Advertisers in ChatGPT can now target users by email address, phone number or mobile ad identifier, and the ads can draw on what people type, 404 Media reported on October 15(opens in a new tab) after testing the ad system. Per Gadget Review's summary of OpenAI's help documentation(opens in a new tab), ads run for Free and Go users in selected regions, while Plus, Pro, Business, Enterprise and Edu accounts see none. The current conversation can shape which ad appears. With personalization on, signals from wider ChatGPT activity can be used too. Advertisers can upload customer lists, with at least 25,000 matched users needed to target an audience.
OpenAI says advertisers never see conversations, cannot tell which individuals matched their lists, and that uploaded files are deleted after processing. Take that at face value and the enterprise problem is still there. It is not that a vendor reads your prompts. It is that your employees' work context is now an input to an ad system you have no contract with.
The exposure sits in the gap between licensed and unlicensed use. An employee pasting a supplier dispute or a draft deal memo into a personal Free account is now feeding the topic of that work into ad selection. Turning personalization off does not end it, since ads still follow the current chat. The ad-free tiers are the paid ones, so the cleanest control is a funded seat, not a policy memo.
There is an upside for the marketing side of the house. ChatGPT is now a paid channel that can reach a known customer list. Before anyone uploads one, have counsel check that your privacy notices and consents cover sharing hashed customer emails with OpenAI for ad matching.
Count how many employees use ChatGPT on personal accounts. Price business seats for the heaviest of them. Then decide, in writing, whether your customer data goes into anyone's ad system.
Action items
ChatGPT now serves ads to Free and Go users, shaped by what they type. Paid business seats carry no ads. Know which of your employees sit on which side.
For the CISO
Count personal ChatGPT accounts used for work, and treat leaked AI API keys like leaked cloud credentials.
For the General Counsel
Before marketing uploads a customer list to ChatGPT Ads, confirm your privacy notices cover it.
For the CAIO
Move high-volume classification to small decision models, and plan the junior pipeline agents are thinning.
Researched and drafted by an automated workflow, then reviewed and edited by a human editor before publication. Every source is linked. See how we use AI here.
Also worth knowing
- 404 MediaHow ChatGPT Targets Ads To You Based On Your Prompts, Phone Number, and Email(opens in a new tab)
ChatGPT advertisers can target by email, phone or device ID, and ads can draw on prompts. Business and Enterprise seats carry no ads, so unlicensed work use is where your exposure sits.
- Help Net SecurityStolen AI API keys are paying for hackers' autonomous scans(opens in a new tab)
Bitsight found a criminal pentest agent, MONST3R-AI, that runs on stolen OpenAI, Anthropic and Google keys, so the key owner pays. Treat a leaked AI key like a leaked cloud credential and set spend caps.
- TechCrunchJev is Vercel's fastest-adopted AI model ever(opens in a new tab)
TypeSafe's Jev, a small decision model for classification, reached 35% of paid Vercel teams in two weeks. Move high-volume classification off general models and onto cheaper specialists.
- AnthropicHow employment is changing at Anthropic(opens in a new tab)
Anthropic says agent hours ran six to one against employee hours by August, while junior technical hires fell from 27% to 9% of hiring. Plan your own entry-level pipeline before agents quietly replace it.