ChatGPT Ads: What We've Learned Running Them

ChatGPT Ads are now available in Latvia and across Europe. They might look familiar in Ads Manager, but in practice they require a very different way of thinking about targeting and creative. We've started running campaigns and are sharing what we've learned so far.

Currently, ChatGPT ads appear only to users who don't pay for a subscription. Plus and Pro users do not see them. This is down to how ad inventory is available right now, not a system error. When you calculate advertising costs, base your budget on the number of free-tier users, not the total ChatGPT user count.

This restriction segments the audience you reach. Consider who stays on the free plan:

• Price-sensitive users who won't pay for an AI subscription, or a second one.
• Low-usage users who never hit the limits that force an upgrade.
• Professionals with paid company access at work who only use the free tier on personal accounts.
• Geography and currency — twenty dollars a month means different things in different markets.

A free user is not necessarily a light user. Compare these factors against your target customer before assuming free-tier status signals a limited budget or low intent.

The single most important lever you have is the context hints you attach to the ad group. This is what tells the model when your ad is even eligible to show.

Most people write these the way they'd write a Google Ads theme: broad and safe. Vague isn't automatically wrong — for reach or brand awareness, broad hints can be exactly right. But if you're running performance and being judged on ROAS, broad is a tax you're choosing to pay.

Specific hints make sure you match conversations where your product is the actual solution rather than a related category.

Specificity in the hints only pays off if the ad carries the same specificity through. If your hints are scoped to marathon training for a sub-3:30 finish but the copy is generic shoe messaging, you paid for precision targeting and threw it away at the creative step. The ad needs to read like it was written for that exact conversation.

If your context hints match, your ad shows up inside a conversation where the user already got an answer from ChatGPT. That changes everything about what the ad copy needs to do.

A keyword-style search ad doesn't work here, because there's no information gap left to fill. A disruptive social ad doesn't work either, because you're not competing against a scrolling feed — you're sitting inside a conversation the user is actively engaged in.

Contextual continuation is what works. ChatGPT's first response is broad by design. Use it as the starting point: identify the next question the user will have and answer it immediately.

And write to the real space you have. The headline caps at 50 characters and the description at 100, but some placements truncate earlier. We work to roughly 24 characters for the headline and 48 for the description as a safer target.

Example. A user asks: "What running shoes should I buy for marathon training?" ChatGPT gives a solid, general answer: cushioning, stability, and a shoe that matches your gait.

You are not interrupting or pitching. You are providing the specific information the user was already looking for. The sale happens because you provided utility at the moment more detail was needed, not because you asked for the transaction.

Write many ads. Many different "next answers" for the same context hints. You don't know in advance which anticipated follow-up question actually matches what real users ask next, so treat this like any other performance channel: hypothesis, variant, test, kill the losers, scale the winners. Teams that only ship two or three variants will read this channel as underperforming when really they just didn't give it enough shots on goal.

None of the above matters if OpenAI's system can't see what happens after the click. Same rule as every other ad platform: no conversion signal, no optimization.

OpenAI ships this as a Google Tag Manager template, "OpenAI Ads Measurement Pixel", available in the GTM Community Template Gallery. It works in two layers:

• A base tag on all pages. It initializes the pixel and covers page views. Trigger it on All Pages and turn off the "send a measurement event" option so it only initializes.
• A separate tag per conversion event — purchase, lead form, signup. Each needs the event's specific ID from OpenAI Ads Manager (under Events) plus a trigger scoped to that exact action.

Install the template, set up the base tag, then add one event tag per conversion mapped to its own ID. Test everything in GTM Preview before you publish the container. It's an easy setup, and it's the difference between OpenAI's delivery algorithm learning what a good click looks like for you versus optimizing toward nothing.

ChatGPT ads reward the same discipline that's always separated good performance marketing from average: precision over breadth, and genuine usefulness over interruption. Teams that treat this like a new flavour of search or social ads are going to burn budget learning that the hard way.

ChatGPT Ads are a new game in town and there's a steep learning curve for everyone. We believe it pays to be at the front of that curve.