Custom audiences are advertiser-uploaded lists of customer identifiers (raw or SHA-256 hashed emails and phone numbers) used as include or exclude targeting filters in ad campaigns. The pattern originated on Meta in 2012, spread to Google as Customer Match in 2015, and arrived in ChatGPT Ads on May 14, 2026. They are the bridge between an advertiser's first-party CRM data and an ad platform's logged-in user graph.
What are custom audiences?
Custom audiences are lists of customer identifiers (emails, phone numbers, or sometimes mobile advertising IDs) that an advertiser uploads to an ad platform and uses as a targeting filter on its campaigns. The mechanic is mature. Meta introduced Custom Audiences on Facebook in September 2012; Google launched its equivalent, Customer Match, on September 28, 2015. The two have been the foundational primitive for first-party-data activation on the dominant ad platforms for over a decade. OpenAI brought the same primitive to ChatGPT Ads on May 14, 2026, nine days after self-serve Ads Manager opened to every US advertiser.
The shape is consistent across platforms. The advertiser exports a list from a CRM or customer-data platform, runs the identifiers through a one-way hash (today's standard is SHA-256), uploads the hashed file, and the platform matches the hashes against its logged-in user graph. The advertiser then references that audience in campaign targeting as either an include filter (show ads only to these users) or an exclude filter (suppress ads from these users). The plaintext customer data never leaves the advertiser's environment; the platform never sees an unhashed email or phone number.
How custom audiences work in ChatGPT Ads
In ChatGPT Ads Manager Beta, custom audiences are created at the ad-account level and referenced from within a campaign during the New Campaign flow. The advertiser clicks Create include audience or Create exclude audience, names the audience, optionally adds a description, picks an identifier type (email or phone, one per audience), and uploads a CSV or TXT file of up to 512 MB. OpenAI's help center documentation on ChatGPT Ads basics describes static tracking parameters and conversion measurement; identifiers may be uploaded raw or already SHA-256 hashed, and OpenAI normalizes inputs during processing (lowercase, trim) and hashes any plaintext server-side before the match.
A 512 MB SHA-256 email file holds approximately 7.8 million hashed identifiers, more than enough for the largest B2B prospect lists most advertisers will plausibly assemble. Once created, the audience can be referenced from any campaign in the account. The campaign's Include slot restricts ad delivery to users matched against the list; the Exclude slot does the opposite, useful for suppressing ads from existing customers or known competitor employees.
Custom audiences vs context hints vs lookalikes vs retargeting
Four related ad-tech primitives sit in adjacent space and are easy to confuse.
| Primitive | What it is | Who provides the data | In ChatGPT Ads today? |
|---|---|---|---|
| Custom audiences | Advertiser-uploaded list of customer identifiers used as targeting filter | Advertiser (CRM data) | Yes (gated rollout, May 14, 2026) |
| Context hints | Advertiser-stated topical and intent signals at the ad-group level guiding when the engine matches an ad to a conversation | Advertiser (strategic input, no individual user data) | Yes (since launch) |
| Lookalike audiences | Algorithmic expansion of a seed custom audience to similar users the platform identifies | Platform (statistical model applied to seed list) | No (not yet announced) |
| Retargeting | A use case, not a primitive: ads served to users with prior interactions, typically driven by a pixel-based or audience-list source | Combination (pixel + custom audience) | Pixel exists; audience-based retargeting now possible via custom audiences |
The sibling-lens distinction with context hints is the most important. Context hints are signals about a conversation; custom audiences are signals about a user. Context hints answer "when should this ad appear in a session;" custom audiences answer "should this ad be eligible to appear at all for this user." A retargeting campaign typically combines both: context hint to pick the moment, custom audience to pick the audience.
Why custom audiences matter for AI Ads
Custom audiences mark the moment ChatGPT Ads stops being a purely contextual ad surface and becomes a first-party-data-activatable channel. Until May 14, 2026, ad delivery was matched on conversation context and intent alone. The advertiser could shape what topics to be relevant for; not who to be relevant to. Custom audiences add the second dimension.
The structural significance, surfaced by Juozas Kaziukėnas on LinkedIn the day the feature was first sighted, is that an identity layer is quietly forming underneath ChatGPT Ads. The advertiser's CRM and ChatGPT's logged-in user graph can now intersect. That intersection is where retargeting, suppression, and persona-based audience activation become possible.
How to use custom audiences in ChatGPT Ads
Four high-level applications cover most of the planned use cases.
- Retargeting: match a list of prior site visitors or trial signups against ChatGPT's logged-in graph; serve a follow-up ad inside relevant conversations.
- Customer suppression: exclude existing customers from prospect-acquisition campaigns to avoid wasting impressions on already-converted users.
- Account-list activation (ABM): for B2B teams running account-based marketing, the target-account contact list can be uploaded once and referenced across multiple campaigns.
- Competitor and employee exclusion: block known competitor staff and your own employees from your own campaigns to clean up reporting.
The hands-on walkthrough (exporting, segmenting, hashing, and preparing CRM lists today even while the rollout is gated) lives in our companion guide on ChatGPT Ads custom audiences.
Common misconceptions
Hashing protects PII completely
Hashing prevents the ad platform from seeing the plaintext identifier. That part is real. The match itself is still the privacy event, though. When you upload a hashed list and the platform reports back that some percentage matched, the platform has confirmed that those specific users in its graph exist in your CRM. That is a real informational transfer; it just does not include the email or phone number itself. Treat hashed-list uploads as a privacy disclosure to your data-protection officer, not as anonymization.
Lookalike audiences are coming next
They might be; nothing has been announced. Meta added Lookalike Audiences about a year after launching Custom Audiences; Google's Similar Audiences product arrived later still. The pattern is well-trodden but not inevitable on the same timeline. Treat lookalikes as a roadmap signal to watch, not a near-term planning input.
This only works on logged-in users
Correct, and that is the structural enabler. Most ChatGPT users are logged in (the product asks for an account on first visit), which is what makes the user-graph match possible. The custom-audience match rate on a clean B2B list will typically be lower than on Meta or Google because the relevant population that uses ChatGPT regularly is smaller, not because the matching is weaker.
Frequently asked questions
#What are custom audiences in simple terms?
Custom audiences are lists of customer emails or phone numbers you upload to an ad platform to tell it which specific users your ads should reach or skip. The platform hashes the identifiers, matches them against its logged-in users, and lets you target or exclude that group across your campaigns. The pattern is the same on Meta, Google, and now ChatGPT Ads.
#How do custom audiences differ from context hints in ChatGPT Ads?
Context hints describe a conversation; custom audiences describe a user. Context hints guide when an ad is eligible based on what the user is talking about. Custom audiences gate whether the ad is eligible at all based on who the user is. Most retargeting campaigns combine both: context for the moment, audience for the person.
#What happens to my customer email list when I upload it as a custom audience?
Your file goes to OpenAI's servers as CSV or TXT, up to 512 MB. You can pre-hash the identifiers with SHA-256 or upload them raw; OpenAI normalizes inputs (lowercase, trim) and hashes server-side before matching against its user graph. The platform reports a match rate but does not return individual matches back to you.
#Can I create lookalike audiences in ChatGPT Ads?
Not yet. The custom audiences feature shipped on May 14, 2026 with include and exclude filters only. Lookalike functionality, where the platform algorithmically expands a seed list to similar users, has not been announced. Meta added lookalikes about a year after launching Custom Audiences, so the pattern is established, but no ChatGPT Ads timeline exists today.
#Are custom audiences available to every ChatGPT Ads advertiser today?
The feature is in gated rollout as of May 14, 2026. Self-serve Ads Manager opened to every US advertiser nine days earlier on May 5, but custom audiences are enabled on a per-account basis. Most advertisers will see them appear in the campaign flow over the coming weeks. Preparing hashed CRM lists in advance is the right move while you wait.
