Beta, Bugs and Breakthroughs: Learning ChatGPT Ads the Hard Way
“There is no better way to learn, when you are at the start of your career, than getting hands on. Today we are sharing a blog from Matt Hoskins who joins the Performance Marketing team from our Graduate program. We asked him to share his experience with the new Chat GPT ads manager platform.”
- David Thompson, Performance Marketing Executive Director
I’ve spent the past few weeks building campaigns in ChatGPT Ads Manager, and I wanted to share what the experience has actually been like. It has been a mix of “this is surprisingly easy” and “why won’t this save?”. I think that’s an honest picture of where the platform is right now, and it’s worth going in with your eyes open.
OpenAI is testing ads in ChatGPT for eligible users on its Free and Go plans, with advertisers able to get started through a self-serve Ads Manager in supported markets. What makes it interesting for businesses is the context. Someone asking ChatGPT a question is often partway through a decision, and the conversation can reveal far more about what they’re looking for than a single search term. The channel is also early, with little published benchmark data, so a small test is a sensible way to learn what works.
Getting started: Campaign setup and the dashboard
When you first open the platform, it feels approachable. The basic structure of a campaign makes sense quickly, and you can be up and running without much training. Compared with Meta or LinkedIn, though, the setup is more basic, with fewer controls to play with. That’s fine for getting started, but you can’t lean on every feature you’re used to.
The platform is also still in beta, and I think that explains a lot of the teething issues I ran into. I found a few bugs while creating campaigns, so I’ve learned to slow down, double-check each step and not assume something has saved just because the screen says so. New features are arriving all the time, so I expect the picture to look quite different in a few months.
Where the platform really impressed me was the dashboard. It’s very user-friendly, and it may well be because the product is so new and hasn’t yet accumulated the clutter that older platforms carry. You can see every data column attached to a campaign at a glance, and you can go deeper using the same metrics at ad group and ad level. For someone trying to understand what is working, that clarity matters a great deal.
What I learned from building and testing campaigns
In my experience, each ad group had to be saved before I could go back in and add the specific ads to it. That’s not a dealbreaker, but it adds up when you’re building several groups. It may not be an issue for everyone, and it may well have been updated by now.
What worked best for me was creating the ad groups and ads directly within the campaign structure, rather than jumping between separate screens. In practice, that meant creating an ad group, saving it as a draft, going back in to create the corresponding ad, saving that as a draft, and then repeating the process for the next group. It’s slightly tedious, but it got the job done.
Plan your ad structure
Planning the structure first makes a big difference. Have your ad copy ready before you start clicking, and the whole process becomes far less frustrating.
The platform can also generate AI-prompted ads based on the ad groups you’ve created. I didn’t use this feature myself, so I can’t speak to how good the suggestions are. But it’s new, and it could be genuinely useful for teams who need to produce a lot of creative quickly or who are starting from a blank page. It’s worth testing alongside your own copy before deciding how much to rely on it.
Get your campaign objectives right
One thing to get right from the start is your campaign performance objective. Once you’ve chosen an objective, such as conversions or clicks, you can’t change it within the campaign. I hit this when I realised my first choice wasn’t the right fit.
The workaround I found was the clone feature. Cloning the original campaign and changing the objective on the copy saved me from rebuilding every ad group and ad by hand. It’s a useful fix, but it shouldn’t be necessary. Take a little extra time to decide your objective before you build anything. I’m hoping Ads Manager adds the ability to change objectives in future.
I also ran into a reporting problem that I’m still trying to explain. For a period, the campaign was serving, but impressions, spend and reach weren’t populating in the dashboard. Cloning the campaign with the same conversion objective but a different billing objective, such as clicks or impressions, got the data flowing again.
I’m not sure whether this was a fault in Ads Manager or something on our side, so I’d treat it as an observation rather than a diagnosis. If you see something similar, it’s worth cloning the campaign with a smaller budget to test the theory before committing real spend. Running two versions side by side can also be a useful way to compare billing options anyway.
Experiment with context hints
Context hints were new to me, and they changed how I thought about targeting. Our rep recommended building each hint from three parts: the product or service, the use case, and a qualifier. It’s a simple formula, and it gave us a clear starting point.
From our previous experience, though, broad hints tended to work better than very specific ones. On one campaign, less was genuinely more. Our rep suggested a different approach, using more conversational hints, up to 30 per ad group, and that worked well too. My takeaway is to test both styles rather than assume one will win.
Plan for creative volume
Our platform rep recommended having up to 100 ads per campaign. That surprised me at first. It’s a lot more creative than most clients are used to producing, and it changes how much copy and imagery needs to be written, produced and approved.
Raise this early. Clients need to be comfortable with the volume of AI-generated ads before the campaign starts, not halfway through it.
Check the policy before anything goes to the client
Before anything goes to a client for approval, it’s worth reading through the ChatGPT advertising policy in full. Several areas are less clear-cut than they first appear. Some categories are permitted in one country and restricted in another, and others require OpenAI’s approval before an advertiser can run them. Where a client plans to target specific markets, both the advertiser and the ads they want to run need to be eligible there. In some cases, the product or service a client wants to promote falls outside the permitted categories altogether, and ChatGPT Ads may not be the right channel. Working through these questions at the start lets the strategy be shaped around the platform’s rules, rather than adjusting creatives after it has already been produced.
Where I’m heading next
ChatGPT Ads is still early in its availability, and I’m treating it as a channel to learn in public. The lessons above are a snapshot of a platform that is changing quickly. I’ll keep sharing what I find as it develops and as I get more hands-on with it.
Get in touch with IDX to find out more about how we are helping our clients.
Let's chat
Whether you're looking for service, support or a future strategic partner - we're here to help.
LDN