ChatGPT Ads Manager lets you measure impressions, clicks, spend, CTR, CPC, CPM, conversions, and ROAS from attributed orders, broken out by device and country. It does not let you measure demographics, reach, frequency, or incrementality, it has no log-level export, and no third party verifies any of the numbers. That is the whole picture as of the date on this page, and the gap between those two lists is the thing to settle before anyone commits a budget.

Budgets are moving into ChatGPT ads faster than the measurement surface underneath them is being built. So we read OpenAI's own help documentation against one question: what can a marketing team actually measure here today? Every line below was checked against those docs on August 25, 2026. This is a fast-moving beta, so treat the date as part of the claim.

This expands on the LinkedIn post I published with the one-page version of this checklist.

Key Takeaways

  • The reporting is enough to spend money and not yet enough to tell you what the spend was worth. Core delivery metrics and ROAS exist; the diagnostic layer underneath them does not.
  • You get impressions, clicks, spend, CTR, CPC, CPM and conversions at campaign, ad group and ad level, revenue and ROAS from attributed order events, a browser pixel plus a server-side Conversions API deduped by shared event ID, an oppref click reference appended to your landing URLs, view-through conversions on a fixed one-day window, and device and country breakdowns.
  • You do not get demographics of any kind, reach, frequency, log-level or scheduled exports, any incrementality product, or third-party verification from IAS, DoubleVerify or the MRC. The platform grades itself.
  • The prompts that triggered your ads stay invisible to you, by design. This is a privacy decision, not a missing feature, so do not plan around it arriving.
  • Close one footgun before the first dollar: an untagged ad click arrives in GA4 as a chatgpt.com referral, indistinguishable from an organic ChatGPT mention.
  • Ads serve on the Free and Go plans only. Plus, Pro and Business subscribers do not see them, which is a targeting fact with budget consequences, not a measurement one.

What ChatGPT Ads Manager Lets You Measure Today

Eight things are actually available. This is the left-hand column, and it is a real column: it is enough to run a campaign and report on it.

  • Core campaign metrics. Impressions, clicks, spend, CTR, CPC, CPM and conversions, at campaign, ad group and ad level.
  • Revenue and ROAS. Order sales and return on ad spend, computed from attributed order events.
  • A pixel and a Conversions API. Browser-side and server-side conversion measurement, deduplicated against each other through a shared event ID. The pattern will be familiar from every other major platform, and the reasons it exists are the ones covered in our explainer on what a tracking pixel is and how server-side measurement changes it.
  • A click identifier. An oppref reference is automatically appended to your landing page URL, and you can capture and pass it server-side.
  • View-through conversions. Reported separately from click conversions, on a fixed one-day window. You cannot lengthen or shorten it. If your buying cycle is longer than a day, the view-through column is structurally understating impact, in the way any fixed attribution window does.
  • Device and country. These are the only two breakdowns in the segment menu.
  • App installs. Through AppsFlyer and Adjust integrations.
  • Your own UTMs. Static parameters persist on the click, and dynamic macros insert campaign and ad IDs, so you can build a naming scheme that survives into your own analytics.

What You Cannot Measure in ChatGPT Ads

Seven absences, and this is the column that decides whether a budget is defensible.

  • Demographics. No age, no gender, no audience reporting of any kind.
  • Reach and frequency. Not in the metric list at all. You cannot tell how many people saw an ad or how often, only how many impressions were served.
  • Log-level data. CSV export is manual, from the table you are looking at. There is no impression-level export and no scheduled delivery, so any recurring reporting has to be pulled rather than pushed.
  • Incrementality. There is no lift, holdout or geo-test product. If you want to know whether the spend caused the orders, you have to build the holdout yourself, using the methods in our guide to measuring marketing lift.
  • Third-party verification. No IAS, no DoubleVerify, no MRC accreditation. Every number in the interface is OpenAI grading its own homework. That is normal for a young platform and it is still worth saying out loud in the meeting.
  • The conversation. The prompts that triggered your ad are invisible to advertisers, by design. There is no query report and there is not going to be one.
  • Paid-tier users. Ads appear on the Free and Go plans only; Plus, Pro and Business are ad-free. Whatever share of your market is on a higher tier is unreachable here.

Notice the shape of the gap. What is missing is not a set of nice-to-have breakdowns, it is the entire layer you would use to answer "did this work, and would it have happened anyway." The left column measures delivery. Only the right column tells you value.

Tag Everything: the GA4 Mistake That Hides Paid Clicks

This is the one operational trap worth closing before the first dollar goes out.

Only the oppref reference is appended to your landing URL automatically. It is a click identifier, not a source label, and GA4 does not read it as one. So an ad click that carries no UTM parameters lands in GA4 as a plain chatgpt.com referral, sitting in exactly the same bucket as the organic traffic you get when ChatGPT mentions you in an answer for free.

The consequence is worse than a mislabeled row. Two things you are actively trying to tell apart, paid placement and organic AI visibility, become one undifferentiated line, and the paid line silently inflates whatever you thought your organic AI referral trend was. If you are tracking AI referral traffic as its own channel, and most teams now are, an untagged ChatGPT ad campaign corrupts that series from the day it launches.

The fix is unglamorous: tag every ad URL with a full UTM set before launch, and use a scheme you already enforce elsewhere rather than inventing one for this channel. Our UTM naming conventions guide covers the structure; the only ChatGPT-specific note is that static parameters do persist through the click, so the tags you set are the tags you get.

Talk to an expert if you are about to launch on ChatGPT and want the tagging and reporting settled before the spend starts, not after.

ChatGPT Ads Reporting in One Dashboard

Because there is no scheduled export and no log-level feed, ChatGPT ad data has a habit of living in a browser tab and a manual CSV, separate from every other channel. That is fine for one test campaign and it stops being fine the moment someone asks how ChatGPT compares to search and social on the same cost basis.

Improvado has an OpenAI Ads connector, so the campaign, ad group and ad level metrics land in the same warehouse as the rest of your paid media and can be normalized against it. The setup guide covers the connection and the available report types. The point is not the pipe itself, it is that a channel with no lift product and no third-party verification is the one channel you most need sitting next to comparable numbers, because comparison is the only diagnostic you have left.

What that does not do, and no connector can do, is manufacture the missing columns. Nobody can hand you ChatGPT demographics or reach, because OpenAI does not emit them. Be suspicious of any tool that implies otherwise.

How to Decide Whether to Spend

Take both columns into the meeting rather than the enthusiasm.

The honest read is that ChatGPT ads are measurable enough to test and not yet measurable enough to scale on platform-reported numbers alone. Delivery metrics and a ROAS figure will tell you whether a campaign is running and roughly what it returned on last-click terms. They will not tell you whether the orders were incremental, who you reached, or how often you hit the same person, and no third party is checking the arithmetic.

That argues for a specific posture: treat the first flight as an experiment with your own holdout rather than as a channel launch, tag it so rigorously that your analytics can answer questions the platform cannot, and set the size of the budget by what you are willing to spend without a lift measurement. That posture is more useful than either the "AI ads change everything" pitch or the reflexive dismissal, and it is the one the measurement surface actually supports today.

One more thing worth saying, since we sell measurement infrastructure and it would be convenient for us to imply otherwise: better plumbing does not close the incrementality gap. A holdout you design and run is what closes it. The plumbing determines whether you can see the result cleanly when you do.

Talk to an expert about bringing ChatGPT ad spend into the same reporting view as the rest of your channels.

Frequently Asked Questions

What metrics does ChatGPT Ads Manager report?

Impressions, clicks, spend, CTR, CPC, CPM and conversions at campaign, ad group and ad level, plus order sales and ROAS from attributed order events, and view-through conversions reported separately on a fixed one-day window. Device and country are the only two segment breakdowns.

Can you measure incrementality or run a lift test on ChatGPT Ads?

Not inside the platform. OpenAI ships no lift, holdout or geo-test product, so the only route to an incrementality read is a holdout you design and run yourself, measured in your own analytics.

Does ChatGPT Ads have an API?

Conversion measurement has a server-side Conversions API, deduplicated against the browser pixel through a shared event ID, and there are app-install integrations with AppsFlyer and Adjust. For pulling reporting data out on a schedule, note that the built-in CSV export is manual only, so recurring reporting needs a connector rather than a native scheduled export.

What is the best way to track spend and conversions for ChatGPT Ads?

Three things in combination: the pixel plus Conversions API for conversion capture, a full UTM set on every ad URL so your own analytics can separate paid from organic ChatGPT traffic, and a connector into your warehouse so the spend sits on the same cost basis as your other channels. Relying on the platform interface alone leaves you with a last-click ROAS figure that nothing else can cross-check.

Can you audit ChatGPT Ads for policy issues and creative compliance?

Not with the third-party verification stack you would use elsewhere. There is no IAS, DoubleVerify or MRC accreditation for this inventory, so brand safety rests on OpenAI's own policy enforcement and on whatever creative review you run internally before the ads go live.

Does ChatGPT Ads report demographics or audience data?

No. There is no age, gender or audience reporting of any kind, and no reach or frequency in the metric list. Device and country are the only breakdowns available.

Can advertisers see what people asked ChatGPT before an ad appeared?

No. Prompts and conversations are invisible to advertisers by design, so there is no query report equivalent to search terms in Google Ads, and none is expected.

Do ChatGPT ads show to paying subscribers?

Not on the higher tiers. Ads serve on the Free and Go plans only, and Plus, Pro and Business are ad-free. If your target buyer is likely to be on one of those plans, that shapes how much of your audience this channel can reach at all.

Why does ChatGPT ad traffic show up as organic in GA4?

Because only the oppref click reference is appended automatically, and it is not a source label. An ad click with no UTM parameters is recorded as a chatgpt.com referral, which is the same way organic ChatGPT referrals arrive. Tagging every ad URL before launch is what keeps the two apart.