ChatGPT Ads Reporting: Metrics, Columns, Exports & Dynamic URL Parameters (2026)

Reporting is where most ChatGPT Ads programs quietly go wrong. Not because the numbers are unavailable, but because the default view answers a narrower question than advertisers assume it does — one combined conversions figure, a 24-to-48-hour lag, and a Segment menu that does not do the thing its name suggests. The result is a lot of correct campaigns being cancelled on incorrect readings.

The metrics Ads Manager reports

Available at campaign, ad group and ad level, in table view, chart view and CSV export.

Two things about that list have changed and both matter. Conversions is no longer a click-only number — when the view-through reporting window is set to 1 Day, eligible view-through conversions are included in the total, and the click-through and view-through components are also available as their own columns. And there are now revenue columns, Order Created Sales and Order Created ROAS, calculated exclusively from attributed order_created events; no other standard or custom event contributes to them, so a lead-gen account will see zeroes there no matter how well it is performing. OpenAI states additional metrics and reporting views will arrive over time; we track shipped changes in the news timeline.

Attribution windows, and what the Conversions total actually contains

This is the single most important thing to get right before anyone in your business reads a ChatGPT Ads number, because the same campaign can report materially different conversion totals depending on two dropdowns.

Ads Manager lets you choose a click-through reporting window of 7, 14 or 30 days, and a view-through reporting window of either 0 Day (disabled) or 1 Day. Any supported combination is allowed. Set the view-through window to 1 Day and eligible view-through conversions are included in the Conversions total. Set it to 0 Day and the total contains click-through conversions only.

What a view-through conversion counts

A view-through conversion is counted when someone converts within one day of an eligible ad impression and no qualifying ad click receives credit for the same conversion. If both an impression and a click qualify, the click gets credit — the two are never both counted for one conversion, which is why the components can be safely summed and why the total is not double-counting.

You will find the split in three places: in the attribution breakdown when you hover the Conversions column, as the optional standalone Click-through conversions and View-through conversions columns added through Edit columns → Conversion Events, and in CSV exports.

To change the windows, open Campaigns, Ad groups or Ads, choose Edit columns, and under Conversion Events find Attribution windows. Pick your click-through and view-through windows and save. The preference then applies to every conversion column displayed — which is the detail that catches people out, because it silently restates every number on the screen at once.

How to report this without misleading anyone

• State the windows next to the number, every time. "412 conversions" is not a fact; "412 conversions on a 7-day click and 1-day view window" is. Two teams pulling the same campaign on different settings will disagree, and both will be right.
• Pick one setting and freeze it for your standard reporting, then change it only deliberately and re-baseline when you do. A window change mid-quarter creates a step in the trend line that looks like performance.
• Keep the components visible. Add the Click-through conversions and View-through conversions columns alongside the total so the mix is legible. A total that grew because view-through grew is a different story from one that grew on clicks.
• Never compare across channels without aligning windows. A ChatGPT 30-day click total against a Meta 7-day click total is not a comparison, it is an artefact.
• Do not sum the total and its components. Conversions already contains click-through plus eligible view-through. Adding them produces a number that means nothing.

Used properly, the view-through component answers one narrow question well — is this creative influencing people who never click? On a conversational surface that is a more common outcome than on a search results page, which is precisely why it is worth watching. Note also that OpenAI has announced a conversion optimisation model that optimises on view-through as well as click conversions, billed on impressions; when that reaches your account, the view-through number stops being purely supplemental and starts feeding delivery. We cover how we structure all of this for clients on the measurement service page.

Breaking the combined Conversions total into individual events

This is the highest-value thirty seconds in the whole interface, and it is behind a menu almost nobody opens. By default, Ads Manager displays one combined Conversions total for the selected scope and date range. If you are firing purchase, lead and registration_completed, all three are summed into a single number that means very little.

1. Open the three-dot menu.
2. Select Edit columns.
3. Expand Conversions & events.
4. Choose the events you want as columns.
5. Select Save changes.

Once selected, those event columns also flow into CSV exports. The headers arrive in a machine-readable form — attributed_event_account_created, attributed_event_registration_completed and so on — which makes them straightforward to join to your own warehouse tables without hand-mapping display names.

The 24 to 48 hour conversion delay

Conversions are not reflected in Ads Manager immediately. OpenAI's documented guidance is to allow 24 to 48 hours for attributed conversions to be included in reporting.

Clicks and spend, meanwhile, appear promptly. The practical consequence is a reporting shape that looks alarming and is entirely normal: day one shows spend and clicks with a conversions column reading zero. We have watched more than one advertiser kill a campaign on that view before the first attributed conversion had any opportunity to land.

Dynamic URL parameters and macros

This is the feature that turns ChatGPT Ads from a black box into something joinable with the rest of your stack, and it is new enough that most accounts we audit are not using it. Ads Manager supports custom landing page query parameters built from your own text or from supported macros, populated automatically at delivery time.

Where to configure them

From the three-dot menu on the Campaigns, Ad Groups or Ads page, choose Edit campaign, Edit ad group or Edit ad. In the edit modal, use Landing page query parameters.

The precedence rule that catches people out

More specific settings take precedence, in this order:

1. Ad URL — highest precedence
2. Ad
3. Ad Group
4. Campaign — lowest precedence

If a parameter is already set at the Ad URL level, campaign, ad group and ad-level settings will not overwrite it. When a macro appears to be silently ignored, this is almost always why — someone hardcoded the parameter into the landing page URL months ago and the campaign-level setting has been losing the fight ever since.

Exports: cumulative vs daily

The first two options use identical cumulative metrics — the only difference is which rows are included. If you are building anything that needs a time series, Export daily values is the one you want, and it is the one to schedule if you are loading ChatGPT Ads data into a warehouse alongside Meta and Google.

Insights charts

1. Select the chart icon next to the three-dot menu.
2. Click View insights.
3. Select the metrics you want — impressions, clicks, spend, or any combination.
4. Set the date range.

Note what the chart view offers: impressions, clicks and spend. Conversions are not among the charted metrics, which is another reason the CSV export rather than the interface is the honest home for outcome analysis on this channel today.

Product-feed reporting: what the Products tab is not

For product feed campaigns, the Products tab is an insights-backed reporting view, not a complete feed inventory. Products appear there only after an active campaign delivers and generates reporting data.

This distinction matters for retailers debugging feed problems. An empty or short Products tab does not mean your feed failed to ingest — it may simply mean those SKUs have not served yet. To confirm feed readiness and eligibility, use the product count shown during ad-group setup, which reflects the feed rather than delivery.

Carousel card metrics: the reporting trap in feed campaigns

When a product feed campaign serves in the multi-product carousel ad unit, Ads Manager reporting now includes carousel card impressions and clicks, and the public Insights API exposes product-level fields alongside them. This is genuinely useful — it is the first time you can see which individual SKUs inside a feed ad are being seen and clicked, rather than reading the ad as one undifferentiated unit.

It is also the easiest way to break a reporting deck this quarter, so read the definition slowly.

One more constraint worth planning around: complete carousel data is available from 20 August 2026, on a rolling 30-day basis. That means two things. Anything before that date is not a fair comparison, so do not build a year-over-year or quarter-over-quarter carousel narrative on it. And because the window rolls, card-level history disappears behind you — if you want a longer series, export it on a schedule into your own warehouse rather than assuming Ads Manager will still hold it next quarter.

Troubleshooting: conversion events reading zero

A conversion event can be firing correctly, be accepted by OpenAI, and still report zero in Ads Manager. There is one dominant cause and it is a configuration mismatch, not a tracking failure.

Troubleshooting: clicks that do not match analytics

Ads clicks measure ad interactions. Analytics sessions measure something materially different, and the two will never reconcile exactly. OpenAI names the usual suspects: page load behaviour, redirects, consent settings, browser blocking, UTM handling, attribution windows and time zone settings.

Before you spend a day reconciling, check event quality. A gap between platform and analytics numbers is expected; a gap caused by a broken integration is not, and OpenAI now tells you which one you have. Open Tools → Conversions and read the warnings on the data source — a stripped click reference, thin identity data or duplicate callbacks will all widen the discrepancy you are about to investigate by hand. The event quality guide covers all nine warnings and what clears each.

1. Align the date range and time zone first. A large fraction of reported discrepancies are a time zone offset and nothing more.
2. Pull the CSV export and compare at campaign and ad level rather than account level — the gap is usually concentrated in one or two ads.
3. Check the redirect chain. The oppref click reference is appended to your landing page URL and needs to survive redirects; a link shortener or a marketing-automation bounce page that strips query parameters will break attribution while leaving clicks intact.
4. Check consent and blocking. Browser-side collection is structurally degraded, which is the argument for running the Conversions API alongside the pixel.

A stable gap is expected and manageable. A gap that widens over time is a signal, and the redirect chain is where we find the cause most often. The mechanics of oppref, first-party cookie storage and server-side event enrichment are covered in the conversion tracking guide.

Sending conversions from a CDP or attribution platform

You no longer have to write the Conversions API integration yourself. Three vendor paths are now documented.

Both are worth considering if you already run one of them. The caveat is the same as with any managed conversion pipeline: you still own the deduplication contract. If a CDP sends an event that your pixel also fires, they need the same event ID or you will double-count — and a double-counted conversion feeding an oCPC campaign does not just misreport, it actively mis-optimises.

What to include when you contact support

OpenAI publishes the list, and sending it up front is the difference between a same-week answer and a fortnight of round-trips. It also happens to be a decent self-diagnosis checklist — assembling it resolves the issue about a third of the time.

• Ad account ID
• Campaign and ad names
• Date range and time zone
• Ads Manager CSV export
• Analytics comparison
• Pixel ID
• Event names
• Sample conversion timestamps
• Screenshots of the discrepancy

A reporting cadence that works

1. Daily — delivery only. Impressions, clicks, spend. Confirm campaigns are serving and pacing. Do not touch bids on a daily conversion read.
2. Weekly — the real review. Export daily values, roll into weekly cohorts, compute cost per conversion yourself, and compare against your control campaign. This is the unit at which decisions get made.
3. Monthly — reconcile Ads Manager against your own analytics and against revenue. Expect a gap; document its size so a change in the gap is itself a signal.
4. On every change — one variable at a time, dated in a change log. Ads Manager records account change logs, and your own annotation of what you changed and why is what makes the weekly numbers interpretable three months later.

Sources and further reading

• OpenAI — Measure Results
• OpenAI — Conversion Measurement
• OpenAI — Conversion-optimized Campaigns
• Hightouch — OpenAI destination documentation
• GPT Ads AI — ChatGPT Ads conversion tracking
• GPT Ads AI — ChatGPT Ads conversion campaigns (oCPC)