OpenAI Ads has no device targeting: you can’t run desktop-only

The setting that isn’t there

A furniture store sells a $1,400 sit-stand desk — a considered purchase its buyers almost always finish on a laptop, not a phone. On Google Ads the move is one line: a mobile bid adjustment of −100%, and the campaign effectively goes desktop-only. In OpenAI Ads Manager, that setting doesn't exist. Ask the API to create a campaign restricted to desktop and it rejects the field:

targeting.device = "desktop"  →  400: unknown field — targeting supports "locations"

The only thing targeting accepts is geography. The campaign targeting docs are explicit: "OpenAI Ads supports country, region, and DMA targeting." No device, no bid adjustment, no exclusion — the desk store and a phone-case store get the same delivery surface.

What OpenAI Ads targets

Targeting on OpenAI Ads is one dimension deep. A campaign carries a targeting.locations.include array — country, region, or DMA — and that is the whole targeting surface. There are no keywords, no device settings, and no bid modifiers. You describe intent in the ad group's context_hints field and the model matches conversations; you fence delivery by geography and nothing else.

So "desktop-only" has nowhere to live. There is no field to set it in, and there is no device bid you could zero out to fake it.

Why device is a report, not a lever

The confusing part is that device is in the platform — just on the wrong side of it. The insights API segments performance by exactly three dimensions: product, country, and device. You can pull impressions, clicks, spend, and conversions split by desktop versus mobile and see precisely how each performs.

That is the asymmetry. OpenAI Ads lets you see the device, not choose it. Google exposes device as both a report and a control; OpenAI exposes it as a report only. The insights endpoint will happily tell you that mobile drove 40% of your desk campaign's spend and none of its conversions — and there is no setting that lets you do anything about it directly.

The reason is structural. Google's device adjustment rides on a per-device auction. OpenAI's match happens inside a model against conversation context, and the ad renders in the same ChatGPT surface whatever the device — so there's no device-level knob to expose in the first place. Same reason there's no search-terms report: the Google lever has no equivalent here.

What to do when you can’t exclude mobile

The device switch is missing, so work the inputs and the outcomes it can't reach:

1. Read the device segment you can't target. Pull insights segmented by device on a schedule. It won't change delivery, but it sizes the problem: if mobile is 40% of spend and 0% of conversions, that's the number that justifies whatever you do next.
2. Steer with context_hints, honestly weakly. A description skewed toward high-consideration, comparison-heavy intent is a soft proxy for desktop buyers, not a device filter — don't oversell it to yourself.
3. Move the device logic post-click. You pay for the mobile impression regardless, so put the control where you actually have it: a responsive landing page, or routing phone visitors to a lighter capture flow. The click is priced; the conversion path is yours.
4. Judge at the campaign level. If a device segment is pure waste, the only real lever is the campaign itself — pause, reframe the offer, or reallocate budget. There is no scalpel below that.

How the check actually ran

The rejection above is the OpenAI Ads API answering for itself: hand it a device restriction and it names the one field it accepts. On an account running on Adjar, targeting is plain-text config, so the absence is visible rather than surprising — a device = "desktop" line has no schema slot to sit in, and the reviewer sees geography is the only fence available:

[campaigns.targeting]
locations = ["US", "US-CA"]   # country / region / DMA — the only targeting there is
# device = "desktop"          # unsupported: OpenAI Ads has no device targeting

The report Adjar pulls back carries the device segment the platform will give you — the desktop-vs-mobile split per campaign — so the number you can't target on is at least a number you can see. You measure the split, act where the platform lets you, and stop pretending there's a switch that isn't there.