How do ChatGPT ads work? The format, targeting, and 'answer independence'

Updated 2026-06-04

ChatGPT’s ad system, which launched February 9, 2026, is architecturally different from search advertising. There are no keywords, no bid auctions on search terms, and no ad injection into the AI’s answer text. Here’s how it actually works.

The placement

Every ad in ChatGPT appears as a clearly labeled “Sponsored” text block positioned below the AI’s answer — not inside it, not interleaved with it. The format is fixed:

• Headline: up to 40 characters
• Description: up to 150 characters
• Link: one clickable URL

The Sponsored label is required and visible. The block is visually separated from the answer. There is no mechanism for advertisers to inject text into ChatGPT’s responses, appear inside them, or otherwise blend with the AI-generated content.

If you want to see exactly how this looks with your own headline and description, the ad mockup tool on this site lets you build and download a preview image.

How targeting works without keywords

Traditional search ads rely on keywords: an advertiser bids on “running shoes,” their ad appears when that term is searched. ChatGPT’s system does not work this way because ChatGPT conversations are not keyword-driven queries.

Instead, advertisers submit “context hints” — plain-language descriptions of the types of conversations or user intents where their ad is relevant. A fitness brand might write something like “users planning a workout routine” or “people asking about injury recovery.” OpenAI’s relevance model then reads the conversation context and decides whether a given session matches any active context hints.

The underlying ad infrastructure is built on Microsoft Advertising, reflecting the existing partnership between OpenAI and Microsoft. Buying options at launch were CPM (cost-per-thousand impressions) and CPC (cost-per-click). Launch CPM rates were approximately $60 and fell toward $25 as inventory scaled. OpenAI’s recommended CPC bid at launch was $3–5.

This context-matching approach means advertisers cannot directly target specific words a user typed, topics pulled from past sessions, or demographic attributes derived from chat content.

Answer independence

OpenAI has stated explicitly that ads do not influence ChatGPT’s answers. According to the company’s published policy:

• Advertisers cannot shape, rank, or alter what ChatGPT says in response to a query.
• Advertisers do not receive user chat content or personal data from conversations.
• The ad matching happens at the session level; the ad unit is appended after the answer is generated.

The practical claim is that the answer and the ad are parallel outputs — the answer is generated by the model without any advertiser influence, and the ad is selected from the eligible pool afterward.

This design is OpenAI’s stated guarantee against editorial corruption of the model. Critics have questioned whether that wall holds in practice, pointing to a December 2025 The Information report that described internal mock-ups suggesting sponsored content could be prioritized in responses. OpenAI denied those mock-ups reflected actual product direction. Researcher Zoë Hitzig resigned from OpenAI on February 11, 2026 — two days after launch — writing in a New York Times essay that she believed the ad system put at risk what she called ChatGPT’s “archive of human candor.”

The format itself — a labeled block below the answer — is verifiable by any user who sees it. Whether the model’s underlying behavior is truly independent of advertiser relationships is a harder claim to audit externally.