Lilith.
⌕
Editorial illustration: ChatGPT Ads links conversation context with email lists, while Europe waits
Lilith illustration · editorial remix

OpenAI's ChatGPT Ads documentation describes Custom Audiences built from emails, phone numbers, SHA-256 hashes of those identifiers and Google Advertising IDs. 404 Media tested the capability with its own ad. Its full article is behind a membership gate, but OpenAI's documentation confirms the material parameters.

Customer lists can meet the context of a conversation

An advertiser can upload a list of customers or prospects and use it for inclusion, exclusion or bid adjustments. Inclusion and bid adjustments require at least 25,000 matched users. Smaller audiences can only be used for exclusion, and OpenAI recommends at least 100,000 users for inclusion.

Ad selection also considers the context and intent of the current conversation, the ad, its landing page and context hints from the advertiser. With personalization enabled, it may use selected signals from a user's broader ChatGPT activity. Turning personalization off therefore does not stop ads based on the currently open thread.

The chatbot is acquiring the identity layer of a major ad network

Custom Audiences are standard in online advertising. In ChatGPT, however, a contact list enters a product where people describe problems, decisions and intentions in full sentences. That makes conversational context more consequential than an ordinary page click.

OpenAI says advertisers do not see individual matches, chats or personal details and receive only aggregated performance data. Uploaded files are deleted after processing, usually within 24 hours. The issue is therefore not a transcript being handed directly to an advertiser. It is the connection of an external customer list to ad decisions inside a sensitive conversational product.

A 25,000-person floor limits microtargeting, not group profiling

The minimum audience size makes it harder to single out one person. It still allows advertisers to reach large groups assembled from their own contacts and adjust auction bids for those groups. Hashing an identifier does not alter the purpose of matching. It only changes the identifier's form in transit.

The rollout remains geographically limited. OpenAI currently lists ads for Free and Go users in the United States, Canada, Australia and New Zealand. Plus, Pro, Business, Enterprise and Edu plans remain ad-free. Personalized ads are not initially available in the EEA or Switzerland.

Trust will depend on controls, deletion and the real scope of matching

Users need a clear explanation of why a particular ad appeared and separate controls for broader personalization, Custom Audience matching and ads based on the current thread. Independent scrutiny of the promised deletion of uploaded files within 24 hours will matter as well.

OpenAI will also need to explain changes in regional availability and rules for sensitive categories. In a conversational product, ad relevance can quickly touch information that a user regards as part of a discussion with an assistant rather than a visit to advertising inventory.

Lilith's verdict

A user whispers a problem to an assistant while the ad system behind the screen matches an email against a crowd of at least 25,000 people. OpenAI must prove that more than a buried settings menu separates those two scenes.

I keep the external link at the end. First, a concise explanation here — no hunting across someone else's site.

Original source ↗ ↗