OpenAI has switched on view-through conversion reporting inside ChatGPT Ads, crediting conversions that follow an ad impression within a window fixed at one day. The metric is walled off from the main Conversions total, and it does not touch bidding, billing or cost per acquisition.

The change surfaced through documentation rather than an announcement. OpenAI's developer pages for the ChatGPT Ads Measurement Pixel and Conversions API now describe view-through attribution alongside the click-through attribution that has been in place since May. The company's help centre article on measuring results in Ads Manager Beta was updated on Tuesday, August 18, 2026, and the companion article on conversion measurement was updated on Wednesday, August 19. Neither carried a blog post, a press release nor an email to advertisers.

A screenshot circulated on LinkedIn on August 19 showed the practical consequence inside the interface: a metric selector listing "View-through conversions (1 day)" and "Click-through conversions (30 days)" as checked options, sitting beside unchecked entries for Order Created ROAS, Post-click CVR and Cost per Conversion.

What the documentation specifies

The mechanics are narrow and, unusually for a young ad platform, spelled out in plain terms.

The same conversion events already flowing into OpenAI support both attribution types. Click-through attribution uses whatever click window an account has configured. View-through conversions run on a fixed one-day window measured from an eligible ad impression, and that window operates independently of the configured click window. Where a single conversion qualifies for both treatments, the documentation is explicit: "the click takes precedence."

Availability is not universal. Both the developer documentation and the help centre qualify the feature as applying to eligible accounts, without publishing the eligibility criteria, the account tiers involved, or the markets in which the reporting has been switched on. Whether view-through reporting appears at all does not depend on the click window an advertiser has set.

No implementation work is required. View-through reporting demands no changes to the Pixel integration, no new event payload fields, and no separate request through the server-side path. An advertiser already sending order_created or lead_created events receives the additional reporting automatically, assuming the account qualifies.

Where the metric appears in Ads Manager

Inside the interface, view-through conversions are reported at campaign, ad group and ad level under the label VTA (1d). Two routes lead to the figure. Hovering over the Conversions column produces an attribution breakdown that includes it. Alternatively, it can be added as a standalone column through the Edit columns control on the Campaigns, Ad groups or Ads pages. The data can also be included in CSV exports.

Individual conversion events remain accessible through a separate path: the three-dot menu, then Edit columns, then the Conversions and events group, then a selection of specific events and a save. Selected event columns carry through to CSV exports, though the help centre notes that availability varies by account and reporting period. The Segment menu governs device and country breakdowns only and does not add conversion-event columns.

Reporting latency has not changed. OpenAI advises allowing 24 to 48 hours for attributed conversions to appear.

A labelling discrepancy

The help centre and the interface do not use the same name for the same number. Documentation calls the optional column VTA (1d). The metric selector visible in the LinkedIn screenshot spells it out as "View-through conversions (1 day)". Advertisers building automated reporting against column headers will encounter both strings.

A second inconsistency concerns the click window. The screenshot shows a click-through metric labelled with a 30-day window. OpenAI's published documentation never names a default click window, referring only to the applicable configured window for each account. The 30-day figure in the interface therefore has no documented counterpart, and OpenAI has not published a default, a permitted range, or the mechanism by which the value is set.

What deliberately does not change

The most consequential detail in the documentation is a list of things that stay where they were.

View-through conversions are excluded from the Conversions metric, which remains a click-through total. Cost per acquisition stays click-through-based. Post-click conversion rate stays click-through-based. So do bidding, billing and conversion optimisation. The help centre describes the new figure as supplemental reporting and states that it does not feed any of those systems.

That construction matters. On most platforms, a view-through window is not merely a reporting line; it is an input to automated bidding and to the cost metrics against which budgets are defended. By fencing the one-day view number outside the optimisation loop entirely, OpenAI has released a metric that advertisers can look at but that the auction cannot act on. The company is, in effect, publishing a number while declining to stand behind it as a performance signal.

Conversion-optimised campaigns, which began rolling out on June 5, 2026 for accounts with at least one conversion event registered by June 1, continue to train on click-attributed events alone.

How the signal is collected

The underlying plumbing is unchanged from the pixel that launched on May 5, 2026 alongside the self-serve Ads Manager.

The Measurement Pixel is a browser SDK loaded from bzrcdn.openai.com and initialised with a Pixel ID created in the conversions tab of Ads Manager. OpenAI recommends placing the snippet near the top of the head section so that early conversions are not lost while other page content loads. Conversions fire through an oaiq("measure", ...) call.

The SDK handles several transport details without configuration. It captures oppref, described in the documentation as a privacy-preserving identifier, from the landing page URL, then stores it in a first-party cookie named __oppref so that later page views on the same domain can reuse it. It attaches the current page origin as source_url, timestamps each event, and batches closely grouped measure calls. Where automatic advanced matching is switched on, it detects supported customer information from website forms, normalises it, and hashes it with SHA-256 in the browser before transmission. Raw customer information is not sent through that mechanism.

Sites enforcing a Content Security Policy need four sources merged into the existing policy: bzrcdn.openai.com under script-src to load the SDK and under connect-src to fetch per-pixel configuration, plus bzr.openai.com under connect-src for fetch and sendBeacon transmission and under img-src for the image-request fallback. OpenAI's guidance rules out adding unsafe-inline for the pixel alone, and specifies that sites defining script-src-elem must add the CDN source and nonce or hash to that directive as well.

Manual matching data can be attached through an optional user object at initialisation, accepting email_sha256, external_id_sha256, country, city and zip_code. Hashes are required as lowercase 64-character hexadecimal strings. Raw email addresses, raw external identifiers, phone numbers and phone number hashes are all excluded.

Duplicate suppression across the browser and server paths relies on a shared event_id, matched against the Pixel ID and event name. For custom events, custom_event_name substitutes for the standard event name in that match. Custom event names are constrained to between 1 and 64 characters, must contain only letters, numbers, underscores or dashes, must begin and end with a letter or number, and cannot duplicate a standard event name.

The server-side path

The Conversions API documentation carries matching language. Web events sent server-side support click-through attribution and, where the account is eligible, view-through attribution on the same fixed one-day basis. View-through attribution requires no separate request and no additional event field.

Unlike the browser pixel, the server-side API does not capture oppref automatically. Advertisers running the API alone are directed to capture the value themselves and pass it with the server event to support click matching. Hybrid deployments are told to read the first-party cookie in the browser, forward it to the server, and include it unchanged with the server-side event.

App lifecycle events occupy an awkward position. app_installed and app_opened are sent server-side with action_source set to mobile_app, using the Pixel ID from an existing web data source. Native mobile SDK setup and dedicated mobile data sources are not supported. That constraint sits alongside the app install attribution integrations that AppsFlyer and Adjust activated across seven markets in late July 2026, which route mobile measurement through the same identity infrastructure.

Modelled conversions enter the picture

The conversion measurement help article adds a line that has not previously appeared in OpenAI's advertising documentation. Where modelled measurement is available, the company may use aggregated patterns from observed conversions to estimate attribution for advertiser-reported events that could not otherwise be attributed. Reported conversion totals may therefore include modelled conversions.

OpenAI does not specify where modelled measurement is available, what proportion of a reported total it might represent, or whether modelled and observed conversions are distinguishable in the interface. The company lists modelled conversion reporting among the reasons its totals diverge from third-party analytics, alongside differing attribution windows, timestamp and time-zone boundaries, browser and consent conditions, and deduplication behaviour.

Modelling is standard practice at Google and Meta, where it compensates for signal lost to browser restrictions on Safari and iOS in particular. Its arrival in ChatGPT Ads documentation indicates OpenAI is following the same route, though without publishing methodology or confidence intervals.

The pixel exposes a consent command. Calling oaiq("consent", false) before initialisation suppresses measurement-event pings; calling it with true afterwards permits future events. Blocked events are not replayed once consent is granted. The pixel initialises consent as true by default unless it is explicitly set to false or finds a stored denial.

That default is significant in any market where prior consent is a legal precondition for tracking. ChatGPT advertising remains outside the European Union. The pilot reached the United Kingdom on June 6, 2026 and Japan and South Korea on June 22, with EU availability unresolved against the General Data Protection Regulation and the ePrivacy Directive. A view-through window that fires on impressions rather than clicks raises the consent question in sharper form, because the conversion is credited to an event the user never acted on.

Why one day, and why separate

Set against the direction of the rest of the industry, a fixed one-day view-through window is conservative.

Meta eliminated its 7-day and 28-day view-through attribution windows from the Ads Insights API with effect from January 12, 2026, retaining click options and a one-day view. It then redefined click-through attribution in March 2026 to count only link clicks for website and in-store conversions. Google Analytics moved the other way in August 2026, replacing its fixed three-day engaged-view conversion window with a configurable value. One platform narrows the definitions on offer; the other widens the range.

OpenAI has landed at the narrow end and locked the door. The window cannot be configured. It is one day, and the documentation states plainly that it is independent of the click setting.

The reasoning is not hard to reconstruct. Post-view credit at premium impression prices has been the single most contested claim in the ChatGPT advertising pilot. When CPMs were drifting between $25 and $60 in April 2026, the argument put to advertisers by measurement practitioners was that post-click attribution is comparatively easy to defend while post-view attribution at $60 CPM is not, absent independent verification of what a ChatGPT impression is worth. OpenAI's response, four months later, is a view-through number that cannot inflate CPA, cannot influence bidding, and cannot be widened.

Six months of measurement construction

The reporting addition lands roughly six months after ChatGPT advertising began, on February 9, 2026, as a closed pilot with a minimum commitment reported at between $200,000 and $250,000 and no conversion measurement whatsoever.

The build-out since has been rapid. The Measurement Pixel and Conversions API arrived on May 5 with the self-serve Ads Manager and cost-per-click bidding. A community Google Tag Manager template followed on May 7, two days later. Custom audiences appeared on May 14, later constrained by a 25,000 matched-user floor. Conversion optimisation began on June 5. LiveRamp connected its Conversions API Hub on June 10, the first third-party server-side route into the platform. AppsFlyer and Adjust went live with mobile attribution in late July. Automatic advanced matching switched on by default for existing web pixels on Monday, August 17, 2026, two days before the view-through documentation update.

Demand data remains contested. Adthena counted 7,378 distinct advertisers inside ChatGPT during the week of July 13 to 20, 2026, with the United States accounting for 60.1 percent, while recording zero paid ChatGPT placements across 169,560 United Kingdom result scrapes during June. Criteo reported 2,000 brands running campaigns through its API connection by June 2026. Click-through rates measured by Similarweb sat at 0.68 percent overall.

Why this matters for the marketing community

For performance teams, the immediate effect is a new column and a familiar argument.

A one-day view-through figure reported at campaign, ad group and ad level gives media buyers something to put next to a conversational placement that, by construction, is often consumed without a click. Someone comparing three products inside a ChatGPT thread may absorb an advertiser's proposition and arrive at the site later through branded search or a direct visit. Under click-only attribution that campaign scores zero. The new metric will show a number.

Whether that number carries weight is a separate question. OpenAI has taken the unusual step of pre-emptively removing it from every calculation that matters commercially. CPA will not move. Bids will not shift. Invoices will not change. An agency reporting a ChatGPT campaign to a client now has two figures to explain rather than one, and the second comes with a documented caveat from the platform itself that it is supplemental.

For measurement leads, the practical work is reconciliation. The label mismatch between VTA (1d) and "View-through conversions (1 day)" will break naive column mapping in reporting pipelines. The undocumented 30-day click window visible in the interface needs verification account by account. And the arrival of modelled conversions inside the reported total, without a published methodology, introduces a second source of divergence between Ads Manager and any independent analytics stack.

The broader signal concerns where the platform sits in its own development. A channel that crossed $100 million in annualised revenue within six weeks of launch, and that added product feed campaigns, custom audiences, mobile attribution and third-party server-side routes inside a single quarter, has now assembled most of the reporting surface that established platforms took years to build. What it has not assembled is the surrounding apparatus - the independent verification, the incrementality testing, the media mix models - that determines whether any of those numbers can be trusted. A view-through metric that the platform declines to use for its own bidding is a candid acknowledgement of exactly that gap.

Timeline

Summary

Who: OpenAI, through its ChatGPT Ads developer documentation and help centre, affecting advertisers, agencies and measurement teams running campaigns inside ChatGPT.

What: View-through conversion reporting has been added to ChatGPT Ads. Conversions occurring within a fixed one-day window after an eligible ad impression are credited as view-through conversions, reported at campaign, ad group and ad level as a metric labelled VTA (1d). The window cannot be configured and operates independently of the click-through window. Where both a click and an impression qualify for the same conversion, the click receives credit. The metric is excluded from the main Conversions total and from cost per acquisition, post-click conversion rate, bidding, billing and conversion optimisation. No changes to pixel or Conversions API implementations are required. Availability is limited to eligible accounts, and OpenAI has not published the criteria.

When: The Measure Results help centre article was updated on Tuesday, August 18, 2026, and the Conversion Measurement article on Wednesday, August 19, 2026. A screenshot of the Ads Manager metric selector circulated on LinkedIn on August 19. The change follows the automatic advanced matching default activation of Monday, August 17, 2026.

Where: ChatGPT Ads Manager at ads.openai.com, with reporting available in table view, the Conversions column attribution breakdown, an optional standalone column, and CSV exports. ChatGPT advertising operates in the United States, Canada, Australia, New Zealand, the United Kingdom, Japan and South Korea, and remains unavailable in the European Union.

Why: Conversational ad placements are frequently consumed without a click, leaving click-only attribution unable to register campaign effect. View-through reporting gives advertisers a figure for that activity. By fixing the window at one day and excluding the metric from bidding, billing and cost calculations, OpenAI has limited the scope for the number to inflate reported performance, in a channel where post-view credit at premium impression pricing has drawn sustained scrutiny since the pilot began.