ChatGPT Ads Custom Audiences: Upload, Match Thresholds, and Bid Adjustments
What Custom Audiences Do in ChatGPT Ads
ChatGPT Ads custom audiences take the lists you already own - customers, qualified prospects, newsletter contacts - and apply them as targeting controls. OpenAI documents three uses. Campaign-level inclusion limits campaign eligibility to people found in one or more selected audiences. Campaign-level exclusion does the reverse: delivery is prevented to people in the selected audiences, the classic suppress-and-save-play against existing customers and recent purchasers. Ad-group audience bid adjustments are the third shape - a bid multiplier raises or lowers an ad group's maximum bid when the viewer matches a selected audience, letting you bid harder for high-value customers and more selectively for low-priority ones. All three are audience-level controls: Ads Manager never shows individual matched users, and you cannot pick specific people from a list.
Preparing a List the Platform Can Match
The file requirements are concrete. Uploads accept valid UTF-8 CSV or TXT files up to 500 MB, with TXT files carrying one identifier per line and CSV files carrying a header row that names each identifier type: email, phone_number, email_sha256, phone_number_sha256, or gaid. The five types mix freely inside a single CSV upload, each in its own column. Hashed formats have exact definitions: email_sha256 is the 64-character SHA-256 hexadecimal digest of the normalized email, while phone_number_sha256 is the digest of the normalized E.164 phone number. The API normalizes and hashes GAID values itself, so raw Google Advertising IDs can go in as they are. This is the same normalize-then-hash discipline as CAPI user data hashing, applied to list membership instead of event payloads. Audiences can grow beyond 5,000,000 members, added in multiple uploads after creation.
Creating the Audience in Ads Manager
The UI path runs through Tools, then Audiences, then Create audience: name the audience, upload the file, pick the identifier type for TXT uploads or let Ads Manager infer it from CSV headers, and confirm. Processing usually takes about 20-30 minutes depending on file size, and the audience status tells you where it stands - Processing, intermediate states like Upload pending or Indexing, then Ready, or Failed when the file needs a re-check against format rules. One accounting detail saves confusion later: uploaded rows are not the same as matched users, because invalid values, duplicates, and identifiers that fail to match never count toward the matched total - the share that does match is the list's match rate.
The 25,000 Matched-User Threshold
The threshold decides what a small list is allowed to do. Audiences with at least 25,000 matched users unlock all three uses - inclusion, exclusion, and bid adjustments. Below that floor, an audience still shows as Ready but carries an Exclusion only note: suppression works, targeting does not. For inclusion targeting or bid adjustments OpenAI recommends audiences of at least 100,000 users, which is the realistic bar for treating a list as a targeting asset rather than a suppression list. The 25,000 floor is what finally settles the softer claims floating in third-party coverage - it is a documented, status-visible gate, not a rumor.
Keeping Lists Current: Add, Remove, Replace, Merge
List maintenance does not require rebuilding anything. After an audience exists, its members can be added, removed, or fully replaced in Ads Manager without recreating the audience or touching the campaigns and ad groups already using it - the audience ID stays stable while membership changes. The API exposes the same operations with a clear contract: Create from a file or a bare name, Add and Remove for incremental changes, Replace for swapping the full list while keeping the ID, and Merge combining 2 to 64 existing audience IDs into a new, independent union audience. Small updates run as inline requests; bulk changes ride on files.
Where Custom Audiences Stop
Two boundaries close the picture. Geographic: custom audiences are not supported for campaigns targeting the European Economic Area or Switzerland, where personalized ads are not yet available - a campaign pointed at those markets cannot lean on list targeting at all. Conceptual: everything stays at audience level, with no visibility into individual matched users and no way to select specific people. And one rule sits above the mechanics: upload only first-party data you have the right to use for ads, with the required notices, consents, and legal bases in place - the list is a liability as well as an asset if the permission chain is broken.
Wire Audiences Into a Working Funnel
Audiences control entry into the auction, and the campaign setup guide shows where to attach them; the conversion tracking stack decides what OpenAI learns from the people who convert. A suppression list without server-side conversion delivery still leaks budget to untracked outcomes, so treat the two as one checklist. MOST delivers tracked conversions server-side to ChatGPT Ads alongside Meta, TikTok, Reddit, Snapchat, and Pinterest, and the free Pixel Activator validates one event at pixel.way2.us before the audience-driven spend scales.