404 Media bought actual ads inside ChatGPT to find out how advertiser targeting really works — and what they discovered affects roughly a billion monthly users. Here's what it means for you and your family.
If you've chatted with ChatGPT lately, you may have noticed something new: ads appearing inside your conversations. Sponsored messages now show up for Free and Go tier users, sitting right alongside answers — and, more recently, next to image generation results too.
But here's the question almost nobody is asking: how does ChatGPT decide which ad to show you? This week, the investigative tech outlet 404 Media decided to find out — not by reading OpenAI's marketing materials, but by doing what any advertiser can do: buying ads themselves and testing the system from the inside.
What reporter Joseph Cox found, published on October 15, 2026, is the kind of story that should make every ChatGPT user pause. Advertisers can target you based on your email address, your phone number, the unique advertising ID on your Android phone — and the topics of your conversations. Let's break down exactly what that means, what OpenAI says in response, and the practical steps you can take.
⚡ The Facts at a Glance
- Who investigated: 404 Media (reporter Joseph Cox), published October 15, 2026 — they ran two real ad campaigns inside ChatGPT to test the targeting system
- Method 1 — Context hints: Advertisers describe conversation topics or keywords; OpenAI's docs say these "guide matching but aren't exact-match targeting rules"
- Method 2 — Uploaded lists: Advertisers can upload raw or hashed emails, phone numbers, and Android advertising IDs to build custom audiences
- The 25,000 rule: Inclusion targeting requires at least 25,000 matched users; smaller lists can only exclude people from ads
- OpenAI's response: Uploaded files are deleted after processing (usually within 24 hours); advertisers can't see which individuals matched
- The catch: The "ads personalization" off-switch doesn't fully work — OpenAI says ads "may still be based on your current chat"
- Scale: ChatGPT has roughly 1 billion monthly users; 404 Media counted at least 43,410 ads across 415,289 placements in the ChatGPT Ad Library
To test the system, 404 Media did something deliciously cheeky: they bought ads about ChatGPT privacy and ran them inside ChatGPT itself. The ads read: "ChatGPT isn't private. OpenAI hires humans to read users' chats. Learn more."
That message wasn't random. It pointed to 404 Media's own earlier investigative series, "Project Lily," which reported that human contractors review some user prompts — a finding that rattled a lot of people who assumed their chatbot conversations were private.
But the stunt was just the delivery vehicle. The real story was what happened behind the scenes in OpenAI's advertiser dashboard — because that's where Cox could see exactly what targeting options any advertiser gets.
Why This Investigation Matters 🔥 KEY CONTEXT
ChatGPT's ad business is still young — ads launched in the US in February 2026, roughly eight months ago, and the program has been expanding to new territories since. That means the targeting infrastructure 404 Media just exposed is the foundation, not the finished building. What advertisers can do today is the floor of what's coming, not the ceiling. Understanding it now — before it becomes as entrenched and opaque as the rest of the ad-tech industry — is exactly the right moment to pay attention.
The first targeting method is about what you're talking about right now. When setting up a campaign, advertisers are asked to provide what OpenAI calls "context hints" — described in OpenAI's own ad product documentation this way:
"Describe the conversations, topics, or keywords where your products or services may be relevant; these hints guide matching but aren't exact-match targeting rules."
Think of it like Google keyword ads, but looser. An advertiser selling, say, a sleep aid might describe conversations about insomnia, bedtime routines, or stress. OpenAI's system then decides which chats are relevant enough to show the ad.
404 Media tested this directly: for their privacy-focused ads, they entered context hints describing "people searching for topics that discuss privacy, or the privacy of ChatGPT chats themselves, or people who are asking about current technology and AI topics." The ads then appeared in relevant conversations.
Here's the part worth sitting with: your conversation topic is an ad-targeting signal. If you're asking ChatGPT about a health concern, a financial worry, or a relationship problem, that topic can determine which sponsored message appears beneath the answer. OpenAI says advertisers don't receive your actual chats — the matching happens inside OpenAI's systems. But the topic of your most personal questions is, functionally, shaping the commercial messages you see.
The second method is the one that surprised even seasoned ad-tech watchers. Advertisers can upload lists of real people's contact details to build what's called a custom audience. According to 404 Media's testing and OpenAI's documentation, the supported identifiers include:
- Raw email addresses — plain, readable emails
- Raw phone numbers — plain, readable numbers
- Hashed versions of emails and phone numbers (digitally scrambled, harder to read)
- Android advertising IDs — the unique identifier assigned to every Android device
OpenAI's system then checks whether those identifiers match real ChatGPT accounts. If they do, those users become eligible to see the advertiser's ads.
Now, before you panic: there are guardrails. The big one is the 25,000-user minimum — an uploaded list must match at least 25,000 ChatGPT users before it can be used to include people in ad targeting. Smaller lists can only be used to exclude people. And OpenAI states that advertisers cannot see which individuals matched their list, nor select specific people within an audience. You can't upload one ex's email and haunt their chats — the system won't allow audiences that small.
But the deeper concern, as 404 Media frames it, is about awareness. Customer-list targeting is standard practice across the advertising industry — Meta, Google, and others all offer versions of it. The difference is that most of ChatGPT's roughly one billion monthly users have no idea this machinery exists behind the friendly chat window. When you see an ad in ChatGPT, you probably don't think: "Someone, somewhere, may have uploaded my phone number into OpenAI's systems." But per this investigation, that's a real possibility.
The Unanswered Hashing Question 🔥 WATCH THIS
One of the most important open questions from 404 Media's reporting: does OpenAI hash the raw personal data advertisers upload? Hashing converts a readable email into a scrambled code, adding a layer of protection. Meta publicly states that it hashes such uploads. OpenAI, per 404 Media, has not said whether it does — even though it accepts raw, readable uploads. OpenAI does state that uploaded files are "deleted after processing, which is usually within 24 hours of upload." Deletion is good; but what happens during those 24 hours, and in what form the data sits, remains undisclosed. This is worth watching.
ChatGPT does offer an "ads personalization" setting. Sounds reassuring — until you read OpenAI's own description of what it does:
"Allow ChatGPT to use your past chats, activity, and preferences to select ads. Ads may still be based on your current chat."
Read that last sentence again. Even with personalization turned off, ads can still be matched to whatever you're discussing right now. As one newsletter covering the story put it: "off" means "less," not "none."
This is a pattern worth recognizing. Across the tech industry, privacy controls increasingly offer the feeling of control without the substance of it. You flip a switch labeled "personalization," reasonably assuming it stops the targeting — but the most immediate form of targeting (your current conversation) continues untouched. For a family audience, this matters: a parent helping a child with homework in ChatGPT, or a teen asking about something sensitive, is still generating ad-targeting signals in that moment, switch or no switch.
Let's bring this home. You're not an advertiser, and you're probably not going to upload anyone's phone number to an ad platform. So why should this story matter at your kitchen table?
Because the mental model most people have of ChatGPT is wrong — and it matters most for kids. Many users, especially younger ones, treat a chatbot conversation like a private diary: a safe space to ask anything. The reality, per this investigation plus 404 Media's earlier reporting, is layered:
- Your topics shape your ads. What you ask about influences the commercial messages shown to you — even with personalization off.
- Your identity can be an audience. If your email or phone number sits in some company's customer list, it can become part of a ChatGPT ad audience.
- Humans may review chats. 404 Media's "Project Lily" reporting found contractors review some prompts — a separate issue, but part of the same privacy picture.
- None of this is clearly disclosed in the moment. There's no banner saying "this ad appeared because of your conversation topic" or "you're seeing this because your email matched an advertiser's list."
This doesn't mean you should stop using ChatGPT. It's an enormously useful tool, and for many families it's become as routine as a search engine. But it does mean the right mental model is: treat a ChatGPT conversation like a search box that can show ads — because that's now literally what it is. You wouldn't type your Social Security number into Google; apply the same instinct here.
It's also worth noting the regulatory backdrop. The same week this investigation published, Washington's new Super Intelligence Force held its first meeting on AI oversight, the FTC's probe of the major AI labs kept expanding, and eight states sued OpenAI over child safety. The rules for AI advertising are being written in real time — which is exactly why informed users have an advantage right now.
1. Turn off "ads personalization" — knowing what it does and doesn't do. In ChatGPT settings, disable it. It won't stop ads or all targeting, but it does reduce the use of your past chats, activity, and preferences. Less is better than nothing.
2. Keep sensitive topics general. You don't need to quit ChatGPT — just apply search-engine instincts. Avoid typing things like full names, addresses, account numbers, medical record details, or other people's private information into any chatbot.
3. Talk to your kids about the "private diary" myth. Teens especially tend to treat AI chats as confidential. A calm, brief conversation — "hey, just so you know, ChatGPT shows ads now, and what you ask about can affect which ads you see" — goes a long way. No lecture needed.
4. Remember the current-chat loophole. Even with personalization off, your current conversation can trigger ads. If you're discussing something sensitive, be aware that a related sponsored message may appear. That's the system working as designed.
5. Watch the hashing question. Whether OpenAI hashes raw advertiser uploads is still unanswered. If stronger data-broker rules arrive — something even 404 Media's sources are calling for — this space could change fast. We'll keep tracking it.
Frequently Asked Questions
How can advertisers target me in ChatGPT?
According to 404 Media's investigation, advertisers have two main methods. First, "context hints" let them describe conversation topics or keywords where their ads may be relevant — OpenAI says these guide matching but are not exact-match rules. Second, advertisers can upload customer lists containing email addresses, phone numbers (raw or hashed), or Android advertising IDs, which OpenAI matches against ChatGPT accounts. Inclusion targeting requires at least 25,000 matched users.
Can an advertiser target me specifically with my email address?
Not individually. OpenAI requires uploaded audience lists to match at least 25,000 users before they can be used for inclusion targeting — smaller lists can only be used to exclude people from ads. OpenAI also states that advertisers cannot see which individuals matched their list or select specific people within an audience. However, if your email or phone number is on a large uploaded list that matches enough ChatGPT accounts, you could be included in that ad audience without ever being told.
Does OpenAI share my ChatGPT conversations with advertisers?
OpenAI says advertisers do not receive your chats or personal details. The "context hints" system works in the other direction: the advertiser describes topics, and OpenAI's system decides which conversations are relevant. That said, 404 Media's reporting notes that your conversation topics do influence which ads you see, and the company has previously confirmed that human contractors may review some chats — a separate practice the outlet exposed in its "Project Lily" reporting.
Can I turn off personalized ads in ChatGPT?
There is an "ads personalization" setting in ChatGPT, but 404 Media found it does not stop ads or tracking entirely. OpenAI's own description of the setting says: "Allow ChatGPT to use your past chats, activity, and preferences to select ads. Ads may still be based on your current chat." In other words, turning personalization off means less targeting, not none — ads can still be matched to whatever you are discussing right now.
Does OpenAI hash the personal data advertisers upload?
This is unclear, and it is one of the open questions 404 Media raised. OpenAI accepts both raw (readable) and hashed uploads of emails and phone numbers, but the company has not said whether it hashes raw uploads itself before processing. Meta, by comparison, publicly states that it hashes such data. OpenAI says uploaded files are deleted after processing, usually within 24 hours of upload.
What can I do to protect my privacy in ChatGPT?
Five practical steps: turn off "ads personalization" in ChatGPT settings (it reduces but does not eliminate targeting); avoid typing highly sensitive personal information into any chatbot; remember that your current conversation topic can still trigger ads even with personalization off; periodically review what data you have shared with AI services; and treat chatbot conversations like a search box that can show ads — because that is now literally what they are.
404 Media's investigation doesn't reveal a scandal in the legal sense — customer-list ad targeting is, as the outlet itself notes, standard industry practice. What it reveals is an awareness gap: roughly a billion people use a product whose ad-targeting machinery most of them have never heard of, controlled by switches that don't fully switch off, built on personal data whose handling has open questions.
The fix isn't panic — it's clarity. Know that your topics shape your ads. Know that your email or phone number can place you in an audience. Know what the off-switch actually does. And teach your kids the same. The most powerful privacy tool isn't a setting buried in a menu — it's an informed user who knows what they're looking at.
More: 8 States Just Sued OpenAI: Is ChatGPT Safe for Kids? | Super Intelligence Force Holds First Meeting | Anthropic Pitches Its Own AI Safety Bill as the FTC Probe Hits | NYC Takes on Grok: Council Hearing Exposes AI Image Scandal
Sources
- 404 Media, "ChatGPT Ads Target Your Phone Number, Email" — October 15, 2026 (original investigation; paywalled)
- Gizmodo, "Here's How ChatGPT Is Targeting You With Ads" — October 16, 2026
- MobileSyrup, "ChatGPT lets advertisers target you based on your personal data" — October 16, 2026
- The Neuron Daily, "ChatGPT Ads Can Target You by Email and Prompts" — October 16, 2026
- Gadget Review, "ChatGPT Can Target Ads With Your Prompts, Email, and Phone" — October 16, 2026
No comments:
Post a Comment