Two people can visit the same website, look at the same product and receive different offers. The difference may not be demand, inventory or timing. It may be what an algorithm believes about them.
That possibility stopped being an abstract privacy debate when some readers of The Washington Post reportedly saw a blunt disclosure beside subscription offers: “This price was set by an algorithm using your personal data.” The wording came from a New York law requiring businesses to disclose personalized algorithmic pricing to consumers in the state.
The notice tells us something important, but not everything. It establishes that an algorithm and personal data were involved in setting the displayed price. It does not reveal which data were used, how heavily each signal was weighted, whether another person received a different offer or whether the goal was to estimate willingness to pay.
The larger technology story is the rise of what the US Federal Trade Commission calls surveillance pricing: systems that combine consumer data, experimentation and automated decision-making to tailor prices, discounts or product displays. These systems can make commerce more responsive. They can also make an ordinary purchase feel like an invisible negotiation in which only one side knows the rules.
Prices have always changed. Airlines adjust fares as seats fill. Hotels respond to seasons and events. Ride-hailing services respond to local demand. Retailers discount products as inventory ages.
That is dynamic pricing: a price changes because conditions change.
Personalized algorithmic pricing goes further. New York General Business Law Section 349-A defines it as dynamic pricing set by an algorithm using personal data linked directly or indirectly to a specific consumer or device. The law requires the notice to appear near the relevant price when the practice applies.
The distinction is simple but important:
Personalization can also operate without changing the sticker price. A system might choose which discount to display, rank more expensive products first, alter the length of a trial or decide who receives a retention offer. The commercial outcome changes even if the base price remains technically constant.
A personalized-pricing system is not one magical AI model. It is usually a pipeline.
The system may receive first-party data such as account history, subscriptions, purchases, abandoned carts and engagement. It may also use device, location, referral, advertising or third-party data, subject to applicable law and company policy.
The FTC's 2025 market study found that pricing intermediaries can use information including location, demographics, browsing patterns, shopping history and even interactions such as mouse movements. The agency stressed that its public examples were aggregated or hypothetical because confidential business information was protected.
Raw activity is converted into variables a model can use: purchase frequency, likelihood to cancel, price sensitivity, device category, recent searches or membership status. Identity systems may attempt to connect activity across sessions and devices.
A model estimates an outcome—perhaps whether a visitor will subscribe, abandon a cart or accept a particular offer. A rules engine or optimization service then selects a price, discount, ranking or promotion.
Businesses often run controlled tests to compare offers. Experimentation can improve products, but it becomes ethically sensitive when the system tests how much different people can be charged without explaining the basis of the difference.
Purchases, cancellations and responses return to the system. Over time, the model learns which offers produce revenue or retention. If the objective rewards only short-term conversion, the system may optimize away trust without anyone explicitly programming it to do so.
Reporting by Washingtonian documented the personalized-pricing disclosure appearing on some Washington Post subscription offers. The disclosure is meaningful because New York law specifies the wording businesses must use.
However, a careful analysis should avoid filling in missing details. The notice alone does not prove that the publication used sensitive traits, inferred a reader's income or deliberately charged the maximum that person could tolerate. It also does not tell us whether the personalized offer was higher or lower than a standard offer. New York's law contains an exception for certain offers to existing subscribers that are below their contracted subscription price.
That uncertainty is exactly why transparency must extend beyond a warning label. Consumers need enough information to understand the category of data used, the purpose of the decision and the choices available to them.
From a product team's perspective, personalization can solve real problems. A discount can retain someone who would otherwise cancel. A regional price can make a service affordable in a lower-income market. An introductory offer can reduce the risk of trying an unfamiliar product.
But the same infrastructure can be used to identify urgency, limited alternatives or a high willingness to pay. The technology does not carry an ethical position of its own; it executes the objective and constraints chosen by people.
A system optimized solely for immediate revenue may learn to extract more from customers who appear less likely to comparison-shop. A system optimized for long-term value may instead favor predictable pricing, useful discounts and lower churn. Product metrics define what the algorithm learns to value.
There is also a measurement problem. A short experiment may show higher revenue while missing the long-term costs of complaints, cancellations, regulatory attention and reputational harm.
A responsible implementation begins before model training.
For consumers, practical protections include comparing prices while signed out, reviewing tracking permissions, limiting unnecessary location access and checking subscription offers directly. These steps cannot rebalance the entire market, but they reduce some of the data available for profiling.
Online commerce is moving from “What does this product cost?” toward “What offer should this person see right now?” That shift makes pricing part of the software experience—and therefore part of privacy, security, UX and AI governance.
The most important question is not whether algorithms should ever influence prices. They already do. It is whether customers can understand the decision, whether organizations can defend it and whether the system creates value without exploiting informational power.
Personalization can make a product more relevant. Once it reaches the price tag, relevance is no longer enough. The system must also be worthy of trust.
I design and build digital platforms with attention to the full system: user experience, backend logic, data handling, integrations and operational reliability. To discuss a product, website or responsible automation project, email hello@massoda.me or contact me on WhatsApp.
Editorial note: The Washington Post disclosure confirms that an algorithm using personal data set the relevant offer. Public evidence does not reveal the model's complete inputs, weighting or intent.