Platform Targeting on ChatGPT Ads: the Setting Nobody Configures
By Keshav Parsai 7 min read
In brief
ChatGPT Ads platform targeting spans five surfaces but offers no surface-level reporting. Learn how to avoid blending unlike audiences.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 December 2026
Five checkboxes sit in the campaign setup, all ticked by default, and almost nobody touches them. OpenAI's campaign documentation lists them plainly: Android app, Android web, desktop web, iOS app, iOS web.
The reason to touch them is not that one surface performs better. Nobody knows that, including us. The reason is that the setting lives at campaign level and reporting has no platform dimension, so whatever you leave on is permanently fused into a single result.
Two audiences, five labels
Start from what a surface actually indicates about the person on the other side of it.
Someone on desktop web is at a computer. For a B2B product that is overwhelmingly a working-hours context: a work machine, a work task, a work problem being thought through in a work session.
Someone in the iOS or Android app is on a phone, usually outside a desk context. That is where personal use concentrates: evenings, commutes, the household admin and general curiosity that make up most consumer ChatGPT usage.
Mobile web sits awkwardly between the two and is the least interpretable of the five. It is a phone, but it is a phone in a browser, which on iOS often means a link opened from elsewhere rather than a session someone started deliberately.
That is the substance of the split, and it is not a claim about performance. It is a claim about context, and context is what a channel matching on conversation meaning is trading on. A B2B advertiser with all five surfaces on is buying two materially different populations under one bid, one budget and one reported number.
The setting is campaign level, and that is the whole problem
If platform were a reporting dimension, none of this would matter much. You would leave everything on, look at the breakdown after two weeks, and switch off what did not work. That is how this decision is made on every other major platform.
Ads Manager reports impressions, clicks, spend, CTR, average CPC, average CPM and conversions at campaign, ad group and ad level. Platform is not one of those levels. There is no device report, no surface breakdown, no segment control.
So the sequence available on Google and Meta is unavailable here. You cannot leave it on, measure, then prune. The only way to see performance by platform on ChatGPT Ads is to decide in advance and run separate campaigns, because the setting sits at campaign level and so does the reporting boundary you would need.
Call it a pre-committed split: a setting you must configure before you have the evidence that would tell you how to configure it.
What to actually do
For a B2B advertiser running a first campaign, the defensible default is desktop web only.
Not because it performs better. Because it is the surface where the buying context is least ambiguous, which means the result you get back has one interpretation instead of several. If desktop web at a four dollar CPC produces an unacceptable cost per qualified lead, you have learned something about the channel for your product. If a blended five-surface campaign produces the same number, you have learned that some unknown mixture of work and personal traffic averaged out badly, which is not actionable.
Then, if budget allows, run app surfaces as a second campaign with their own budget rather than as a hedge inside the first. The point of the second campaign is not to be fair to mobile. It is that a separated result is readable and a blended one is not.
Where this advice breaks
Two problems, and neither is small.
Volume. The most common failure mode on this channel is not spending at all. Restricting to one of five surfaces cuts your match surface substantially, on a platform where new campaigns frequently struggle to exhaust even a 25 USD daily budget. If your campaign is already underdelivering, tightening platform targeting makes the primary problem worse to solve a secondary one.
Budget arithmetic. Splitting into two campaigns divides the money, and we have argued elsewhere that a divided budget on this channel often means neither half accumulates enough spend to read. A 100 USD a day advertiser splitting across desktop and app campaigns ends up with two campaigns that each fail to reach a conversion-readable sample.
So the honest rule is conditional on budget. Below roughly 100 USD a day, pick one surface and accept that you are sampling rather than comparing. Above that, split into separate campaigns. In neither case leave all five on and expect to work out later what happened, because later never arrives.
This is reasoning, not measurement
Worth being blunt, because the argument above is built entirely on inference.
OpenAI publishes no performance data by platform. No advertiser has published a controlled comparison of desktop web against app surfaces. There is no cross-advertiser benchmark to appeal to, and InPromptAds has no first-party account data. The claim that desktop web carries a cleaner B2B context is drawn from how people generally use computers and phones, not from any observed difference in ChatGPT Ads results.
It is also possible that the surface matters less than it would on a display channel, because matching happens on conversation meaning rather than on browsing context. Someone researching compliance software in the iOS app at nine in the evening may be exactly the buyer you want, doing the work they could not do during the day. That reading is equally consistent with the documentation, and nothing published resolves it.
What survives either reading is the structural point. Whatever you believe about surfaces, you cannot measure the difference inside one campaign, so the setting deserves a deliberate answer rather than a default.
What we cannot tell you
- Whether any platform performs better than another. OpenAI publishes no performance data by platform and no advertiser has published a comparison.
- How ChatGPT usage divides between app and web. OpenAI has not published a surface breakdown of its user base.
- What share of desktop web sessions are work sessions. Not published, and not inferable from anything in Ads Manager.
- What mobile web traffic represents. The least interpretable of the five surfaces, with no published characterisation.
- Whether platform reporting will be added. Not on any published roadmap.
Quick answers
What platform targeting options does ChatGPT Ads have? Five: Android app, Android web, desktop web, iOS app and iOS web. You select one or more at campaign level.
Can I see performance by platform? No. Reporting runs at campaign, ad group and ad level, with no platform or device dimension. Separating surfaces requires separate campaigns.
Which platform should a B2B advertiser choose? Desktop web, as a default rather than a finding. It carries the least ambiguous working-hours context, which makes the result easier to interpret. No performance data supports the choice.
Should I leave all five platforms on? Only deliberately. All five on means work and personal contexts are fused into one number with no way to separate them afterwards.
Does restricting platforms hurt delivery? It reduces your match surface, and underdelivery is already the common failure on this channel. Below roughly 100 USD a day, do not split into separate campaigns as well.
Is platform targeting set per ad group? No. It sits at campaign level, alongside objective, budget, dates, locations and custom audiences.
Sources
| Claim | Source | Tier |
|---|---|---|
| Advertisers select one or more of Android app, Android web, desktop web, iOS app or iOS web | OpenAI Help Center, Create Campaigns for ChatGPT Ads, 2026 | Confirmed, primary |
| Platform targeting is a campaign-level setting | OpenAI Help Center, Create Campaigns for ChatGPT Ads, 2026 | Confirmed, primary |
| Reporting runs at campaign, ad group and ad level | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| No platform or device reporting dimension exists | OpenAI documentation, by absence | Absent |
| No published performance data by platform | OpenAI documentation, by absence | Absent |
| Minimum daily budget of 25 USD | OpenAI Help Center, Create Campaigns for ChatGPT Ads, 2026 | Confirmed, primary |
| Desktop web as the cleanest B2B context, and the pre-committed split framing | Newtation reasoning from usage context | Inference, ours |
Related reading
- How to Structure a ChatGPT Ads Account When There Are No Keywords
- Location Targeting Inside ChatGPT Ads: Country, Region, DMA and Postal Code
- How to Measure ChatGPT Ads, and What You Cannot Measure Yet
- How Many Ad Groups Should One ChatGPT Campaign Have?
Changelog
12 September 2026, v1.0. First publication. Names the pre-committed split created by campaign-level platform targeting with no platform reporting dimension behind it.
Field kit
Tools
Site
OpenAI (primary)
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.
