Short answer
OpenAI has explained that ads in ChatGPT are chosen by a language model that runs separately from the system writing answers. The model turns the conversation's context into search queries, finds suitable ads and sends them to a separate auction. An auction winner can still go unshown if it fails relevance or timing checks.
Highlights
- The system choosing ads runs on separate models and code from the one writing ChatGPT's answers.
- An auction winner is not shown if it fails the conversation-fit check or misses the delivery window.
- OpenAI cited a $1 billion annualised revenue run rate but shared no data on advertiser returns.
2 min readAuthor: UNIT Journal EditorEditor-in-chief: Uğur Deniz İlhan
As Search Engine Journal reports, OpenAI has published a technical post explaining how ChatGPT ads are chosen and served from conversation context. This is OpenAI's own description, not independent confirmation of how it works in practice.
Do ChatGPT ads influence the answer?
According to OpenAI, the system selecting ads runs separately from the one writing the answer; the two use different models and code. If the ads system is slow or stops, the answer is not delayed. When an ad is shown it carries a "sponsored" label below the response.
How is an ad chosen from a conversation?
A language model called the "orchestrator" manages the selection. It turns relevant parts of the conversation into search queries, which are converted to embeddings so ads can be found without matching wording. Eligibility checks follow, such as location targeting, policy restrictions, budget and pacing. Candidates then go to a separate auction tool, which decides ranking and price, not the model.
Why might an auction winner not be shown?
After the auction, a further check looks at whether the ad fits the conversation. An ad that misses the delivery window is also dropped. Low impressions may therefore stem from relevance or timing problems, not only from the bid.
What should advertisers do?
Because ads are matched to a conversation as it develops, specific customer needs (for example, no charge for guest users) may matter more than a generic price message. Use your existing search terms and customer conversations to decide which needs to stress; the approach of performance marketing teams managing budgets by conversion data applies here too.
- If impressions are low, ask at which stage the ad dropped out (no candidates found, eligibility, auction, late delivery) before raising bids. OpenAI says it logs these stages internally but does not say how much advertisers can see.
- Make the landing page back up the promise in the ad copy; the post stresses that the page needs to support the offer.
- Keep budgets small and treat them as learning: OpenAI shared no performance benchmarks, so track your own conversion and cost data separately.
Frequently asked
- What do the revenue and advertiser figures show?
- OpenAI says ChatGPT Ads reached a $1 billion annualised revenue run rate in under 200 days after launch and that more than 100,000 advertisers have run ads. The post includes no CPA, ROAS or conversion-rate data on advertiser returns.
Sources
- Search Engine Journal ·
Follow UNIT Journal
What's new in search, AI and technology, in your feed every day.
