ChatGPT Ads now counts views in its conversions

KEY TAKEAWAYS

Since late September, ChatGPT Ads conversion totals can include one-day view-through, so older and newer reports are not comparable.
Compare ChatGPT and Search on the same rung. Click-only against click-only, with matching windows.
The gap between platform conversions and your own matched orders is a tracking diagnostic, not a performance result.
Published wins use undisclosed benchmarks. Scale only on order-matched or incremental CPA.

The Conversions column in ChatGPT Ads changed meaning at the end of September. The spend did not change. The clicks did not change. The count of conversions did, because it can now include people who saw an ad and never clicked. Any CPA you compare with Search has to account for that first.

What changed in ChatGPT Ads reporting?

Three changes, in order.

18 and 19 August 2026. OpenAI's help articles added one-day view-through reporting. PPC Land, reading those pages at the time, reported it as a separate column outside the Conversions total, with no effect on CPA or bidding.
30 September 2026. OpenAI's API changelog, as reported by TechWyse, said conversion totals now include eligible outcomes after impressions as well as clicks. Totals for the same campaign and dates can differ from earlier reports.
Now. OpenAI's Measure Results page offers click windows of 7, 14 or 30 days and a view window of zero or one day. With the view window at one day, view-through conversions sit inside the Conversions total.

Two more details matter. The model is last-touch, and the page says it is not multi-touch. And OpenAI's conversion measurement page says reported totals may include modelled conversions where available, without saying how many.

So a ChatGPT Ads CPA can now rest on clicks over 30 days, views over one day, and an unknown modelled share. A Search CPA in most accounts rests on clicks alone.

Why do the published results not settle the question?

Because each one sits on a different basis, and none states it fully.

Published result	What it measures	What is missing
WeightWatchers, attributed CPA 15.3% lower	DV Rockerbox attribution against a benchmark	OpenAI calls the benchmark blended paid search; DV's release calls it established. Neither gives period, model or volume
Dose, significant lift, 67% of incremental purchases from new customers	WorkMagic geo test	Spend, dates and interval
Portland Leather, 93% of visitors new	Triple Whale visitor data	Cost and conversions
A Common Thread Collective client, ROAS 3.3x to 6.8x	Same $9,620 of spend, different attribution models	Which model is closer to lift

The last row is the useful one. Search Engine Journal reported it in September. The same spend produced a return that doubled depending on the model chosen. That spread is far larger than the 15.3% gap in the WeightWatchers result. Model choice alone can decide whether the channel looks good.

Two other figures circulate in coverage of OpenAI's 5 October announcement: a 2.3x incremental-orders multiple and a share of view-through conversions within an hour. We could not find either on the OpenAI page we read, so we leave them out.

Where do the sources disagree?

On whether measurement works. In June, agencies quoted by AdNews said the new pixel and Conversions API had changed the commercial case, and one said results seen off the platform mattered more than any gap. In August, MediaPost reported one agency seeing 57 platform clicks against fewer than 20 analytics visits, and zero conversions. On 1 October, Digiday reported Accuracast seeing about 100 platform clicks against 20 in its own tracking, and form fills arriving while the platform showed zero conversions. AdRoll, Jellyfish and another agency told Digiday they saw no click gap.

Our reading: all of these are small tests. The positive accounts mostly rely on the platform's own counts. The negative ones compare the platform with independent tracking, which is the stronger check. But since some agencies find no gap, the gap is usually a setup issue in a specific account, not a constant. Treat the platform total as a ceiling until you have reconciled it.

On the default click window. Enalitica, a vendor, says the interface defaults to seven days. TechWyse says API pulls default to 30 days when no window is passed. OpenAI's help page names no default. The safe answer is to read the setting in your account and pass windows explicitly in every export.

On the offline window. LiveRamp's documentation for OpenAI's invitation-only offline conversions programme refers to a 21-day attribution window. That does not match the 7, 14 or 30 days on OpenAI's help page. If you send offline conversions, ask which window applies before you compare them with online ones.

How do you put a ChatGPT CPA on one basis?

Build a ladder. Each rung divides the same spend by a stricter count of conversions.

CPA at rung = spend ÷ conversions counted at that rung

Platform CPA. Conversions as Ads Manager reports them, with views and any modelled conversions included.
Click-only CPA. The same report with the view window set to zero days.
Order-matched CPA. Orders or leads in your own system that carry ChatGPT's click reference, called oppref, or your UTM tag, inside the same window.
Incremental CPA. Spend divided by the extra conversions a geo holdout attributes to ChatGPT.

Then compute one ratio: platform conversions ÷ order-matched conversions. Call it the gap ratio. It tells you how much of the platform's count you cannot see in your own data.

Hypothetical example. These numbers are invented to show the method. A retailer spends $10,000 on ChatGPT Ads in a month. Its click-based non-brand Search CPA is $100.

Rung	Conversions	CPA	Against Search at $100
Platform (30-day click, 1-day view)	125	$80	20% cheaper
Click-only (30-day click, no view)	90	$111	11% dearer
Order-matched (oppref in backend)	70	$143	43% dearer

The gap ratio is 125 ÷ 70, or about 1.8. The same month reads as a win, a small loss or a clear loss, depending on the rung.

Now weight for new customers. Suppose 60% of the 70 ChatGPT orders are new, against 35% of Search orders. Cost per new customer is $10,000 ÷ 42, or about $238, on ChatGPT. On Search it is $100 ÷ 0.35, or about $286. On that basis ChatGPT wins, even at rung three. This is the question the published new-customer figures point at, and the one a CPA comparison alone misses.

Which rung should you compare with Search and Meta?

Match the rung to the other platform's default.

Google Search. Google's help page sets the click-through window default at 30 days and the view-through default at one day. Most teams report Search CPA on clicks. If yours does, compare it with ChatGPT's click-only CPA at 30 days.
Meta. Jon Loomer's guide gives Meta's default as 7-day click and 1-day view. Compare it with ChatGPT at 7-day click and 1-day view. Our Meta attribution guide covers which Meta setting to report.

A ChatGPT CPA with views included, set against a click-only Search CPA, flatters ChatGPT by construction.

Does the billing choice change what the total means?

Yes. OpenAI's conversion-optimised campaigns page says campaigns billed per click optimise toward conversions after a click. Campaigns billed per impression optimise toward a broader set that includes conversions driven by views.

So an impression-billed campaign is steered toward the same view-through conversions that now sit in its total. It is partly grading its own work. For those campaigns, judge on rungs two to four. You cannot switch billing in place; OpenAI says to clone the campaign. We covered a similar view-through question in Google's Demand Gen.

What should you do in each situation?

What you see	What it most likely means	What to do
Platform CPA beats target, click-only does not	Views carry the result	Hold spend. Test view value with a geo holdout
Gap ratio above about 1.5	Conversions you cannot see in your data	Check tracking before judging performance
Platform shows zero, your backend shows leads	Tracking not linked or event mismatch	Fix setup. OpenAI says mismatched history will not backfill
Platform clicks far above sessions	Lost click references or slow landing pages	Compare by landing page; test redirects
Order-matched CPA beats target	A real candidate	Raise budget in steps, then confirm with a holdout
Impression billing on a conversion campaign	Bidder targets view-through	Judge on click-only and order-matched rungs

The gap ratio threshold of 1.5 is our rule of thumb, not a published figure. Set your own from the first clean month.

How to audit a ChatGPT Ads test in seven steps

Record the settings. Click window, view window, billing type and the single conversion event, which OpenAI says cannot be changed after creation. Mark 30 September on any trend chart that crosses it.
Check the plumbing. Confirm oppref survives every redirect. Send pixel and Conversions API events with the same event ID so OpenAI can deduplicate. If you use a partner link, Singular warns that editing one parameter stops conversion data reaching OpenAI with no error shown.
Export twice. Pull the same date range with the view window at one day and at zero days. In API pulls, pass both windows explicitly.
Drop the last two days. OpenAI asks for 24 to 48 hours before attributed conversions appear, and dates them to the ad event by default.
Match your own records. Count orders or leads carrying oppref or your UTM tag in the same window. Compute all four rungs and the gap ratio.
Compare like with like. Use the rung that matches the other platform's default, then weight by new-customer share.
Hold out before you scale. Since 22 May 2026 you can target US campaigns by state, DMA or ZIP. BlueAlpha, a measurement vendor, suggests at least four weeks of treatment after a long pre-period. Haus says there is "no existing benchmark for how attribution compares to true incremental lift." Our geo test guide covers which design fits.

What this means for a test budget

Accuracast told Digiday that open questions stop it moving client test budgets of £10,000 to £20,000 up to £100,000. The ladder does not fix that. It does tell you which number to defend when you ask for more, and which to ignore.

ChatGPT ads and ChatGPT answers reach the same people. Ethos Engine is building tools to measure brand visibility in AI answers. If that would help, join the waitlist.

Questions

Does ChatGPT Ads count view-through conversions?

Yes, if the view-through window is set to one day. OpenAI's help page says eligible view-through conversions are then included in the Conversions total. Set the window to zero days and the total holds click-through conversions only. When a click and an impression both qualify, the click gets the credit.

What attribution windows can I choose in ChatGPT Ads?

Click-through windows of 7, 14 or 30 days, and a view-through window of zero or one day. OpenAI says these are reporting settings and do not change bidding, billing or optimisation. Its help page names no default, and secondary sources disagree on it, so read the setting in your own account.

Why does ChatGPT Ads report more clicks or conversions than my analytics?

Some agencies have seen large gaps, such as about 100 platform clicks against 20 in their own tracking, while others report none. Common causes are a click reference lost in a redirect, an event that does not match the campaign's conversion, view-through credit, and the 24 to 48 hour reporting delay.

Can I run an incrementality test on ChatGPT Ads?

Yes, as a geo test. OpenAI added US state, DMA and ZIP targeting in May 2026 and names Haus, Measured and WorkMagic as geo experiment partners, while calling that work early. There is no built-in lift study, so you split regions yourself or through a partner.

Sources

OpenAI Help Center, Measure Results help.openai.com
OpenAI Help Center, Conversion measurement help.openai.com
OpenAI Help Center, Conversion-optimized campaigns help.openai.com
OpenAI, Building advertising for the way people use AI openai.com
OpenAI, New ways to buy ChatGPT ads openai.com
DoubleVerify, DV Launches Attribution Measurement and Advances Brand Suitability for ChatGPT Ads s206.q4cdn.com
Google Ads Help, About conversion windows support.google.com
PPC Land, ChatGPT Ads gains view-through conversions with a fixed 1-day window ppc.land
PPC Land, ChatGPT Ads gains app attribution in 7 markets as AppsFlyer, Adjust go live ppc.land
TechWyse, ChatGPT Ads Gives Advertisers More Scale and a New Conversion Baseline techwyse.com
Search Engine Journal, 6 Months Into ChatGPT Ads, Advertisers Still Don't Know What Good Looks Like searchenginejournal.com
Digiday, OpenAI's measurement gaps are keeping ChatGPT ads budgets at test level digiday.com
MediaPost, Data Reveals ChatGPT Ads Not Performing mediapost.com
AdNews, Advertisers say ChatGPT Ads have closed the measurement gap adnews.com.au
Enalitica, ChatGPT Ads Attribution enalitica.com
LiveRamp, The OpenAI Conversions API Program for Offline Conversions docs.liveramp.com
Singular, OpenAI Ads (ChatGPT) Attribution and CAPI support.singular.net
MarTech360, Haus partners with OpenAI to provide incrementality testing and measurement for ChatGPT Ads martech360.com
BlueAlpha, How to Measure Whether ChatGPT Ads Actually Drive Revenue bluealpha.ai
Jon Loomer, Meta Ads Attribution Setting: A Complete Guide jonloomer.com
Relevant Audience, OpenAI names 20 ChatGPT Ads measurement partners and tests image-led ads in the US relevantaudience.com

Esh

Founder, Ethos Engine. Builds the Digital Authority Score and runs the placement marketplace. esh@ethosengine.io

NEXT STEP

See where your brand stands.

Your Digital Authority Score, who AI recommends for your target questions, and the placements that close each gap. Your first report is free.

Get your free report →

ChatGPT Ads now counts views in its conversions. Restate the CPA before you compare it with Search