A view from the pilot — and why the next paid channel looks nothing like the last one.
For the past few months I’ve been part of OpenAI’s ChatGPT Ads pilot, along with a small group of other advertisers around the world. If you run paid media in-house and you’re trying to figure out whether this channel is worth paying attention to yet, this is the post I wish I’d read six months ago.
ChatGPT Ads is not a new ad network. It’s a new surface.
This is the first mistake I made, and I’ve seen other people make it since.
The instinct is to treat ChatGPT Ads like Google Ads on a different platform. Bid on keywords, write a good headline, get the click. It doesn’t work that way. The user isn’t looking at ten results and picking one. They’re asking a question and getting an answer. If your brand is going to show up at all, it needs to be part of the answer — not sitting underneath it waiting for a click.
That changes most of the things paid media usually assumes:
- There is no results page to rank on. There’s an answer to be included in.
- Impressions and clicks are not what matters. Being mentioned inside a useful answer is.
- Keywords barely apply anymore. The user’s question is the context.
- The funnel shrinks. The user is making up their mind during the conversation, not after.
If you bring a Google Ads mindset to this, you’ll build the wrong campaigns and read the data wrong. I know because I did.aw the wrong conclusions from the data. I speak from experience.
Measurement is where most teams will come unstuck
I’ve spent most of my career on measurement — MMM, incrementality, testing — and that’s turned out to be more useful here than any paid search knowledge. Here’s why.
The normal move is to plug the new channel into whatever attribution you already use and see what it says. Don’t. Three reasons:
The path from click to conversion looks nothing like paid search. People interact with the channel in ways that don’t leave the usual trail. Your attribution tool will underreport what it’s actually doing.
There are knock-on effects in other channels. Being seen inside ChatGPT seems to lift brand search and direct traffic in ways last-click will miss completely. If you only look at the platform dashboard, you’ll turn off spend that was actually working somewhere else.
The data is early and messy. Drawing big conclusions from a few weeks of data on a channel that didn’t exist a few weeks ago is a quick way to make an expensive mistake.
What’s worked for me: treating the pilot as an experiment from day one. Geo holdouts where possible. Watching brand search and direct before and after. Incrementality as the main signal, not ROAS on the platform. And updating the MMM to include the new channel properly, not just reading numbers off a dashboard.
If your measurement is still mostly last-click, you won’t be able to judge this channel fairly. Fix the measurement first, then worry about the spend.
The creative question is the one nobody has a clean answer to yet
Every conversation I have with other paid media leaders about AI channels ends up in the same place: what does the creative even look like?
Honest answer: nobody fully knows yet, including the platform. But a few things are already clear from being inside the pilot.
Clear product copy matters more than before. If a model is going to include your brand in an answer, your product description and positioning have to be obvious. Vague brand copy falls apart the moment a model tries to match it to a real question. A lot of brands are going to find their copy is vaguer than they realised.
Useful information beats decorative copy. The brands that show up well are the ones whose information is actually helpful when it’s pulled apart and put back together. Fluff disappears. Substance sticks.
Strong brands win. The better known and more distinctive your brand, the better you do when a model picks what to reference. This is a point Byron Sharp and Mark Ritson have been making for twenty years, and it’s about to matter a lot more. If you already have a strong brand, this channel rewards you for it. If you don’t, catching up is going to get expensive.
That last one is the point I’d most want in-house marketers to take seriously. AI channels will reward brands that have done real brand-building and will punish brands that only ever chased performance. If you’re a performance-only brand, you’ll feel this first.
A few operational things nobody writes about
Some things that surprised me and aren’t in the trade press:
Getting the pixel right matters more than you’d think. The tracking setup is still maturing. Mess it up early and you lose weeks of learning. Get it right and the platform learns faster.
Cross-team coordination takes over your week. Paid media, analytics, dev, legal — everyone has to move together. If your company can’t work at that speed, the technical setup drags and the pilot is over before you’ve learned anything useful. This is a common problem.
Reporting is going to be custom for a while. I’ve ended up building my own PDF and slide reports in Python because the ready-made tools don’t exist yet. Plan for the time or you’ll end up reporting on feelings.
The platform is changing fast. What’s true this week might not be true in six. Treat your notes as a living document, not a finished thing.
What I think this actually means
I don’t want to oversell this. Google and Meta aren’t going anywhere, and anyone saying otherwise is going to look silly in about eighteen months.
But there’s a new channel opening up alongside them, with its own rules, its own creative logic, and its own measurement needs. The in-house marketers who’ll benefit most are the ones already doing proper measurement, already building a real brand, and already set up to test new channels without it taking a quarter to get a pixel live.
If you run paid media in-house and you’re trying to figure out how to approach ChatGPT Ads, I’m happy to compare notes. Half of what I’ve learned so far has come from talking to other people working through the same thing, and that’s honestly been the best part.