A post shared on October 16, 2026 says OpenAI "lets anyone target specific people with ads via phone or email." The underlying tool is real in outline: ChatGPT Ads has a custom-audience feature where advertisers upload contact lists. What we could not verify is the strongest reading of the claim, that an advertiser can pick out an individual.
We could not open the original report behind that post, so this is a claimed-versus-verified breakdown built from OpenAI's own help center and trade-press coverage of the feature. For background on the ad product itself, see our ChatGPT ads launch guide.
TL;DR
| Question | What the sources say |
|---|---|
| Does a contact-list targeting tool exist? | Yes, reported by trade press since spring and early summer 2026 |
| Can an advertiser target one named person? | Not supported by the sources we read; a 25,000 matched-user minimum is reported |
| What can be uploaded? | Emails or phone numbers, raw or SHA-256 hashed (PPC Land) |
| Who sees ads? | Reported: free and Go tier adults; Plus, Pro, Business ad-free |
| Can users opt out? | OpenAI documents a Marketing Privacy control for its own promotion |
| Primary OpenAI page on audiences | We did not find one |
What exactly is the claim?
The viral framing is that anyone can aim ChatGPT ads at specific individuals by supplying a phone number or email. That is a statement about precision. Two different things hide inside it: matching a list against OpenAI accounts, which is standard in ad tech, and singling out one person, which is a much bigger privacy story.
How custom audiences work, according to the trade press
PPC Land reports that Juozas Kaziukenas spotted the feature on May 14, 2026 in a gated rollout, after Ads Manager opened to all US businesses on May 5. Other search-industry outlets describe an early July sighting and a short-order announcement afterwards, so the exact dates conflict across sources and you should treat them as unsettled.
The mechanics, as PPC Land describes them:
- Advertisers upload a CSV or TXT file of up to 512MB containing emails or phone numbers and name the audience.
- Raw or SHA-256 hashed values are accepted; OpenAI normalizes identifiers on ingestion.
- At the ad-group step, an include option limits delivery to the list, and an exclude option keeps ads away from it.
- Lookalike audiences do not appear to exist yet.
Several outlets, including Shopifreaks, report that an audience needs at least 25,000 matched users before inclusion targeting and bid adjustments work, with 100,000 recommended. PPC Land says minimum sizes are not addressed in its piece, so this figure comes from other outlets and we have not seen it in an OpenAI document.
Why the 25,000 floor matters
If the floor is accurate, uploading one person's email gets you nothing: the audience will not run. That is the main evidence against the "target specific people" reading. It matches how Google and Meta handle customer-match lists, where minimum sizes exist partly to prevent individual-level targeting.
It does not make the system harmless. A list of 25,000 can still be a narrow group such as a company's employees, a political donor file, or a leaked database. Exclusion works the other way: an advertiser can suppress ads to people on a list. And advertisers learn only an approximate matched count, which reveals how many of their contacts have ChatGPT accounts. Whether that counts as a privacy leak depends on your view of match-rate disclosure, which every major ad platform reports.
A short timeline of ChatGPT ads and data use
The custom-audience tool is one step in a rapid build-out, using dates reported by PPC Land:
| Date (2026) | Event |
|---|---|
| March 26 | Ads pilot expands beyond the US to Canada, Australia and New Zealand |
| April 30 | OpenAI updates its US privacy policy: it discloses receiving advertiser purchase data for measurement, sharing user information with marketing partners for third-party targeting, and using personal data to promote its own products |
| May 5 | Ads Manager opens to all US businesses |
| May 14 | Custom audiences spotted in a gated rollout (other outlets say early July) |
| July | Ads launch for advertisers more broadly, covered in our launch guide |
| October 16 | Viral post reframes the feature as targeting "specific people" |
Read in order, the pattern is the standard ad-platform playbook: first contextual placement, then measurement, then audience controls. The privacy-policy change in April is the document to read if you want OpenAI's own wording on what data flows to advertisers.
Questions people are asking
Is this the same as Facebook's custom audiences? Structurally, yes. Customer-list matching is a long-standing industry technique: an advertiser hashes its customer emails, the platform hashes its own users' emails, and matches are counted. The new part is the surface, a chat product where people may have shared far more personal context than they would on a social feed. Whether OpenAI uses conversation content for ad targeting is a separate question; the help-center page above says conversations are not shared with marketing partners for OpenAI's own promotion, and trade-press descriptions of custom audiences mention only list uploads, not conversation signals.
Does hashing protect me? Partly. A hashed email cannot be read directly, but anyone who holds a list of candidate emails can hash them and compare. That is exactly how matching works, so hashing prevents casual exposure rather than determined matching. Our OBI cookie post makes the same point about hashed identifiers.
Can I find out which advertisers have my email? None of the sources we read describes such a feature. PPC Land says the platform does not show individual matched users to advertisers, and says nothing about showing users which lists they appear on.
Will paid plans stay ad-free? Search-industry reports say Plus, Pro and Business are ad-free today. That is a product decision OpenAI can change, so check the plan page before assuming.
Why does the 25,000 number matter? It is the best available evidence that the tool is built for groups, not individuals. If OpenAI lowered the floor, the story would change, and we would update this post.
What OpenAI itself documents
OpenAI's help-center page, How We Promote OpenAI on Third-Party Properties, covers a different flow: OpenAI sharing hashed identifiers with marketing partners to advertise its own products elsewhere. Key points from that page:
- OpenAI says it does not share conversations, or uploaded documents, images and videos, with those partners.
- It shares identifiers such as a cookie ID, an IP address, or a hashed email or phone number, plus basic commercial and browsing information.
- Users can opt out under Settings, Data Controls, Marketing Privacy.
This page does not describe advertisers uploading lists to ChatGPT, so it cannot confirm or refute the viral claim. It is relevant because it shows the same hashing and matching approach in use, and because it contains the one opt-out control OpenAI documents. We covered the related web tracking in ChatGPT's OBI cookie and cross-site ad tracking.
What is not confirmed
- Whether the 25,000 minimum is current policy, and whether it applies to every ad type.
- The countries where custom audiences are live (one report says seven, another does not say).
- Whether free-tier users can opt out of being matched to an advertiser's list, as distinct from OpenAI's own marketing.
- How raw (unhashed) uploads are stored and for how long.
- Whether the October 16 post reflects new reporting or the older feature.
How we weigh it
The claim is partly true and partly overstated, on the evidence available. A contact-list upload exists. Single-person targeting is not documented, and the reported minimum audience size argues against it. The legitimate concern is broader: a product that holds conversational data is now an ad platform, and the controls users have are still thin. That tension has been building through the product's history, from the $1 billion ads revenue run rate to visual ad tests and the Notch founders joining the ads team.
What should you do?
If you use ChatGPT: open Settings, Data Controls and check Marketing Privacy. Remember that ads are reported to appear on free and Go tiers; paid tiers are reported ad-free. Avoid putting sensitive identifiers in a free account if that matters to you.
If you are an advertiser: hash identifiers before upload, keep consent records for the contacts you upload, and expect only a portion of your list to match.
If you write about this: ask for the primary source. A screenshot of an Ads Manager upload screen shows a list tool, not individual targeting.
Method note
We searched OpenAI's help center, Ads Manager coverage in search-industry outlets and the viral post's subject. We did not find an OpenAI announcement page for audiences, so everything about the feature's mechanics is secondhand. If OpenAI publishes documentation or the original report becomes accessible, we will add an update section here.
Details are accurate as of October 16, 2026 and may change; verify against OpenAI's ads documentation. Follow @explainx_ai for updates.