ChatGPT ads can now use your customer list. Read the fine print

Advertisers in ChatGPT can now build Custom Audiences from their own customer data, according to OpenAI's help page on setting them up. A list can hold email addresses, phone numbers, SHA256 hashed versions of either, or Google advertising IDs from Android phones. OpenAI matches the list against ChatGPT users and then lets the advertiser do one of three things: show ads only to those people, skip them, or bid more or less when they appear.

Those bid changes run from 0.1x to 10x. That means a business can pay ten times its normal bid to reach a known customer, or pay a tenth to keep showing up without spending much. OpenAI says uploaded files are deleted after processing, usually within 24 hours, and that Ads Manager never shows which individual people matched.

The feature got attention this week after Gizmodo walked through how ChatGPT targets ads, citing an investigation by 404 Media, which ran two ad campaigns of its own to see how the system works from the advertiser's side.

Here is the catch. A list must match at least 25,000 ChatGPT users before it can be used to include people or adjust bids, and OpenAI recommends 100,000 or more. Anything smaller can only be used to exclude people.

Note the word matched. If you upload 25,000 emails, you will not get 25,000 matches. Some of your customers do not use ChatGPT. Some used a work email with you and a personal one with OpenAI. Some are on plans that never see ads at all. A dental office with 6,000 patients, or a local firm with 2,000 past clients, is nowhere near the bar.

So for most small and midsize businesses, list targeting in ChatGPT is not a tool yet. List exclusion is.

Paying to advertise to people who already buy from you is one of the oldest leaks in paid media. Someone who booked with you last month asks ChatGPT a question in your category, your ad appears, they click it to find your phone number, and you pay for a customer you already had. Your report calls it a conversion. Your bank account calls it a donation.

An exclusion list removes that waste from day one, and it works at any size. Upload current customers, recent buyers or active patients, and your budget goes toward people who have never heard of you. That is the whole point of paying for reach.

Two practical notes. First, OpenAI says to avoid custom audiences for campaigns that target the European Economic Area or Switzerland, since ads personalization isn't available there yet. Second, only upload contacts you collected with a privacy policy that covers advertising use. If your intake form never mentioned it, talk to whoever handles your compliance before you upload anything.

Not everyone. OpenAI's ads basics page says ads do not appear on Plus, Pro or Business plans, and are not shown to users under 18. That means the people you reach skew toward free users, which is still a huge crowd.

Ads are matched to the conversation itself: what the person is asking about and what they seem to want. Advertisers can add what OpenAI calls context hints, which describe the kinds of conversations they want to appear in. OpenAI is clear that these are not exact match keywords. You are describing a situation, not bidding on a phrase.

That changes how you write an ad. In Google you might bid on emergency plumber. In ChatGPT the person may have typed three paragraphs about water under the sink and a landlord who will not call back. The ad that wins answers that situation, not a keyword.

OpenAI bills by views or clicks, and you can optimize for conversions, which are still billed per click. Its help page suggests a maximum bid of $3 to $5 per click as a starting point. Ads run in a relevance weighted, second price auction, which means relevance counts alongside your bid, and the winner pays just above the next highest bid rather than their own full bid.

UTM tracking is supported, so add UTM tags to every ad link and you can see ChatGPT traffic in Google Analytics next to everything else. Without them, those visitors may get lumped in with direct or referral traffic, and you will be guessing whether the channel works. Guessing is expensive at $4 a click.

If you are testing ChatGPT ads, or plan to, here is a sensible order:

• Export your current customer list and upload it as an exclusion audience before you spend a dollar.
• Write your context hints as real situations your buyers describe, in their words.
• Put UTM tags on every link so ChatGPT shows up as its own source in your reports.
• Start small, judge it on booked customers rather than clicks, and give it a few weeks of data before deciding.

If your list is already past 100,000 people, inclusion targeting and bid boosts are worth testing for win back campaigns and upsells. Everyone else gets the cheaper, less exciting benefit, which in advertising is usually the one that pays.

If you want help deciding whether ChatGPT ads fit your business, and setting them up so they bring in new customers instead of old ones, our founders can walk you through it. Talk to them below. See you in the next edition.