LLM ads are coming — measurement is fuzzy
Marketers are debating whether ads inside large language models will be a real performance channel, and measurement standards are still unclear, even as OpenAI reportedly forecasts massive ad revenue growth. That means testing hypotheses, clear success metrics and a cautious stance on attribution will be central to early experiments. (marketingdive.com) (benzinga.com)
OpenAI is reportedly telling investors it could make $2.5 billion from advertising in 2026 and $100 billion a year by 2030, which is a sharp turn for a company whose chief executive Sam Altman spent years sounding skeptical of ads. (axios.com) (reuters.com) That forecast lands while marketers are still arguing over a simpler question: if an ad appears inside a large language model, what exactly counts as success. Brian Quinn of AppsFlyer wrote this week that buyers still do not know whether these placements can be judged with the same rigor as search, social media, or retail media. (marketingdive.com) The measurement problem starts with how people use these tools. A Google search often ends with a click, but a ChatGPT session can end with a summarized answer, a product comparison, or a next question, which means the platform may shape a purchase without sending obvious referral traffic. (marketingdive.com) (openai.com) OpenAI has already built the kind of product that blurs that line. Its shopping features show product images, details, prices, and links to merchants, while the company says those product results are selected independently and are not ads or influenced by partnerships. (help.openai.com) (techcrunch.com) That creates an awkward middle ground for advertisers. If ChatGPT recommends a blender, a hotel, or a software tool before any paid unit appears, brands may already be competing for inclusion in an answer engine that does not look like a normal ad auction. (help.openai.com) (openai.com) It also scrambles attribution, which is the bookkeeping system marketers use to decide which touchpoint gets credit for a sale. If a user asks ChatGPT for “the best running shoes under $150,” then later buys through Google, Amazon, or a store app, each platform can claim part of the same conversion. (marketingdive.com) The safest early tests will probably look less like old search advertising and more like controlled experiments. Marketers can compare regions, holdout groups, or time periods, then watch for lift in branded search, direct traffic, or sales instead of assuming every click from a chatbot deserves full credit. (marketingdive.com) (appsflyer.com) There is another reason buyers are cautious: OpenAI’s own product line still mixes paid subscriptions, commerce features, and now ad infrastructure. The company’s help pages for shopping research say ads are separate from shopping research, which suggests OpenAI is trying to keep sponsored placements distinct from its organic recommendations. (help.openai.com 1) (help.openai.com 2) If OpenAI does build a large ad business, it will be competing for budgets that already flow to Alphabet’s Google and Meta’s Facebook and Instagram, where buyers have decades of click, conversion, and auction data. The pitch for large language model ads will need to beat not just those companies’ reach, but their reporting systems. (axios.com) (reuters.com) So the near-term story is not that chatbot ads are proven. The near-term story is that one of the biggest artificial intelligence companies in the world appears to see ads as a future giant, while the people who would buy those ads are still working out how to measure one. (axios.com) (marketingdive.com)
