A third of ChatGPT's ads land in conversations that don't fit

A study from AI visibility platform Searchable found that a third of ads served inside ChatGPT conversations had no connection to what users were discussing, AI News reported. Searchable analyzed more than 11,000 ads served in real ChatGPT conversations between July 4 and August 4, pairing each ad with its conversation and sorting the match into one of three bands: unrelated, contextual, or a direct match to the product the user asked about.

The breakdown came out uneven. Just 27% of ads directly matched what the user was asking about, while 40% landed in the contextual band, tied to something raised earlier in the thread rather than the question in front of the user, and 33% missed the conversation entirely. Counted by conversation instead of by ad, 40% of ad-carrying chats contained at least one unrelated ad, and 28% saw every single ad served come back unrelated. More than two-thirds of ads, 68%, showed up in conversations where the user never signaled any intent to buy, book, hire, or compare anything.

The mismatch traces back to how ChatGPT's ad system works. Paid search lets an advertiser bid on the exact keywords a buyer types. ChatGPT's platform instead runs on context hints, where advertisers describe the conversations and topics they want to appear near and OpenAI matches those descriptions against a live conversation happening inside a tool people also use to write, study, and troubleshoot unrelated problems. Performance varied sharply by sector. Marketing and B2B services ads were unrelated 47% of the time, and software and SaaS ads 45%, the two worst rates measured. Data brokers and background-check services came closest to a majority direct-match rate, and no sector crossed 50%.

• Ads analyzed: more than 11,000, served between July 4 and August 4
• Share of ads with no connection to the conversation: 33%
• Share that directly matched the requested product: 27%
• Ads appearing in conversations with zero purchase intent: 68%
• Worst-performing sectors: marketing/B2B services (47% unrelated) and software/SaaS (45% unrelated)

OpenAI's own numbers point the other way. Denise Dresser, OpenAI's chief revenue officer at the time, said in June that the rate at which users dismiss ads inside ChatGPT had fallen by half since the ad business launched in February, and the company treats dismissal rates as its stand-in for relevance. Searchable's window runs through the following August, roughly six months into that rollout, and the irrelevance rate it measured held close to a third the whole time, without moving as ad volume grew. The mismatch matters more as OpenAI leans further into commerce: separately, wallets built to let ChatGPT buy things on a user's behalf are already live, and an assistant recommending the wrong product inside its own answer carries different stakes than an ad a user can simply scroll past. It's also worth naming who ran this particular study: Searchable sells AI-visibility and answer-engine-optimization services, and its own pitch to advertisers is that ranking inside an assistant's organic answer beats paying for a placement next to it, which is close to what this data set is built to demonstrate.

This article is for informational purposes only and does not constitute investment advice.