ChatGPT Ads Target Emails, Phone Numbers and Prompts
A 404 Media test found that ChatGPT advertisers can upload contact lists and use conversation topics to guide ad targeting.
👁 118 reads
Journalists at 404 Media did something no newsroom had done before: instead of asking OpenAI how ChatGPT ad targeting works, they became advertisers. They bought two campaigns, went through the ad dashboard and reported which fields they had to fill in. Their ad copy was pointed in the best journalistic sense: “ChatGPT isn’t private. OpenAI hires humans to read users’ chats”, linking to their earlier investigation into Project Lily.
The experiment revealed details that aren’t apparent from OpenAI’s public documentation. Most notably, advertisers don’t just choose broad interest categories: they can upload lists of specific people using email addresses, phone numbers or Android Advertising IDs. They can also suggest the kinds of conversations in which their ads should appear.
Three types of identifiers and a 25,000-record threshold
I counted the different identifiers OpenAI accepts from advertisers, because that’s the heart of the issue. There are five options across three categories: plain-text email addresses, hashed email addresses, plain-text phone numbers, hashed phone numbers and Android Advertising IDs. “Plain-text” means exactly what it sounds like: an advertiser enters real email addresses or phone numbers in OpenAI’s dashboard, and OpenAI checks which ones belong to actual ChatGPT accounts. Those people can then be targeted.
There’s also a threshold. To use a list to include people in a campaign, it must contain more than 25,000 records. Below that number, it can only be used to exclude people. In theory, the threshold guards against targeting individuals or very small groups. In practice, it also means that only organizations with a database of at least 25,000 records can use this form of targeting. Many ordinary businesses won’t have a list that large; a debt collection company, a loan provider or a data broker might.
That threshold is worth keeping in mind when weighing who can use the system. It is a substantial barrier for small advertisers, but it doesn’t change the underlying fact that uploaded contact details can be matched against ChatGPT accounts.
Context hints: ads tied to what you ask about
The other feature 404 Media tested was a field called “context hints.” OpenAI describes it this way:
“Describe the conversations, topics, or keywords where your products or services may be relevant; these hints guide matching, but are not exact-match rules.” - OpenAI advertising documentation, quoted by 404 Media
The journalists entered topics such as people looking for information about privacy, the privacy of ChatGPT conversations, and current technology and AI topics. The ad appeared. That suggests the content of a prompt - what users type into a chat window, often assuming they’re speaking privately to a machine - can act as an advertising signal. It may not be a literal keyword match, but it can still guide ad delivery.
Zach Edwards of DecryptAds, a company that collects online advertising data for research, told 404 Media that OpenAI’s ad products work like those of many large companies beginning to monetize their audiences: they “boil the frog” and hope users don’t notice the privacy consequences.
How ChatGPT ads compare with Google and Meta
Uploading a customer list to an advertising platform isn’t new. Google calls its product Customer Match; Meta calls its version Custom Audiences. Here’s how the three platforms compare on the data advertisers can upload and the audience thresholds reported in this comparison.
| Platform | Accepted identifiers | Minimum list size | Content signal |
| ChatGPT (OpenAI) | Plain-text and hashed email addresses; plain-text and hashed phone numbers; Android Advertising ID | 25,000 to include people; no minimum to exclude them | Prompt content through context hints |
| Google Ads | Hashed email addresses and phone numbers, mailing addresses, mobile identifiers | About 1,000 matched users | Search history and activity across Google services |
| Meta Ads | Hashed email addresses and phone numbers, mailing addresses, mobile identifiers | About 100 initially | Activity in apps and on websites with a pixel |
The difference that stands out is OpenAI’s acceptance of plain-text email addresses and phone numbers. Google and Meta have long required advertisers to hash data on their own systems first, so the platform receives a hash rather than the original contact details. That’s not just a technical distinction; it’s a form of data minimization. According to 404 Media’s test, OpenAI allows advertisers to enter data in plain text, meaning the company receives the advertiser’s contact list in readable form and checks it against its own user database.
The other difference is the threshold. Google’s and Meta’s lower thresholds may sound worse for privacy, but those platforms have other safeguards, including restrictions on targeting sensitive categories, consent requirements and audits. OpenAI’s 25,000-person threshold is the main visible limit, and it also screens out smaller advertisers.
What the GDPR means for ChatGPT ad targeting
Three GDPR provisions are particularly relevant. Article 6 requires a lawful basis for processing. For ad targeting based on matching customer lists, EU supervisory authorities have consistently taken the position that legitimate interests alone are not enough and that consent is required. Article 5(1)(c) requires data minimization; accepting plain-text contact details when a cryptographic hash would suffice may be difficult to justify. Article 22 concerns automated decision-making.
There’s another question that deserves more attention: who controls the data in a list uploaded by an advertiser? If a debt collection company uploads 40,000 phone numbers belonging to debtors to show them ads in a chatbot, it is processing those details for a purpose the debtors may never have agreed to. The advertiser bears responsibility, but OpenAI, which receives and checks the list, would not necessarily be a neutral intermediary in that arrangement.
As of October 16, 2026, no public proceeding by a national data protection authority against OpenAI over its advertising system had been announced in the account covered here. OpenAI’s European headquarters are in Ireland, making Ireland’s Data Protection Commission the lead authority under the GDPR’s one-stop-shop mechanism. Other EU authorities may receive complaints from people in their countries and refer them through the relevant process. The Italian Garante’s 2023 ChatGPT case ended in a €15 million fine, but it concerned the legal basis for training models, not advertising.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →What you can do while the questions remain open
OpenAI launched ChatGPT ads in the US in February and is gradually expanding them to other markets. Ads are shown to users on the Free and Go plans, and, starting this month, alongside image generation too. Availability varies by market, so check OpenAI’s current information for your country.
As a practical step, open your ChatGPT account settings and go to Data Controls. Check the controls for personalization and for using your conversations to improve the model. Turning off the latter does not remove ads, but it limits what is used for further processing. If you’re on the Free plan, upgrading to Plus is the lowest-cost technical way mentioned here to avoid ads: it costs $20 per month, or $240 per year.
Also consider the email address linked to your ChatGPT account as an advertising identifier. If you use the same address for newsletters, loyalty programs and shopping accounts, an organization with a list of more than 25,000 records may be able to match it to your ChatGPT account. Using a separate address for chatbot accounts may be a sensible privacy measure.
If you use ChatGPT on Android, you can find an option in your device settings to delete the advertising ID. That removes one of the five targeting identifiers described in 404 Media’s test.
What concerns me most is what people ask ChatGPT: questions about symptoms, divorce, debt or whether a manager is breaking the law. They may ask because talking to a model doesn’t feel like talking to another person, so the usual instinct to be discreet doesn’t kick in. The 404 Media test is a reminder to bring that instinct back - not because someone on the other side is necessarily malicious, but because an advertising dashboard now has a field for describing the conversations in which an ad should appear.
Facts checked October 16, 2026. Sources: 404 Media - How ChatGPT Targets Ads To You Based On Your Prompts, Phone Number, and Email; GDPR, EUR-Lex; Garante per la protezione dei dati personali; Google Ads - Customer Match.
