Custom audiences & lookalikes for ChatGPT Ads

Relevance is the cheapest performance lever there is. With custom audiences you put your ChatGPT ads in front of exactly the people most likely to convert — existing customers, newsletter subscribers, cart abandoners — instead of anyone at all. And with lookalikes you reach people who resemble those customers. This article explains what both are, how they work technically and what to use them for in practice.

What is a custom audience?

An audience built from your own first-party data: a list of customers you upload and that OpenAI matches against the ChatGPT user base. Whoever appears in both sets becomes a member of the audience. You don't learn who that is — you can only use the audience as a whole in ad groups, to target or to exclude.

Uploading a list

• As a CSV with a header row and the columns email and phone; every row needs at least one value.
• Identifier types are detected automatically, you don't have to label anything.
• On upload, Chatvertise passes along filename, type and size — the API requires this metadata, otherwise it rejects the request.
• The exact format, the statuses and three pitfalls are covered in our upload guide.

After upload, matching runs asynchronously: the audience first reads “processing” and switches to “ready” once membership is built. Depending on list size, that takes minutes to hours. While it's processing it can neither be archived nor meaningfully tested in an ad group.

Lookalikes: reach with intent

From an existing customer list you can build a lookalike audience — users whose behavior resembles your best customers. That's the way to gain cold reach without losing relevance: instead of “everyone in Germany” you address “people like my buyers in Germany”. The quality of the lookalike depends on the source. A list of your most valuable customers produces a better lookalike than a list of everyone who ever opened a newsletter.

Typical uses

• Retargeting: re-engage cart abandoners or newsletter subscribers — in their own ad group with a higher bid, because the conversion probability is higher.
• Exclusion: exclude existing buyers from new-customer campaigns so budget doesn't flow to people who'd buy anyway.
• Scaling: lookalikes from high-value customers for cold reach, once the retargeting ad group shows which ads work.
• Seasonal reactivation: reach last year's buyers again before Black Friday — upload that list weeks ahead, because matching takes time.

A sensible start

Three ad groups in one campaign: one with the customer list and a slightly higher bid, one with the lookalike at the standard bid, one without an audience as a control. After a week, CPA and conversion rate per ad group show where the budget should go. Which metric stands for what is explained in our metrics glossary.

In Chatvertise you upload lists, build lookalikes and see the status of every membership job in one place — failed jobs with their reason, running ones with resume and cancel. Connect your API key and start for free.