Small customer lists just became useful for ChatGPT advertisers, even when they are too small to target. OpenAI has amended the 25,000 matched-user floor it set on custom audiences in July: that minimum still blocks small inclusion lists, but audiences below it can now be used for exclusions. Advertisers with only a handful of thousand past customers or lapsed buyers can finally block those users from a campaign, rather than losing the list entirely.
The change is one of seven items in a product update PPC Land reported OpenAI circulated to advertisers during the week of August 31, 2026. The rest of the update matters more for agencies managing scale than for any single campaign: list members can now be added or removed without rebuilding the audience, identifier types can be mixed in one request, lists can exceed 5 million members, and audience-size ranges report at finer granularity. Google Advertising ID joins email and phone number as a supported match key, extending targeting to app advertisers who lack contact details.
Conversion matching also gained fields. The Measurement Pixel now takes a hashed phone number along with first name, last name, postal code and region, while the Conversions API gains plural list fields, so one customer record can carry several hashed emails or phone numbers at once. That lifts match rates, but it also broadens the pool of personal data collected right as automatic advanced matching, which pulls form data in hashed form without asking, is already running on existing pixels. OpenAI has not explained how these newly added fields intersect with consent requirements in the European markets it just opened.
The most consequential item is not a metric but a tool. A new ChatGPT Ads Manager plugin lets advertisers create, edit and troubleshoot campaigns from inside ChatGPT or Codex using plain-language instructions, and it previews every change before applying it. That confirmation step puts OpenAI on the cautious side of a boundary advertising platforms have spent the year sketching out. Meta gave Claude and ChatGPT write access to its ad system through an open connector in April and opened the underlying protocol to any developer in July. X shipped a 23-tool Ads MCP server, ten of them writes, in August. Microsoft went the other direction, launching a read-only Advertising MCP server in June.
OpenAI’s plugin, by contrast, runs only inside its own surfaces. Whether the plugin offers an endpoint built on the Model Context Protocol, one that a competitor’s agent could call, is not addressed, even though OpenAI’s own coding tools now call Meta’s. Until OpenAI clarifies that, agencies standardizing on one agent across platforms cannot fold ChatGPT Ads into that workflow the way they can Meta’s or X’s.
Two other pieces round out the release. Campaigns running on a total budget now smooth their spending across the whole flight instead of burning through it early, though OpenAI has not published a variance cap, unlike Google, whose equivalent controls publish a ceiling set at double the average daily spend. And a future campaign type will bill on impressions while optimizing toward view-through conversions, reversing a two-week-old design that walled off post-view data from bidding and billing entirely.
For search and performance teams running paid alongside organic, the practical read is narrower than the announcement. The exclusion fix removes a real operational blocker for suppression campaigns, and the plugin signals OpenAI is racing toward the agentic ad-buying model Meta and X already ship. Teams testing ChatGPT Ads should audit suppression lists now that small ones qualify, and hold off building automation around the plugin until OpenAI states whether it will expose a protocol other agents can call.
PPC Land reported the update, based on a product email OpenAI circulated to advertisers, on September 3, 2026.
