A few months ago, we broke down what it looks like to advertise on ChatGPT. At the time, it was all about “context hints” – getting your message in front of people based on the topics they were typing into the prompt box. While it opened up an exciting new frontier for advertising, it lacked the precise audience control that data-driven marketers need to justify their budget.
That just changed.
OpenAI recently rolled out Custom Audiences for ChatGPT Ads. This update changes the game entirely, shifting the platform from an interesting contextual experiment into a legitimate performance channel where you can put your first-party CRM data to work.
Here is a quick breakdown of what this means for brands and how marketers can use it.
What’s the New Feature?
To put it simply, advertisers can now upload customer or prospect lists into the OpenAI Ads Manager to control who sees your ads.
However, this tool is strictly for first-party data. OpenAI’s terms are explicit: list rentals, third-party broker data, or scraped leads are completely off the table. This feature is designed for brands to securely activate the data they actually own.
You can upload standard customer identifiers like emails or phone numbers, including secure SHA-256 hashed files. Once processed, you can use the list in three ways:
- Campaign-Level Inclusion: Restrict a campaign so it only shows up for people on your list. This is great for hitting warm leads or loyalty members.
- Campaign-Level Exclusion: Block specific groups from seeing your ads. This is huge for acquisition campaigns where you don’t want to waste money showing ads to people who just bought from you last week.
- Audience Bid Adjustments: You can set bid multipliers from 0.1x all the way up to 10x at the ad group level to bid up aggressively on high-value VIP lists.
The Big Catch: You Need Scale
Before you try running a highly-targeted campaign, check the fine print. Each custom audience must have at least 25,000 matched users before OpenAI will let you use it.
Because you will always lose some data to bad email addresses, opt-outs, and match-rate drops, OpenAI recommends starting with a list of at least 100,000 users. For now, this tool is built for brands with decent-sized CRM databases, larger e-commerce lists, or robust loyalty programs.
How to Run a Clean Pilot
If you want to test this out, don’t just dump your entire database into the system and hope for the best. Treat it like a real CRM experiment.
Because OpenAI requires verified first-party records, the heavy lifting happens before you ever hit upload. That’s where the Response Labs team comes in. We work with our clients to cleanse, format, and segment data to ensure high match rates while strictly keeping your brand compliant.
When launching a pilot, we typically recommend defining three target groups to start:
- Recent Leads (0–90 days): An inclusion list for hot prospects who need a nudge to finish converting.
- Dormant Leads (91–365 days): A reactivation segment where you test a slightly different, incentive-driven message
- Existing Customers: A strict exclusion list to protect your acquisition budget from ad waste.
From there, we round out the pilot with a strong framework:
Combine Audiences with “Context Hints”
ChatGPT ads don’t use traditional keyword matching. You have to pair your audience list with conversations where your ad belongs. For a flooring brand, this could be “user planning a home renovation and looking for flooring ideas” or “user comparing local installation companies.”
Keep the Creative Helpful
People go to ChatGPT for advice, not to look at flashy banner ads. Focus your creative angles on cost clarity, decision support, or convenience, and use specific landing pages that match the user’s intent.
Measure for Incrementality
OpenAI lets you buy on a CPC or CPM basis. For a meaningful 30-to-45-day pilot, we recommend a media budget that allows you to gather statistically significant data (think $15,000 to $25,000). Most importantly, hold out 10% to 20% of your list from the campaign to compare downstream actions – like actual purchases – between the people who saw your ChatGPT ads and the people who didn’t.
The Compliance Guardrails
OpenAI’s terms are explicit: you must own this data, and your privacy policy must cover using consumer data for paid media matching. Strip out any sensitive CRM notes or commentary before you upload; the system only wants clean identifiers like emails or phone numbers.
The Bottom Line
With Custom Audiences, ChatGPT is moving past the experimental phase and becoming a serious player in performance media. It gives you a direct line to engage your audience exactly when they are asking deep, conversational questions relevant to your business.
Ready to see if your CRM data meets the scale requirements for a ChatGPT pilot? Contact the Response Labs team today to map out a strategy that proves real bottom-line lift.
