is likely deceptive where shoppers reasonably expect one price for everyone, and it concedes the FTC cannot ban the practice outright.
- Three disclosures carry the whole policy. A personalized price should say that it is personalized, what the basis for personalization is, and which types of data were used, clearly and conspicuously.
- Industry split on direction, not on the facts. NetChoice asked the FTC to withdraw the draft, CCIA and J. Howard Beales asked it to carve out discounts, and Consumer Reports said disclosure alone is not enough.
- The open question is who owns the disclosure. CCIA argued the duty sits with whoever determines the price, not the tool supplier. In practice the retailer cannot disclose a basis the vendor's model cannot explain.
The short answer
The FTC's draft covers businesses that use personal data to set an individual's price where shoppers expect a uniform price. They should disclose that the price is personalized, its basis and the data types used. It is a nonbinding proposal. Comments closed on 25 September 2026, and no final version had appeared when we checked.
This post is for founders of B2B companies selling AI pricing, promotion or loyalty software. The draft addresses retailers, but its weight lands on the vendors whose models set the prices.
The FTC personalized pricing draft, clause by clause
The document is formally the Proposed Enforcement Policy Statement Regarding Personalized Pricing, matter number P034101. The FTC published it on 19 August 2026 and opened docket FTC-2026-1057 on regulations.gov. The original deadline was 18 September. On 3 September the FTC extended it by seven days to 25 September, without giving a reason.
The FTC release quotes Chairman Andrew Ferguson saying the agency "does not have the legal authority to ban personalized pricing in all circumstances". So the draft is a disclosure policy built on Section 5 of the FTC Act, which bars unfair or deceptive practices.
The definition and the trigger
The FTC release defines personalized pricing as "the use of personal data to set prices according to the amount that a company believes an individual consumer is willing to spend". Hogan Lovells reads it as covering inferences too, such as a shopper's likelihood of comparison shopping. Morgan Lewis notes the draft is not expressly limited to prices generated by algorithms or AI.
The trigger is expectation. Per Hogan Lovells' reading, the concern arises "in markets where consumers reasonably expect the same contemporaneous price to be available to others". Arnold & Porter quotes the draft's plainer version: a retailer listing should show the same price to anyone browsing it.
The draft does not ask whether a price varies. It asks whether the shopper had a reason to think it did not.
The three disclosures and what counts as clear
Where the trigger applies, the draft expects a clear and conspicuous disclosure covering 3 things. The price is personalized. The basis for the personalization, such as willingness-to-pay modelling. The types of data used. Paul Weiss lists the same 3 and warns that vague or buried disclosures may not suffice.
The draft also gives a floor and a ceiling. Hogan Lovells reports that calling a price "specially selected" would likely be insufficient. Arnold & Porter quotes footnote 9, which gives an example the FTC says would likely be adequate if accurate and complete. It is a price "based on a consumer's estimated willingness to pay derived from data about that consumer's previous purchases from the same retailer through the same login account".
Read that example as a specification. It names the model's objective, the data source, the data owner and the identity link. A disclosure that specific needs a pricing system that can produce those 4 facts for each price it sets.
Deception, unfairness and consent
The deception theory has 3 branches, quoted by Arnold & Porter from footnote 6. Representing a price as static or widely offered when it is personalized. Failing to disclose personalization where the shopper reasonably believes the price is static. And misrepresenting the basis or effect, for example presenting a price derived from inferences about disposable income as a loyalty discount.
That third branch is the likeliest to catch a promotions product. A tool that labels every individualized offer a reward, while some sit above the shelf price, writes the deceptive claim for its customer.
Unfairness is the second theory. A higher personalized price can be a substantial injury that shoppers cannot reasonably avoid, because they cannot see or dispute the data behind it. The draft also says, per Kelley Drye, that basing a price on personal data "without sufficiently verifying that consumers consented to the collection of those data for that purpose" may independently violate Section 5. That sentence reaches every third-party data feed a pricing model ingests.
| Clause | What the draft says | What it implies for the tool |
|---|---|---|
| Definition | Personal data used to set a price by estimated willingness to pay, including inferences | Model inputs at the individual level are in scope, whatever the model type |
| Trigger | Shoppers reasonably expect the same contemporaneous price for others | Each deployment context needs a written expectation assessment |
| Disclosure | Fact, basis and data types, clear and conspicuous | The tool must output a plain-language basis and data list per price |
| Deception | Static claims, silence, or a markup dressed as a discount | Every offer needs a logged direction against a reference price |
| Unfairness | Unavoidable higher price, not outweighed by benefits | Shoppers need a way to see and dispute the inputs |
| Consent | Data used without verified consent for pricing may violate Section 5 | Consent provenance has to travel with every data feed |
What the draft leaves out of surveillance pricing
The FTC has used the phrase surveillance pricing since its 2024 study, while the draft says personalized pricing. Either way, its scope is narrower than the name suggests.
Practices the draft names as normal
Arnold & Porter summarises the draft as accepting that prices vary because of supply and demand, geography, taxes, market conditions and risk-based underwriting. Paul Weiss adds rideshare surge pricing tied to real-time market conditions. Insurance and credit are described as markets where individual pricing "is a long-established norm".
So a demand forecast that moves the price for a whole region is outside the draft, as is a markdown optimiser clearing stock. The line is crossed when the input is about the person, not the market.
The draft also lists cases it considers troubling, which the FTC calls illustrative only. Arnold & Porter's summary of footnote 11 includes a delivery app charging more to someone data suggests cannot leave home. It also lists a grocer pricing milk higher for households with children and a hotel pricing higher on inferred funeral travel. Others are a rideshare price raised because no competitor apps are installed, and a security retailer charging a recent crime victim more.
The draft also declines, per Arnold & Porter and Hogan Lovells, to say whether a fully disclosed personalized price could still be unfair. That reservation is deliberate: it leaves the FTC free to challenge a disclosed practice later. Disclosure lowers your risk without closing it.
The comment record on algorithmic pricing at the FTC
The docket drew trade associations, academics and consumer groups. What follows covers only the filings we could read or that a named outlet reported.
The industry case for withdrawal or narrowing
NetChoice asked for the most. MediaPost reported on 28 September that the group, which represents platforms including Amazon, Meta and Google, urged withdrawal. Its filing says the FTC lacks evidence of "a widespread problem warranting economy-wide enforcement policy". It also argued that "mandating alarm-inducing language in beneficial transactions does not correct a deception".
CCIA took a narrower line in its 25 September filing. It asked the FTC to distinguish personalized price increases from discounts, coupons, loyalty rewards and membership benefits. It also argued the duty should rest with the entity that determines the price, and that supplying tools should not oblige platforms to monitor sellers' pricing.
J. Howard Beales, a former director of the FTC's Bureau of Consumer Protection, filed a comment published by the International Center for Law & Economics. He asked for a safe harbor exempting data-driven discounts from disclosure. On timing he writes: "Disclosure when a price is offered is in the wrong place and at the wrong time."
Ad groups ANA, 4A's and AAF called the draft too vague, per MediaPost.
The advocates' case for going further
Consumer Reports opposes personalized pricing and, per MediaPost, says disclosures alone will not protect consumers. The National Consumer Law Center's 25 September comment went elsewhere, asking the FTC to apply the policy to debt collection discounts and settlement offers. It says these are increasingly calculated using AI or analytics.
Our read is that a discount carve-out is the change most likely to survive into a final version. CCIA and Beales both asked for it, it costs the FTC little, and it would sharpen the draft's existing line against a markup dressed as a discount.
Mapping the draft onto AI pricing vendors
The draft names no software vendor. That does not put vendors outside it. The FTC built its 2024 study around them. On 23 July 2024 it voted 5-0 to send 6(b) orders to 8 intermediary firms: Mastercard, Revionics, Bloomreach, JPMorgan Chase, Task Software, PROS, Accenture and McKinsey.
The preliminary findings in January 2025, per Morgan Lewis, found those intermediaries worked with at least 250 clients. The tools could draw on location, demographics, browsing patterns, shopping history, mouse movements and items left in an online cart. Morgan Lewis advises businesses that provide pricing, analytics or loyalty technology services to evaluate the draft directly.
Who owns the disclosure
CCIA's comment asks the FTC to put the disclosure duty on whoever determines the price. On paper that is the retailer, which we think is legally right. Commercially it changes little.
A retailer cannot disclose the basis of a price, or the data behind it, unless its vendor's system can say. If the model is a black box trained on mixed first-party and broker data, the retailer has nothing accurate to disclose.
That is why we expect the draft to show up first in procurement, not in enforcement. Retail legal teams will ask pricing vendors for the facts behind the 3 disclosures before they ask the FTC anything. The questions resemble a good AI agent procurement checklist, pointed at prices instead of actions.
What an AI pricing vendor will be asked to show
The table below is our translation of the draft into vendor evidence, not FTC language.
| Question a retail customer will ask | Weak answer | Answer that holds up |
|---|---|---|
| Does the model use personal data at the individual level? | "We use AI to optimise prices" | A feature list split into market-level and person-level inputs, with a switch to disable the second set |
| Can you state the basis for a given price? | Aggregate model documentation | A per-price reason code in plain language that the checkout page can render |
| Which data types fed this price? | Reference to the privacy policy | A data-type list per price, matching the categories the disclosure will name |
| Was consent collected for pricing use? | "Our data partners are compliant" | Consent provenance per feed, with purpose recorded, and feeds without it excluded |
| Is this offer a discount or a markup? | Everything labelled a reward | Every offer logged against a reference price, with direction and size |
| Can sensitive inferences be blocked? | Not addressed | Exclusion rules for the draft's example categories, such as household children or crisis travel |
Two rows matter most. The discount-or-markup log is cheap to build and directly answers the deception branch the FTC spelled out. The consent provenance row is expensive and where many pricing products are weakest, because broker data rarely arrives with purpose-specific consent.
The logging pattern is the same discipline as an agent audit log specification: every decision recorded with its inputs. If your product cannot reproduce why it quoted a price last Tuesday, it cannot support a disclosure about that price today.
B2B contract quotes look outside the draft, since every example in it is a consumer purchase. But sole traders buying on consumer-style storefronts blur that line, and the FTC has not drawn it.
Algorithmic pricing rules already in force at state level
The FTC draft is not the strictest rule a pricing vendor faces. New York's Algorithmic Pricing Disclosure Act, General Business Law section 349-a, already requires a fixed sentence next to qualifying prices. The wording, per Benesch, is "THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA".
The National Retail Federation challenged that law on First Amendment grounds. Judge Jed Rakoff of the Southern District of New York dismissed the challenge on 8 October 2025, finding the disclosure factual and uncontroversial. Benesch noted that enforcement had been stayed during the litigation and the law would take effect absent a stay pending appeal.
Other states have moved too. Hunton describes Maryland as the first state to directly prohibit certain personalized pricing in groceries, with New Jersey following. Connecticut amended its privacy law to add disclosure duties and substantive bans for retailers and delivery services. California's Attorney General has opened an investigative sweep under the CCPA. The obligations Connecticut places on AI deployers and the California AI laws that bind software vendors have their own posts.
These rules stack. New York asks for a fixed sentence, the FTC draft for a basis and data types. A retailer selling into New York could need both, and the FTC's version asks far more of your model.
How much an FTC policy statement on pricing weighs
A policy statement explains how an agency plans to use powers it already has. Hogan Lovells calls it nonbinding, creating no private rights. Paul Weiss is blunter: even when final, it will not have the force of law.
That understates its effect. Morgan Lewis notes that courts and state enforcers often treat an FTC policy statement as indicative of what Section 5 means. Paul Weiss also flags that the draft may implicate the Restore Online Shoppers' Confidence Act and the FTC's Fees Rule, both of which carry their own remedies.
The vote count matters too: the 2024 study orders passed 5-0, the draft 2-0 with only two sitting commissioners voting. A future Commission with a different majority could finalise it, harden it or drop it. Build to the 3 disclosures anyway, because the state laws do not depend on who sits at the FTC.
The FTC has form on turning soft guidance into cases. Its AI accuracy statement earlier this year followed the same pattern, covered in the analysis of the FTC's Section 5 theory and Colorado's AI law. The SEC's parallel record on AI claims is in the AI washing enforcement log.
Where this reading is weakest
Our argument is that the draft's real burden falls on vendors' ability to explain their models. Here is where it could be wrong.
We did not read the draft's own text. The FTC's PDF yielded no extractable text for us, so every quotation attributed to the draft comes via the FTC press release or a named law firm. Arnold & Porter, Hogan Lovells, Kelley Drye and Paul Weiss agree on the 3 disclosures and the deception branches, which gives us some confidence. Footnote numbering and exact wording still need checking against the primary document.
The comment record we cover is partial. The regulations.gov docket refused automated access, so we cannot say how many comments were filed or how they split. A search result attributed comments to the US Chamber of Commerce and the food retail group FMI, but we could not open either filing, so neither appears above. Six filings out of a full docket is a sample, not a census.
The vendor burden may be overstated. If the FTC adopts CCIA's position that the price setter owns the disclosure, vendor explainability could become a commercial nice-to-have, not a compliance need. A retailer could also stop personalizing on person-level data, removing the trigger and the vendor problem together. Some will.
The draft might never be finalised. With a 2-0 vote, a contested comment record and no rulemaking behind it, withdrawal is a real outcome. The state laws remain if it is.
Frequently asked questions
What is the FTC personalized pricing policy statement?
It is a proposed enforcement policy statement the FTC published on 19 August 2026 by a 2-0 vote. It says businesses that use personal data to set individual prices, where consumers reasonably expect a uniform price, are likely acting deceptively under Section 5 if they do not disclose it. It does not ban personalized pricing. Comments closed on 25 September 2026, and the statement remains a proposal until the FTC finalises, revises or withdraws it.
What disclosures does the FTC expect for personalized pricing?
The draft expects a clear and conspicuous disclosure of 3 things: that the price is personalized, the basis for the personalization, and the types of data used. Law firm readings say vague labels such as "specially selected" would likely be insufficient. The draft's own example of an adequate disclosure names the model's objective, the data source and the account link, for instance willingness to pay estimated from prior purchases on the same login.
Is surveillance pricing illegal in the US?
Not as a general matter. The FTC chairman has said the agency lacks authority to ban personalized pricing in all circumstances, so the FTC draft relies on deception and unfairness. Some states go further. Maryland and New Jersey restrict certain personalized grocery pricing, Connecticut added disclosure duties and bans, and New York requires a fixed disclosure on prices set by an algorithm using personal data.
Does the FTC personalized pricing policy apply to AI pricing software vendors?
The draft is addressed to businesses that set prices for consumers, not to software suppliers by name. CCIA asked the FTC to keep it that way and place the duty on the price setter. In practice vendors are exposed indirectly, because a retailer can only disclose the basis and data types if the vendor's system can produce them. The FTC's 2024 study also targeted 8 pricing intermediaries directly.
What did industry groups say about the FTC personalized pricing proposal?
NetChoice urged the FTC to withdraw it, arguing there is no evidence of a widespread problem. CCIA asked the FTC to separate personalized price increases from discounts and loyalty rewards, and to put the disclosure duty on whoever sets the price. Former FTC official J. Howard Beales asked for a discount safe harbor. Advertising groups called it too vague, while Consumer Reports said disclosure alone is not enough.
Is dynamic pricing covered by the FTC personalized pricing draft?
Mostly not. Law firm readings of the draft say it accepts price changes driven by supply and demand, geography, taxes, market conditions and real-time rideshare surge pricing. The draft targets prices set from data about the individual shopper, such as browsing history, location or inferred willingness to pay. A model that moves prices for everyone in a market is outside it. A model that moves one person's price is not.
Where to start this quarter
Pull 20 prices your system set last week for a single retail customer. For each, write the disclosure the draft describes: that it was personalized, why, and which data types fed it. If your team cannot pull all 3 facts from logs in an afternoon, that gap is the product work. It is also the first thing a customer's legal team will find.
Then split your feature list in two. Market-level inputs on one side, person-level inputs on the other, with consent provenance written next to every person-level feed. Ship a configuration that switches the second list off. Some customers will take it; the rest will need the evidence from the first exercise.
Related on AI disclosure
Disclosure can raise trust or erode it depending on how it is written. See what the evidence says about AI disclosure and customer trust.
References
- Federal Trade Commission, FTC seeks comment on enforcement policy statement regarding personalized pricing, 19 August 2026, updated 31 August 2026. Used for the definition, the 2-0 vote, the chairman's statement and the docket.
- Federal Trade Commission, FTC issues orders to eight companies seeking information on surveillance pricing, 23 July 2024. Used for the 8 intermediaries and the 5-0 vote.
- Arnold & Porter, All eyes on personalized pricing: what stakeholders need to know, 8 September 2026. Used for the disclosures, footnotes 6, 9 and 11, the exclusions and the extension.
- Hogan Lovells, FTC draft personalized pricing policy would require more than a personalized price disclosure, 25 August 2026. Used for the trigger, the specially selected example and nonbinding status.
- Morgan Lewis, FTC proposes enforcement policy on personalized pricing, seeks public comment, 28 August 2026. Used for the 2025 findings, vendor scope and the weight of policy statements.
- MediaPost, Wendy Davis, NetChoice to FTC: withdraw personalized pricing proposal, 28 September 2026. Used for the NetChoice, ad industry and Consumer Reports positions.
- J. Howard Beales, Comment on FTC's proposed enforcement policy statement regarding personalized pricing, International Center for Law & Economics, 25 September 2026. Used for the discount safe harbor and timing argument.
- Benesch, New York algorithmic pricing law survives First Amendment challenge, 14 October 2025. Used for the New York disclosure wording, exclusions and ruling.
The weakest point in this source base: the draft's text is quoted via the FTC press release and law firm readings, not the PDF. The comment record covers six filings, not the full docket. CCIA, NCLC, Paul Weiss, Kelley Drye and Hunton were also read and are linked or named in the text. This is analysis, not legal advice.
Related reading