Are ChatGPT ads and OpenAI Ads the same thing?

Yes. OpenAI's product is named OpenAI Ads and runs at ads.openai.com; the ads themselves render inside ChatGPT conversations, which is why almost everyone searches for "ChatGPT ads". Same auction, same API, same ad account.

How are ChatGPT ads targeted?

There are no keywords and no interest graph. Targeting is a campaign-level geo block plus context hints on each ad group — short free-text phrases describing the conversations the ad should appear in. The relevance model reads those hints together with the card's title and body, so specific creative is part of the targeting.

What do ChatGPT ads cost?

OpenAI quotes a typical $15-60 CPM. At the reported ~0.91% click-through rate that works out to an effective CPC of roughly $1.65 to $6.59, around $3.30 at the midpoint. Budgets and bids are set in micros of the account currency (25000000 = 25.00).

Is there an API for ChatGPT ads?

Yes — the OpenAI Ads Advertiser API at api.ads.openai.com/v1, authenticated with an Ads API key scoped to one ad account. Cogny exposes 32 of its endpoints as MCP tools, so an agent can read insights and push campaign, ad group and ad changes without you writing an API client.

How do I report on ChatGPT ad traffic in GA4?

Tag every card's target URL and give each ad its own utm_term so creatives stay separable when they share an ad group and landing page. Then cross-check sessions where the source is chatgpt against the platform's own click counts — the gap between them is usually a tagging or consent problem, not a delivery one.

ChatGPT Ads: What They Are, What They Cost, and How to Run Them From an Agent

OpenAI named the product OpenAI Ads. Nobody searches for that. Pull up Google Trends for Sweden over the past month and "chatgpt ads" peaks at 100 while "openai ads" tops out around 22 — a roughly 4–5× gap on two names for the same auction. So: ChatGPT ads are OpenAI Ads. Same ad account at ads.openai.com, same Advertiser API, same chat cards.

This post is the channel explainer. If you want the integration details instead — every tool, every unit convention — read OpenAI Ads Is Live in Cogny.

What a ChatGPT ad actually is

A chat card: a title, a body, an image and a target URL, rendered below a model response and labelled Sponsored, with a clickable favicon next to it.

What sits behind that card is where the channel stops resembling search:

No keywords, no interest graph. Targeting is a campaign-level geo block plus context hints on each ad group — short free-text phrases describing the conversations the ad belongs in.
The relevance model reads the hints and the card copy together. An ad that echoes the situation the user is actually in outscores a generic one that could sit under any conversation. That inverts the usual instinct to write three strong ads and let the machine sort it out: here, breadth of specific creative is the targeting.
Three bidding types — impressions, clicks, conversions. Conversions requires event settings wired to a pixel or the Conversions API first.

What it costs

OpenAI quotes a typical $15–60 CPM. Reported ChatGPT-ads CTR sits around 0.91%, which puts the effective CPC in a $1.65–6.59 band, ~$3.30 at the midpoint. That band is the useful number: it tells you whether an ad group's max bid can win at all, and it is the first thing worth checking when a group logs budget but no impressions.

Two unit conventions bite everyone once:

Budgets and bids are micros of the account currency — 25000000 = 25.00. Insight metrics are not: spend, cpc and cpm come back in plain currency units, and ctr is a fraction (0.0226 = 2.26%).
An insights end_date must be a completed day in the account timezone. Ask for a range ending today and the API answers time_ranges.end cannot be in the future. "Month to date" means through yesterday.

How the campaign should be built

OpenAI publishes two guides — the Creative Launch and Scaling Guide and the Context Hints Guidance. The structure they describe is unusually prescriptive, and conformance is most of the job in the first month:

Element	Guidance
Ad groups	One product, theme or intent area each — hints and bids are set here
Ads	~10 per ad group, ~100 per campaign
Context hints	5–10 per ad group, 3–6 words each: `[product or category] + [use case] + [qualifier]`
Card title	16–24 characters (API allows 50)
Card body	32–48 characters (API allows 100)
Creative image	Not the logo as primary visual — the favicon is already on the card
Bidding	Launch on Clicks; move to Conversions once tracking is live
Tagging	A unique `utm_term` per ad
Customer lists	25,000 matched users minimum, 100,000+ recommended

The floors matter as much as the ceilings. A 9-character title is not "concise", it is unused space in a fixed-size card.

A hands-on build walkthrough: our partner agency Growth Hackers Sweden wrote up how they built a campaign on this channel end to end — Build an OpenAI Ads campaign. Their run was one campaign, ten ad groups, 100 ads and 80 context hints, with a build script that fails on any card or hint outside the bands above. Worth reading before you open Ads Manager.

Running it from an agent

Ads Manager shows one campaign at a time and no export of ad-level daily rows, which makes the weekly pass exactly the kind of work you should not be doing by hand.

OpenAI Ads is a live MCP integration in Cogny. Create an Ads API key at ads.openai.com → Settings → API keys, paste it on the OpenAI Ads card under warehouse Settings → Cogny Managed MCPs, and 32 tools over the Advertiser API become callable from Cogny chat, the Cogny CLI, or any MCP client pointed at app.cogny.com/mcp. One key covers one ad account, so there is no account picker.

cogny tools call mcp__OpenAI_Ads__tool_get_insights --input '{
  "aggregation_level": "ad_group",
  "start_date": "2026-08-01",
  "end_date": "2026-08-31",
  "segment": "country"
}'

Read tools are auto-approved; writes — create, update, pause, archive — ask first, like every other ads MCP in Cogny.

Two scheduled templates run the recurring work:

OpenAI Ads (ChatGPT) Optimization Audit grades performance: context-hint relevance crossed with CTR, cards over 2,000 impressions under 1% CTR, geo and device fit, bids that cannot win, budgets that never deliver, and whether the conversion pixel is still receiving events.
OpenAI Ads Launch & Scaling Playbook grades structure against the table above and drafts the fixes as ready tool_update_ad_group / tool_create_ad payloads — left paused, because passing format validation is not launch approval.

Each finding names the campaign, ad group or ad id, so the ticket it becomes can be executed through the same MCP once you approve it.

Where this fits with the rest of AI search

Buying chat cards is the paid half of showing up inside an assistant. The unpaid half is being the source the model cites in the first place — which is GEO, and a different set of work entirely. Run both and the same question — does this brand exist inside ChatGPT? — gets two answers instead of none.

Connect OpenAI Ads from Settings, or run npx @cogny/cli init and let your agent do it.