OpenAI’s Ads Manager reports seven metrics on your ChatGPT ad spend: impressions, clicks, spend, CTR, average CPC, average CPM, and one rolled-up conversions number, all in aggregate. Seven numbers, which Ads Manager can break down by country and device. As of September 2026 that is the native surface of ChatGPT ads analytics. No query report sits behind those numbers. The click carries OpenAI’s own click reference, oppref, not a gclid. And GA4 quietly misfiles part of the traffic.
The verdict, up front. ChatGPT ads can be measured. The platform will not do it for you. The channel is measurable enough to pilot responsibly. But only if you build the measurement before the first ad runs, and only if your CRM outranks GA4 when you read results. The rest of this piece is the inventory and the build. For the buying side (formats, targeting, budgets), see our guide to how to advertise on ChatGPT.
The seven numbers Ads Manager reports
The list above is complete. You get those seven metrics at campaign, ad group and ad level, split by country or device, and no query or user detail beneath them. Everything else in your reporting stack you bring yourself.
Data table
| Measure | Value |
|---|---|
| native metrics in Ads Manager: impressions, clicks, spend, CTR, avg CPC, avg CPM, conversions | 7 |
| fixed view-through conversion window, reported as VTA (1d) | 1 day |
Conversion measurement runs through a data source in Ads Manager, fed by two tools. The OpenAI Pixel is the browser tag. The Conversions API sends the same events server-side, and OpenAI describes it as the more reliable path. Underneath, a click reference called oppref, which OpenAI appends to the landing page URL, connects eligible ad clicks to later conversion events.
Two recent additions deserve a date stamp. View-through conversions run on a fixed 1-day window, reported in the interface as VTA (1d); the window is reported as non-configurable. Automatic Advanced Matching, or AAM, matches more conversions using hashed customer info from your site forms. Search Engine Land reported on August 7 that it had become the default for new pixels, and that OpenAI would switch it on for existing pixels on August 17, 2026 unless the advertiser opted out. Hashed form data flowing to an ad platform deserves a privacy review. The opt-out exists for a reason.
What it cannot give you, as of September 2026
Start on the input side. You cannot see which conversations your ads appeared in, nor which context hints matched. A search terms report has no equivalent here, so the weekly feedback loop you run in Google Ads has no data to run on.
User-level data is a hard no rather than a missing feature. OpenAI frames aggregated-only reporting as privacy by design, and advertisers never receive conversations or personal details. The click carries less than you are used to: OpenAI appends its own oppref click reference, and Ads Manager fills ID macros such as campaign_id into landing page parameters you add, but the UTM values are yours to set. Attribution in GA4 rests on those UTMs plus whatever referrer survives. OpenAI’s docs describe click-through attribution, plus view-through attribution where available, against a click window you configure.
Independent verification is the last gap. OpenAI’s documentation, checked in September 2026, describes no DoubleVerify or IAS integration, no viewability standard and no incrementality tooling. Every absence below carries a September 2026 date stamp. Re-check this section first when OpenAI ships something new.
| Capability | ChatGPT Ads (as of September 2026) | Google Ads (baseline) | Workaround |
|---|---|---|---|
| Click ID / auto-tagging | oppref, appended to the landing page URL | gclid | UTMs and ID macros in your landing page parameters, plus oppref |
| Query report | None | Search terms report | None; design the account so you can live without it |
| Conversion tracking | OpenAI Pixel plus Conversions API | Mature | Run both; OpenAI describes the server-side path as more reliable |
| View-through conversions | Fixed 1-day window (reported) | Configurable windows | Read directionally |
| Third-party verification | None documented | DoubleVerify, IAS | Treat platform numbers as unverified |
| Incrementality | None in the docs | Experiments tooling | Holdout or geo-split design |
| User-level data | None, by design | Restricted | CRM-source truth (framework below) |
The GA4 problem: where your ChatGPT ad clicks land
Attribution here hangs on two threads: manually placed UTMs, and the referrer. Neither survives reliably. UTM passthrough is inconsistent, with parameters sometimes stripped before the click resolves. The referrer fares no better. A Clickport study, with a snapshot saved on April 23, 2026, shows how noreferrer links and referrer policies strip it before GA4 ever sees the session. So paid clicks land as generic chatgpt.com referral traffic, or as Direct. Nobody has a defensible percentage. We will not print one.
The consequence deserves plain words. GA4 undercounts this channel. A GA4-only read will kill a working pilot or excuse a failing one, and you will never know which happened. The standard workaround is a custom channel group keyed on utm_source=chatgpt plus the chatgpt.com referrer. Build it on day one. It stays a partial fix, which is why the framework below treats your CRM as the deciding record.
Advertisers on the record
The complaints are on the record. Read them before your CFO does. The Information reported advertisers calling the early buying process low-tech and saying they still lacked data proving the ads worked. Two agency executives told eMarketer they had yet to prove measurable business outcomes. Digiday framed the mood as advertisers flying blind. Digiday’s senior platforms reporter Krystal Scanlon said some advertisers were not getting even a tenth of their $250,000 commitments used.
The other side of the ledger is real too. OpenAI shipped the OpenAI Pixel, the Conversions API, an oCPC beta, AAM and view-through conversions inside a single year. Vendors are moving in as well. Adthena launched a competitive-intelligence product for ChatGPT Ads because advertisers cannot see competitors’ ads or the prompts that trigger them. Positive direction, real gap. Bank the direction, plan around the gap.
The measurement framework that works today
None of this needs exotic tooling. It needs the discipline any young channel demands, applied harder because the native reporting is thinner. Five pieces, all in place before the first ad runs.
- One UTM convention, written down. Fix utm_source, utm_medium and utm_campaign for every final URL before launch, and mirror the scheme in a GA4 custom channel group. A convention agreed on after launch is archaeology.
- OpenAI Pixel and Conversions API together. Browser plus server-side, test events verified before a single impression. Decide the AAM position deliberately: run the privacy review, and opt out if your compliance team says so.
- Landing-path isolation. Give each campaign its own landing path. When the referrer and UTMs vanish, the destination still tells you who came from where.
- Incrementality by design. OpenAI’s documentation describes no incrementality tooling. So the experiment design is the tooling: a holdout or geo-split (run markets against held-out markets), with brand-search and direct-traffic lift read alongside.
- CRM-source truth. Add a self-reported attribution field (“How did you hear about us?”) and carry the oppref and UTM trail into the CRM. Judge the pilot on CRM-sourced pipeline. Platform conversions and GA4 sessions are inputs.
This is the discipline COSEOM® runs on every channel inside an international PPC program. This channel just needs it earlier. We set it up before a dollar of spend moves. If you want it built and defended for you, that is what our ChatGPT Ads management service covers.
FAQ: ChatGPT ads analytics
What metrics do ChatGPT Ads report?
Seven. Impressions, clicks, spend, CTR, average CPC, average CPM, and a rolled-up conversions number, in aggregate at campaign, ad group and ad level. As of September 2026 Ads Manager breaks them down by country and device, with no query-level or user-level breakdown beneath them, so pair Ads Manager with your own tracking stack.
Do ChatGPT Ads have a pixel?
Yes. The OpenAI Pixel is a browser tag. The Conversions API sends the same events server-side, and OpenAI describes that path as the more reliable one. Both feed a conversions data source in Ads Manager. Run both and verify test events before launch. A tag installed after spend has started explains nothing retroactively.
Why does my ChatGPT ad traffic show as Direct in GA4?
Because UTM passthrough is inconsistent and several HTTP-level mechanisms can strip the referrer before GA4 sees the session. Paid clicks then land as generic chatgpt.com referral traffic or as Direct. The workaround is a custom channel group keyed on utm_source=chatgpt plus the chatgpt.com referrer, backed by dedicated landing paths per campaign.
Is there a gclid equivalent for ChatGPT ads?
Not a gclid, but OpenAI appends its own click reference, oppref, to the landing page URL, and Ads Manager fills ID macros such as campaign_id into the landing page parameters you add. oppref connects eligible ad clicks to later conversion events for the OpenAI Pixel and Conversions API. Treat oppref as plumbing. Your UTM convention is the reporting layer.
Can you track ChatGPT ad conversions into a CRM?
Yes, and the CRM should be your deciding record. Carry UTM values into hidden form fields, send conversion events through the Conversions API, and add a self-reported attribution field such as “How did you hear about us?”. CRM-sourced pipeline is the number that survives a CFO’s scrutiny. Platform conversions and GA4 sessions are supporting evidence.
Is there third-party verification for ChatGPT ads?
Not in OpenAI’s documentation as of September 2026, which describes no DoubleVerify or IAS integration and no viewability standard. Until that changes, treat platform-reported numbers as unverified and lean on your own conversion and CRM data.
How do you measure incrementality on ChatGPT ads?
You design it yourself, because OpenAI’s documentation describes no incrementality tooling. Run a holdout or geo-split: serve ads in some markets, hold comparable markets back, and compare pipeline. Read brand-search and direct-traffic lift alongside the split. It is the oldest method in media measurement, and on this channel it is currently the only one.
COSEOM Team
COSEOM is a multilingual B2B growth agency. Since 2008 the COSEOM team has run international SEO, generative engine optimization, paid media and demand generation programs for B2B software and technology companies, in English, German, Spanish and Portuguese. This account publishes our research, playbooks and product guides. More about COSEOM