How to Measure ChatGPT Ads, and What You Cannot Measure Yet
By Keshav Parsai 11 min read
In brief
Ads Manager reports impressions, clicks, spend, CTR, CPC, CPM, and conversions. It does not show the conversation your ad sat beside — or any benchmarks.
Last verified 13 September 2026. Version 1.1. Next scheduled review 13 October 2026.
A lot of published commentary on this channel describes it as unmeasurable. That was closer to true in February 2026 than it is now, and the documentation has moved underneath the commentary.
OpenAI's advertiser documentation, verified 13 September 2026, states that Ads Manager Beta reporting includes impressions, clicks, spend, click-through rate, average CPC, average CPM and conversions. Eligible accounts may also receive a separate one-day view-through conversion metric. Advertisers can set up conversion measurement in Ads Manager Beta, and landing-page parameters can use both static values and supported dynamic macros.
That is a materially better position than "aggregated data only," which is what most guides published earlier this year still say, and which an earlier version of this page said too.
What has not changed is the harder part: there is no benchmark to compare your numbers against, and you do not see the conversations your ads appeared beside.
What you can actually see
| Metric | Available | Caveat |
|---|---|---|
| Impressions, clicks, spend | Yes | In Ads Manager Beta reporting |
| CTR, average CPC, average CPM | Yes | Reported directly rather than derived by you |
| Conversions | Yes | Requires conversion measurement set up in Ads Manager Beta |
| One-day view-through conversions | Eligible accounts | Reported separately; do not add them to the main click-through Conversions metric |
| Order Created Sales and ROAS | Yes | Calculated only from attributed order_created events |
| Static and dynamic URL parameters | Yes | Supported macros include campaign, ad group, ad and ad-account IDs |
| Delivery status | Yes | Ads tab; hover a Not serving ad for the reason |
| The conversation your ad appeared beside | No | Advertisers receive aggregated performance information |
| Cross-advertiser or industry benchmarks | No | OpenAI states none exist |
Both buying models are supported. ChatGPT Ads offers CPM and CPC. A Views objective (formerly Reach) optimises for impressions and pays per thousand impressions; a Clicks objective optimises for clicks and pays per valid click; a Conversions objective bills either way, as oCPC or oCPM. So the channel is not view-priced by nature, and reporting that framed it that way was describing the pilot rather than the product.
The auction is a relevance-weighted, second-price auction. Your bid is not the only input, and relevance is weighted into selection, which means creative and landing page quality affect what you pay rather than only what you win.
The seven-hour rule, and the twenty-four-hour rule
Two numbers explain most of the panic in this channel's first weeks.
Reporting lags delivery by up to seven hours. An ad can be serving while the dashboard shows nothing. This is documented and it is not a bug.
Attributed conversions may take 24 to 48 hours to appear. This is a different clock from the delivery-reporting delay and explains why an event can test successfully before the Conversions column moves.
OpenAI's own guidance is to allow 24 hours before flagging a delivery issue. Before that window closes, a campaign showing zero impressions is not evidence of a problem.
The same latency explains a failure that gets misread constantly: a product feed uploads successfully and the Products tab stays empty. The Products tab is a reporting view, not an inventory view. Products appear only after an active feed campaign delivers, and then only after reporting catches up.
If your numbers disagree with each other, the cause is usually one of two things before it is a platform fault. Confirm the ads are marked Serving, and confirm you are comparing the same campaigns, filters and date range across the table view, the CSV export and the chart. Those three views are easy to leave on different settings.
What the agencies are doing instead
Given the above, the pilot participants are not running conversion tests. They are running learning budgets, and it is worth being precise about what that means in practice, because it is a legitimate strategy and not a euphemism for wasted spend.
Buying at a $60 CPM to establish creative patterns, category presence and internal capability in a surface that will be larger later is a defensible use of money. It is defensible because the buyer has said that is what they are doing, has sized it accordingly, and is not reporting it upward as performance.
The trap is a mid-market advertiser copying the tactic without copying the framing. A holding company allocating a learning budget across a client portfolio is doing something structurally different from a single brand spending its Q3 test budget and needing a ROAS number in October.
What is still genuinely missing
Three gaps survive the August documentation, and they are the ones that matter for a spending decision.
No benchmark. OpenAI states there are no performance benchmarks across advertisers, industries or campaign types. You can measure your own conversions precisely and still have nothing to compare them against.
No conversation-level context. Advertisers receive aggregated performance information and do not see the conversations their ads appeared beside. Search advertisers tune copy against query reports. There is no equivalent here, which makes creative iteration slower and more inferential.
No published view of the ad-eligible audience. Reach cannot be sized, because the exclusions are documented and their magnitudes are not.
The one integration that changes the picture
On 3 March 2026, Criteo became OpenAI's first adtech partner. Brands can access ChatGPT inventory through Criteo's commerce media platform, which means buying inside a workflow they already run and applying performance benchmarks they already trust.
This matters for measurement specifically. It does not create conversion attribution inside ChatGPT. It routes the buy through a platform that already has the advertiser's commerce data, which is a different and more practical answer to the same problem.
Criteo's aggregated data reports that traffic referred from LLM platforms including ChatGPT converts at roughly 1.5 times the rate of other referral sources. Read that carefully: it is a statement about referred traffic quality, not about ad performance. It describes what happens after someone arrives from an LLM, whatever brought them. It is the most useful number available and it is not a ChatGPT ads benchmark.
The measurement plan we would actually recommend
Not a framework. Four things to decide before spending.
Decide what the test is for, and write it down. Learning or performance. If learning, define what you will know at the end that you do not know now, and accept that the answer will not be a ROAS figure. If performance, the reporting available today may not support the decision you want to make.
Set your own baseline before launch. There is no external benchmark to compare against, so the only comparison available is your own before-and-after. That means measuring organic AI visibility, branded search and direct traffic for a period before the first impression serves. A test launched without a baseline cannot be read afterwards.
Set up conversion measurement before launch, not after. It exists in Ads Manager Beta and it is the single highest-value setup step. Add UTM parameters to landing page URLs as well; supported macros can populate {campaign_id}, {ad_group_id}, {ad_id} and {ad_account_id} at delivery. Preserve oppref through redirects. Run the crawler check at the same time, because a landing page blocking OpenAI user agents is a listed cause of ad rejection and will end the test before it produces data.
Budget for enough conversion events to distinguish signal from noise. The campaign minimum is 25 USD a day, and a two-week test at the floor produces single-digit conversions for most businesses. That is not a cheap test, it is an unreadable one.
What we cannot tell you
How much modeled measurement contributes. OpenAI now documents modeled conversion reporting where available, but does not report what share of an account's total is modeled or expose a control-group comparison.
Whether the paid surface lifts or cannibalises organic citation. The only measurement pointing either way is Seer Interactive's report that organic clickthrough rose 156% for a client brand while paid ran alongside and fell back when the paid placement was removed. One account, one brand, and it points at interaction rather than the independence that OpenAI's Answer Independence principle asserts. Nobody has run the test properly.
What a good CPM looks like. OpenAI's default max CPM bid is 60 USD, the pilot is reportedly clearing around $60 for premium positioning, and observed clearing is reported at 25 to 45 USD. Those are not enough data points to describe a market.
Whether we have run this. We have not. Everything above is documentation, published reporting and our own reading of it.
Quick answers
Does ChatGPT ads have conversion tracking? Yes. Advertisers can set up conversion measurement with the browser Pixel, the Conversions API, or both. Reporting includes conversions alongside impressions, clicks, spend, CTR, average CPC and average CPM, with separate one-day view-through reporting for eligible accounts.
Are ChatGPT ads priced on clicks or impressions? Both. ChatGPT Ads supports CPM and CPC buying. A Views objective (formerly Reach) optimises for impressions and pays per thousand impressions; a Clicks objective optimises for clicks and pays per valid click. OpenAI recommends starting CPC campaigns with a maximum bid of 3 to 5 USD per click.
How does the ChatGPT ads auction work? OpenAI uses a relevance-weighted, second-price auction to select among eligible ads. Advertisers set a maximum bid at the ad group level, and Ads Manager may provide bid-strength guidance indicating whether a bid is likely to be competitive or may limit delivery.
Why is my ChatGPT ads reporting showing nothing? Reporting lags delivery by up to seven hours, and OpenAI advises allowing 24 hours before flagging a delivery issue. If numbers disagree across views, confirm the ads are marked Serving and that campaigns, filters and date range match across the table, CSV and chart.
Why is my Products tab empty after uploading a feed? The Products tab is a reporting view. Products appear only after an active feed campaign delivers, and then only once reporting catches up, which can take up to seven hours.
What benchmarks should I compare ChatGPT ads against? OpenAI states no performance benchmarks exist across advertisers, industries or campaign types. The only comparison available is your own pre-launch baseline, which has to be measured before the first impression serves.
Can I buy ChatGPT ads through a platform I already use? Criteo became OpenAI's first adtech partner on 3 March 2026. Brands already buying through Criteo's commerce media platform can access ChatGPT inventory inside that workflow.
Sources
| Claim | Source | Tier |
|---|---|---|
| Reporting lags delivery by up to 7 hours; allow 24 hours before flagging | OpenAI Help Center, Troubleshooting, updated 31 July 2026 | Confirmed, primary |
| Products tab is a reporting view | Same | Confirmed, primary |
| Reporting mismatch checks: Serving status, matching filters across views | Same | Confirmed, primary |
| Reporting includes impressions, clicks, spend, CTR, average CPC, average CPM and conversions | OpenAI Help Center, Ads in ChatGPT: The Basics, updated ~8 Aug 2026 | Confirmed, primary |
| Conversion measurement can be set up in Ads Manager Beta | Same | Confirmed, primary |
| One-day view-through conversions for eligible accounts; 24-to-48-hour attributed-conversion delay; event-specific columns; Order Created Sales and ROAS | OpenAI Help Center, Measure Results, verified 13 September 2026 | Confirmed, primary |
| Dynamic URL parameters with campaign, ad group, ad and ad-account ID macros | Same | Confirmed, primary |
Pixel, Conversions API, oppref, advanced matching and modeled measurement |
OpenAI Help Center, Conversion Measurement, verified 13 September 2026 | Confirmed, primary |
| CPM and CPC buying via Reach and Clicks objectives; recommended starting CPC bid 3 to 5 USD | Same | Confirmed, primary |
| Relevance-weighted, second-price auction; bid-strength guidance in Ads Manager | Same | Confirmed, primary |
| Advertisers receive aggregated performance data, not chat access | OpenAI communication to users, reported 14 Aug 2026 | Reported |
| No benchmarks across advertisers, industries or campaign types | OpenAI advertiser FAQ | Confirmed, primary |
| Default max CPM bid 60 USD; recommended starting max CPC 3 to 5 USD | OpenAI advertiser documentation | Confirmed, primary |
| CPMs priced on views rather than CTR; advertisers can see clicks | eMarketer, 10 Feb 2026 | Reported |
| Omnicom Media, Dentsu, WPP Media in pilot at roughly $60 CPM; learning over ROI | eMarketer, 10 Feb 2026 | Reported |
| Criteo first adtech partner, 3 Mar 2026; ChatGPT inventory via Criteo commerce media | eMarketer, 3 Mar 2026 | Reported |
| LLM-referred traffic converts ~1.5x other referral sources | Criteo aggregated data, via eMarketer | Reported, vendor |
| Organic CTR rose 156% alongside paid, fell back on removal | Seer Interactive | Reported, single account |
| Landing pages blocking OpenAI user agents are a listed rejection cause | OpenAI Help Center, Troubleshooting | Confirmed, primary |
Related reading
- Do ChatGPT ads actually work? — the three public numbers and what each measured
- What a readable ChatGPT ads test actually costs — why a floor-budget test is unreadable
- ChatGPT ads troubleshooting — every documented failure mode
Changelog
13 September 2026. v1.1 added dynamic URL macros, event-level conversion columns, the 24-to-48-hour delay, one-day view-through reporting, Order Created ROAS, advanced matching and modeled measurement.
15 August 2026. v1.0 published.
Field kit
Tools
Site
OpenAI (primary)
Share this
Keshav studies how AI systems retrieve, verify, and cite brand information. At InPromptAds, he leads source research and turns platform documentation into practical guidance for advertisers.
