Should You Test ChatGPT Ads Yet? A Small-Budget Answer

What's changed

Digiday reports (5 October) that OpenAI began selling ChatGPT ads in February, and advertisers have complained about measurement, limited formats and not enough inventory. The latest updates respond to those complaints:

Image ads. OpenAI plans to test image ads in the US later in October. They'll appear during image generation, be clearly labelled, and stay separate from ChatGPT's generated images. Until now, ads have used a text-and-logo format with an optional image.

Better measurement. OpenAI is expanding its own tools and partner integrations, including Kochava and ten other partners, so advertisers can look at web and app outcomes beyond last-click attribution.

Incrementality testing. Geo-based tests are in early experimentation with Haus, Measured and WorkMagic.

Brand suitability. Pilots with DoubleVerify and Integral Ad Science are testing suitability in controlled environments, without giving the vendors access to real user conversations.

What hasn't changed

The harder problem is inventory. Digiday cites about 2.5 billion ChatGPT prompts a day, against about 15 billion Google searches. And as Shamsul Chowdhury of Zeno Group put it: "There's only one ad placement on ChatGPT whereas Google has multiple."

In plain terms, even if you wanted to spend more, there isn't much room to do so yet. That's a limit on how much you could learn too.

Measurement progress isn't measurement proof

Credit where it's due: moving beyond last-click is exactly what was needed. But incrementality testing is still at the experimental stage, so I'd treat any early results with caution. It's the same lesson as with TikTok's new ad network: a new channel announcing better measurement is not the same as a new channel proving its value.

How I'd approach a test

This is my opinion, not a rule:

Ring-fence a small budget. Make it an amount you're comfortable writing off, funded from your experimental line rather than your core spend.

Decide what success looks like first. Agree your measure (new customers, incremental sales, cost per acquisition) before the first ad runs.

Insist on incrementality. If you can run a holdout or geo test, do. If you can't, be honest that your results are indicative only.

Keep it short and review it. Set a review date, learn what you can, and make a decision rather than letting it drift on.

Check availability. The reporting describes a US test of image ads, so confirm what's available to UK advertisers.

The bottom line

ChatGPT ads are worth watching and, for the right brand, worth a small test. They're not yet a channel to plan around. If your audience is already using AI tools to research products, the experiment is cheap. If you're still working out your retail media and creator mix, put your energy there first.