First large study of ChatGPT ads finds low-income accounts see more of them

How researchers observed the ads

ChatGPT began showing ads to users of its free and low-cost tiers in February 2026. In early October, a research team published what Brookings describes as the first large-scale analysis of these ads. The authors are Sorelle Friedler, Emma Lurie and Danaé Metaxa.

Because ChatGPT is a closed platform with no separate mechanism for observing prompts, results and associated ads, independent researchers had to build their own. The team used "sock puppets", a common AI audit approach in which accounts are created so that the behavior of logged-in users can be observed with controlled prompts.

Ninety-one such accounts were created between January and March 2026. They were spread across three racial and ethnic groups (Black, Hispanic, and white) and three income terciles (low, medium, and high), with at least nine accounts in each race-income combination. Group membership was varied by using geographic region as a proxy, with accounts accessed from zip codes with the associated characteristics.

Each day, the active accounts were prompted with content across a range of topic categories. These included topics where ads may reasonably be served, such as "Recipe for lasagna", and topics where OpenAI said it would not serve ads, such as mental health.

What the study found

The team collected more than 3,600 ads. They ranged across sectors but were especially prevalent in response to queries about products, health, fitness, and cooking. Some top ad-triggering prompts were obvious fits, such as "What's the best streaming service for sports?" Others were stranger matches, such as "Give me tips for work-life balance." A request for a step-by-step guide to make pad Thai turned up an ad for DoorDash food delivery.

The key finding is that accounts from lower-income zip codes were significantly more likely to receive ads. Among the 53 accounts that received at least one ad, location income clustered at lower levels, with a mode around $60,000. Accounts that were not exposed to ads had a broader distribution, peaking near $80,000. According to the authors, the gap suggests the pattern is not driven by a few outliers but is consistent across the sample.

The study did not identify evidence that race or ethnicity affected whether an account received ads. It also found no obvious problems with OpenAI's guardrails, which include not serving ads on violent, political or hateful content. The authors caution that, given the small sample sizes, these null results must be interpreted carefully, especially as advertising practices on the platform are certain to change in the coming years.

The study may also have been cut short. About a month in, all accounts stopped being shown more ads, which the authors say perhaps suggests OpenAI detected them, effectively ending the study.

Policy recommendations and what it means

The authors argue that continued independent research access to ChatGPT and other closed AI platforms will be critical as they evolve, and that policymakers should ensure and protect it. Mandatory transparency requirements and independent researcher access to served ads would, in their view, bring public accountability to opaque platforms and make auditing more reliable, more accurate and less expensive than the sock-puppet method.

They point to ways AI advertising may differ from social media advertising. The conversational tone of chatbots is persuasive, and repeated interactions and chatbot "memory" will allow AI companies to build complex user profiles and serve ads inconspicuously integrated into ordinary responses. People increasingly treat chatbots as confidants and trust their recommendations, which the authors say leaves them open to manipulation and discrimination in private interactions.

The authors draw a parallel with Facebook, where repeated investigations showed its advertising was discriminatory and that the company had not fixed the problem. A settlement eventually led to third-party oversight. They say AI advertising may be about to rerun this playbook and urge policymakers to act now to protect independent monitoring and constrain potentially deceptive advertising practices.

Brookings notes that Meta is a general, unrestricted donor to the institution, and that the findings are solely those of the authors.