ChatGPT Ads Reporting: 7 Metrics, Zero Search Terms, and What I Did About It

I put my own money through ChatGPT Ads over a couple of weeks in September 2026, purely to poke at it. Impressions came in. Clicks came in. Spend came in.

Then I went looking for the bit that tells you what your ads actually matched to, and there isn’t one. Not buried in a menu. Not behind a toggle. It does not exist.

I’m Jamie. I run Umped, I’ve managed $28M+ in ad spend over nine years across Google and Meta, and I run ChatGPT Ads for clients now too.

So this is a rant with a point.

Short version. ChatGPT Ads reporting gives you seven aggregate numbers and nothing underneath them. There is no search terms report, no query data, no placement breakdown, at any level, ever. And here’s the part most people get wrong: that is not a beta gap that gets patched next quarter. OpenAI has framed it as a privacy decision baked into the architecture. So stop waiting for the report. Rebuild the control somewhere else, which I’ll show you how to do.

What does ChatGPT Ads reporting actually show you?

Seven metrics, at three levels, and that is the whole surface. Per OpenAI’s own documentation as of mid-2026, Ads Manager reports impressions, clicks, spend, click-through rate, average CPC, average CPM, and one rolled-up conversions number, available at campaign, ad group and ad level, with CSV export and an Insights API on top.

That sounds fine until you line it up against what you’re used to.

What you want to know	Google Ads	ChatGPT Ads reporting
What triggered the ad	Search terms report	Nothing. No equivalent exists
Which targeting matched	Keyword and match type	Nothing. Context hints are not reported back
Where the ad showed	Placement and network reports	Nothing
Who saw it	Demographics, device, location	Nothing beyond country
Blocking bad matches	Negative keywords	No negation at any level
Spend, clicks, CTR, CPC, CPM	Yes	Yes
Conversions	Per conversion action	One rolled-up number

So ChatGPT Ads reporting is solid on pacing and absent on diagnostics. You can tell whether you’re spending. You cannot tell whether you’re spending on the right conversations.

Which raises the obvious question, and the answer is more interesting than “it’s a beta”.

Why is there no ChatGPT Ads search terms report?

Two reasons stacked on each other, and only one of them is about privacy.

The first is structural. There are no search terms, because there are no keywords. ChatGPT Ads targets on context hints: plain-language descriptions you write at ad group level about the conversations where your product is relevant. OpenAI matches those semantically against the live conversation and runs a relevance-weighted auction. There’s no exact match, no phrase match, no match types at all. So a “search terms report” in the Google sense has nothing to report on, because the unit it would report doesn’t exist.

Fine. But that argument only gets OpenAI so far, because the obvious replacement is a conversation-context report. Show me the anonymised themes my ads landed beside. Meta manages placement reporting without handing over anyone’s DMs.

Which brings you to the second reason, and it’s the real one. OpenAI’s stated position is that advertisers receive aggregated, non-identifying performance data only, and never conversations, chat history, memories or personal details. Chats are the most sensitive text most people have ever typed into a computer. OpenAI has decided the ad product doesn’t get to touch them, even in aggregate.

Agree with it or not, it’s a coherent position. It’s also the one that quietly kills the report you’re waiting for.

Is a search terms report coming in a later update?

Probably not, and planning as if it is will cost you money. Every other gap in this platform reads like a roadmap item. Country-only geo will tighten. More ad formats will land. This one is different, because OpenAI has framed aggregated-only reporting as a design principle rather than a missing feature.

There is a difference between “not built yet” and “decided against”, and advertisers keep filing this one under the wrong heading.

My honest read as of September 2026: expect better conversion tooling, better bidding, better geo. Do not build a plan that depends on ever seeing what conversation your ad appeared in. If OpenAI ships something in that direction, treat it as a bonus, not a milestone.

Which leaves the part that actually keeps me up: what you can’t see can absolutely be wrong.

What is the real risk of running ChatGPT Ads blind?

Paying for relevance you can’t verify. On Google, a bad match shows up in the search terms report within days and you negate it. Here, the same bad match is invisible, and it keeps costing you at somewhere around $4.65 to $7.75 AUD a click on the recommended starting bids.

It is a big issue, and you would never know. Your dashboard would show healthy impressions, a respectable CTR and steady spend. Every number would look fine. The traffic would be wrong and the only symptom would be a conversion rate you’d probably blame on the landing page.

The counter-argument, which I made myself, is that the model should be smart enough to work out that a bathroom renovator shouldn’t show up beside a roof question. Maybe it is. That’s exactly the problem: I have no way to check, and neither do you. “Trust the matching” is a position, not a measurement.

So you rebuild the control somewhere the platform can’t take it off you.

How do you get control back without a search terms report?

You stop asking ChatGPT Ads reporting for it and move it upstream into structure and downstream into your own data. Five things, in order of how much they matter.

1. One theme per ad group. No exceptions. The ad group is your only unit of thematic control, because you cannot negate and you cannot see what matched. Mix bathrooms, kitchens and roofing hints into one ad group and you have permanently blended three data sets with no way to separate them. Split them and the performance difference between ad groups becomes your search terms report, by proxy.
   
2. Write hints narrow, then widen deliberately. Start tighter than feels comfortable. A hint that describes one specific buying situation gives you a cleaner read than five broad ones. Widening later is a decision you can measure. Starting wide is a mess you cannot unpick.
   
3. Send every ad group to its own landing page. Two reasons. The landing page is scored as part of ad selection, so a tight page improves matching, not just conversion. And a page-level conversion rate per ad group is the closest thing to relevance feedback this platform will give you.
   
4. UTM everything by hand, and isolate the channel in GA4. There’s no gclid equivalent, no auto-tagging and no macro substitution at click time. Your UTMs are the attribution. Expect GA4 to undercount, because referrers get stripped in app WebViews and copy-pasted URLs, so treat platform clicks and GA4 sessions as two different numbers that will never agree.
   
5. Judge on lead quality, not platform metrics. Feed real outcomes back from your CRM. If bathroom leads are arriving and roofing enquiries aren’t, you’ve answered the relevance question without ever seeing a query. That’s the whole workaround: you can’t audit the input, so you audit the output harder than you would anywhere else.

On the conversion side, OpenAI does document a proper measurement path, including a pixel, a Conversions API and modelled conversions where attribution is incomplete. Read OpenAI’s own conversion measurement documentation before you launch, not after. Zero conversions on this channel is almost always broken tracking rather than broken ads.

Should you still run ChatGPT Ads in Australia?

Some businesses, yes. Most local ones, no, and thin ChatGPT Ads reporting is not even the main reason.

The reason is geo. Targeting is country-level only. No state, no city, no suburb, no radius. If you’re a bathroom renovator in Western Sydney at roughly $6 a click, you’re paying Perth and Darwin prices to reach Penrith. That’s a slow leak, not a channel, and it outranks every other objection on this page.

Where it does work right now: national eCommerce, national B2B, travel, digital products. Businesses where reaching all of Australia is the point rather than the tax.

If that’s you, run it as a defined test with a written question, a budget you can afford to lose, and a decision you’ve agreed to make if the answer comes back null. Not a core channel line item. Not yet. I’ve written up what ChatGPT Ads actually cost in Australia if you want the numbers before you commit, and how it stacks up against Google Ads if you’re deciding between the two. If you’re earlier than that and just want the plain explanation of the channel, start with ChatGPT Ads in Australia.

Here’s the bit I’ll be straight about, because nobody else on this topic will be. I am not going to show you a ChatGPT Ads case study with a ROAS number on it. OpenAI has published no CPC, CPM, CTR, CPA or ROAS benchmarks by vertical, the channel is months old, and anyone waving a chart at you is either extrapolating from a handful of accounts or making it up. When I have results worth showing, dated and grounded, you’ll get them. Not before.

What I will do is run it properly while it’s still cheap to be early, with the structure above, and tell you honestly when it isn’t working. That’s the job. I run it as a ChatGPT Ads specialist alongside the Google and Meta work, hands on the account, not handed to a junior.

The reporting is thin. The channel might still be worth being early on. Both things are true, and pretending otherwise in either direction is how people lose money on new platforms.

Say G’day and I’ll tell you straight whether your business should touch it yet.

ChatGPT Ads reporting: frequently asked questions