News · 2026-10-16

ChatGPT ads use conversation context and advertiser customer-list matching

ChatGPT’s advertising system uses current-conversation context to select ads and can match advertiser-uploaded customer identifiers against eligible accounts. OpenAI’s documentation requires at least 25,000 matched users for audience inclusion or bid adjustments, distinguishing audience-level matching from an advertiser selecting or identifying one particular person.

Key facts

• Advertisers can supply email addresses, phone numbers, hashed versions and supported advertising identifiers.
• Inclusion targeting and bid adjustments require at least 25,000 matched users, not merely 25,000 uploaded rows.
• OpenAI says uploaded audience files are usually deleted within 24 hours after processing.
• Primary sources: OpenAI’s custom-audience guide and ads documentation, corroborating 404 Media’s own ad test.

404 Media bought and ran two sponsored messages about ChatGPT privacy, then examined how the system selected conversations in which to display them. That experiment makes the story more concrete than speculation about a future advertising business. The outlet reports operating the advertiser interface and seeing its advertisements appear in relevant chats.

There are two separate mechanisms. Conversation relevance concerns the subject of a chat and the likely usefulness of an advertisement. Customer-list matching concerns whether an account belongs to a group supplied by the advertiser. They can influence the same advertising system without being the same data flow. Confusing them can create the false impression that an advertiser receives a transcript or can buy access to a named individual’s conversation.

For contextual selection, advertisers provide what the interface calls “context hints.” These describe situations or conversations in which an offering may be relevant. OpenAI’s documentation says they guide matching rather than operate as exact keyword triggers. A hint about a category of interest is therefore not a guarantee that a particular sentence will display a particular advertisement.

A useful analogy is a shopkeeper choosing an advertisement based on the aisle a customer is browsing. The aisle supplies context without the advertiser necessarily receiving the customer’s full shopping history. A conversation can be much richer and more sensitive than an aisle, however, so the analogy explains the mechanism without settling the privacy question. The platform’s use of context is itself consequential even if outside advertisers do not see the underlying text.

Customer lists add a different route. Under OpenAI’s audience documentation, an advertiser can upload supported identifiers to match existing customers or prospects to eligible users. Raw or SHA-256-hashed email and phone fields are accepted, along with supported advertising identifiers. A hash transforms a value for matching; it should not be mistaken for a guarantee that the original identity has become irrelevant to the operation.

The numerical threshold is a critical boundary. At least 25,000 users must actually match for inclusion or bid adjustments; the documentation recommends larger lists. Uploaded rows that fail to match do not satisfy the requirement. Smaller lists may be used for exclusion. Advertisers cannot see which individuals matched or select an individual from the matched audience.

OpenAI says advertisers receive aggregate performance information rather than chat content or user identity. Its general advertising guide also describes relevance signals beyond the current chat, including selected broader-experience signals when personalization is enabled. Switching personalization off does not remove the current conversation as a contextual signal, nor does it automatically remove advertisements.

Plan details matter as well. The reviewed documentation describes ads for Free and Go users, with Plus, Pro, Business, Enterprise and Edu ad-free. It also describes a reduced-access no-ads option for some Free users, with availability varying. Those distinctions make a blanket statement that every ChatGPT user sees the same targeting or advertising setup inaccurate.

The strongest counterargument to alarmist coverage is that contextual selection and customer audiences are established advertising techniques, and the published controls expressly limit advertiser visibility and individual selection. The strongest privacy concern is that the conversational setting produces unusually revealing context inside a service people may use for personal questions. Both points can be true without asserting a data sale the sources do not establish.

Zach Edwards of DecryptAds told 404 Media that the rollout resembles other large platforms’ gradual monetization and argued for stronger deletion rights and data-broker rules. That is an attributed expert position, not a regulatory finding. OpenAI says ads do not shape answers and are ineligible near specified sensitive subjects; the dossier does not independently test every boundary of those policies.

The verified story is a concrete monetization architecture: current-chat relevance on one side and audience-level identity matching on the other. Readers can evaluate those mechanisms against their privacy expectations without being told, inaccurately, that advertisers can read all prompts or target a single named person through the documented audience tool.

Primary source, verified: read the paper →

Ground Truth. (2026, October 16). ChatGPT ads use conversation context and advertiser customer-list matching. Ground Truth. https://groundtruth.day/news/chatgpt-ads-combine-conversation-context-with-customer-list-matching.html

@misc{groundtruth:chatgpt-ads-combine-conversation-context-with-customer-list-matching,
  title  = {ChatGPT ads use conversation context and advertiser customer-list matching},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/chatgpt-ads-combine-conversation-context-with-customer-list-matching.html}
}

Topics: openai · advertising · privacy · chatgpt · platforms