How two months running ChatGPT Ads had me unlearning Google Ads

There are no experts in ChatGPT Ads

You may be the most capable, experienced Google Ads manager around, but not every skill you have refined over the years is going to port over to ChatGPT Ads.

Lets set the scene

I have been running Adthena's ChatGPT Ads Manager account since the end of May. Bringing AI Search Intelligence products to market is my day job, performance marketing is my background, so taking this new platform under my wing was an easy call.

None of my following words are criticisms of such a young platform still in beta. In fact, this article will age like a vampire on a sun bed with the pace of change seen so far in 2026. My points rather refer to a few places where a capable paid search practitioner may still get tripped up or surprised.

Four key points

Firstly, write for a conversation, not a query. In one A/B test on the account, a rich, conversational context hint returned a 1.07% CTR against a 0.77% for a thinner, keyword style list. You cannot exact match a conversation. This makes sense once you step back and think about how the platform reports back. OpenAI gives aggregate performance, not a prompt-level view. No search query report exists, and dare I say it? It never will. The closest we could hope for in my opinion is a report of prompt volumes linked or categorised into themes or topics.

Daily budgets (if you have enough headroom) behave more like a promise than a suggestion. Every campaign in my account on one test has spent exactly the daily budget, every day, and without fail. A minimum budget test of $25/day (for a theme I suspected there would be headroom for) delivered consistent days of exactly $25 spent, not $24.83, or $25.67, $25 exactly. This feels like a bug on the first couple of days, then rationally I develop a sense that when demand exceeds the cap, we're throttling across the day to our total lands exactly on the max spend cap. I expect we have all been burnt by Google at some point and wished this was the case there for a daily budget. I won't source anyone, but anecdotally I know this is not the case for a lifetime budget where spend can be utilised very quickly, with minimal pacing mechanism in place, so do be cautious!

The US and UK are significantly different markets right now, but I expect not for long. In another A/B test I ran the same campaign, context hint, ad combination but split by geography. Average CPC in the US was $4.52, against $3.05 in the UK. The US trial has been live around 4 months longer than the UK, so a gap this size after such a small head start says a lot about how quickly we're moving here. Lighter density in the UK right now is an opportunity, not a permanent feature. A tiny bit of insider information as of July when I write this, the number of US advertisers on ChatGPT is around 300% bigger than the UK, some of this attributed to the head start, some purely on population and advertiser numbers between the two.

My last point is around metrics. We have impressions, clicks, cost, CPC, conversions etc. It's a primary school set of available data at this stage. If you do want anything more exotic such as Conversion Rate, or CPA, you're going to have to download the data and do this yourself. (Although this guy named Claude does it for me now).

In summary

None of my four points above are permanent laws of ChatGPT Ads, they're more of a snapshot of where we are today. None of this unlearns or makes redundant the Google capability we hold either. Personally I'm rather excited to see the testing, optimisation and ingenuity which will come from wider rollout, adoption and usage within the SEM community.