The Drum
Thrad CCO shares his perspective on The Drum
Roger Dunn, Thrad's chief commercial officer, spent most of his career buying and selling search and retail media. In a member article for The Drum, he sets that experience against what Thrad has learned running paid placements inside AI conversations across a network of more than 50 AI publishers, and frames the difference as ice cubes and snowflakes: keywords are identical and mass-produced, while every prompt is shaped by one person's situation.
A keyword became a good trading unit because it is the same everywhere, which is also why every brand in a category bids on the same few terms and prices only rise. A prompt carries far more. Someone planning a first marathon who mentions an injury and a budget hands an advertiser the need, the constraints and how close they are to buying in a single turn. Citing SimilarWeb, the piece notes that a typical search runs to a few words while prompts to AI assistants often reach sixty or more, and that AI sessions are growing on top of a search market that has held flat since 2020. That is new demand, and no brand owns it yet.
In a conversation, the whole thread does the targeting. A relevance threshold stands in for match types, exclusions apply to topics rather than single words, and brand safety is judged at the level of the conversation. No third-party profile is needed: “The ad reads the conversation, not the person.”
Live campaigns have also shown how often the conversation overturns the brief. A B2B software campaign aimed at senior engineering leaders found its strongest conversations among junior engineers asked to bring their managers a plan for moving off legacy code. A mobile game found its players in conversations about unwinding after work and commuting without signal, not about games. Travel buyers were reachable while a trip was still a daydream, well before dates and budgets existed.
On measurement, Dunn argues that last-click attribution misses most of what happens in a conversation, where people read the answer, ask a follow-up and decide before anything is clicked. Thrad is building a measure it calls attention shift, which compares what a conversation is about after an ad appears with what it was about before. Clicks are still reported, with attention shift on top.
The article also makes the case that the answer is becoming the new shelf. Brands appear either inside the model's answer or in a clearly labelled sponsored unit right after it, across general assistants, in-app chat and retailers' own shopping assistants. The formats that earn attention follow a genuinely helpful answer, are labelled as ads, match the tone of the chat and give the user a next step, such as a follow-up prompt.
For teams planning next year's media, the starting point is simple: take your top 20 keywords and write out the sentence a real customer would say instead. That list becomes the brief for conversational AI, and it usually sharpens search copy along the way.
The Drum published the article on 29 October 2026 in its Industry Insights section, under the headline “Keywords are ice cubes. Prompts are snowflakes.”