Sponsorship Deals Inside AI Answers, Negotiated and Audited

The clause that decides whether one of these deals is worth signing is not the rate. It is what happens when the model declines to render your placement, which it does on a share of matched queries no publisher puts in writing until asked — the assistant, not the ad server, makes the final call on whether a sponsored element appears in an answer. Sponsored answer inventory currently prices between roughly thirty and eighty dollars CPM depending on the publisher and the query class, against eight to fifteen for the same advertiser's open display, and that difference is only defensible if unrendered impressions settle in cash rather than in more inventory. We negotiate that first and the rate second. What we negotiate: Exclusivity granularity. Keyword-level exclusivity is close to worthless on these surfaces, because the model paraphrases the user before it matches anything. We push for query-class exclusivity with a written definition and a sample of 200 representative queries attached to the contract as an exhibit, so a dispute six months later is settled by reading rather than by opinion. Rendering and make-good. A defined render rate, a monthly floor, and cash settlement below it. Publishers open by offering additional impressions, which compounds the original problem rather than curing it. Reporting. Query-level or log-level, delivered to your side at a cadence written into the agreement, with an audit right that names an auditor and a timeframe. A monthly PDF is not reporting, and we have walked away from otherwise good economics on that single term. Adjacency. Your placement can appear beside an answer that is wrong about your product, your pricing or your regulatory status. We negotiate a takedown obligation with a response window rather than a general brand-safety recital. Termination. Model and routing changes are the normal case, not the exception. We write a termination right triggered by a defined drop in render rate, so a deal that stops working can be ended without an argument. Rate benchmarking: We benchmark against what comparable advertisers in your category are actually paying, expressed as effective cost per rendered placement rather than per matched query. On every surface we have bought, the two differ by a wide margin, and the gap is where most of the money is lost. What we will not do: We do not structure undisclosed placement. If the reader cannot tell that a commercial relationship produced the answer, it is not a deal we will put our name to, whatever the local market permits. We do not take rev share on placement we have not seen disclosed. We do not sponsor answers to symptom, dosage or eligibility questions in health. And we do not take token promotion, unregistered offerings, or any crypto engagement where the asset itself is the thing being advertised. Who this is not for: Advertisers who want a signed deal this month. A negotiation with a serious publisher on these terms runs six to ten weeks, and the terms above are what takes the time. Brands treating this as a test budget: at these rates, a test that cannot be read is a donation. And anyone who needs a guarantee of fixed position, which no publisher on an assistant surface can honestly offer — any that does is selling a product they do not control. How we work: We negotiate on your paper or theirs, sit in the calls, and hand you a term sheet with every term above marked walk-away, negotiable, or nice to have, before the first conversation with the publisher. You keep the relationship. We take no commission from the publisher and will not accept one.