Sponsorship Terms for Prompt and Response Placements

The clause that decides whether one of these deals is worth signing is the log clause, and it disqualifies most sellers before price is discussed. I ask for response-level records — response ID, timestamp, model version, hashed query, placement position, click — delivered daily to a bucket you own. Roughly two in three publishers I approach cannot produce them, which makes it the fastest qualification filter in this channel. Everything else in the term sheet is downstream of whether you can count what you bought. Who actually sells this inventory: Newsletters with an assistant attached, wrapper apps with retained weekly users, vertical copilots in law, medicine, logistics and trading, developer assistants, and agent directories. I qualify on retained weekly users and query mix, never on registered accounts or download totals, and I ask for the query distribution before I ask for the rate. A seller who will show you traffic but not query mix is selling you a category they cannot describe. Pricing structures I will sign: Per qualifying response against a defined intent set, with the definition written into the contract rather than agreed on a call. CPM against rendered impressions of the sponsored block, where response volume is stable enough for that to mean anything. Flat monthly with a volume floor and an automatic make-good in inventory or cash when the floor is missed. Exclusivity, when it is worth buying at all, is bought per intent cluster and per quarter, never per category, which is how sellers charge you for traffic that was never going to arrive. Clauses I insist on: Model-version notice within a set number of days, because a swap changes what fires. An audit right with a sampling protocol you can execute without the publisher's help. Brand safety defined by adjacent query rather than by topic list. Fraud defined to include synthetic query volume, with the burden of proof on the seller. A kill clause with pro-rata refund if measured qualifying responses fall below the floor for two consecutive periods. Verification after signature: Logs land daily. I reconcile qualifying responses against your own endpoint weekly and raise a make-good the week the floor is missed, rather than at quarter end when sellers expect to negotiate it away. The reconciliation query is written for your warehouse and stays there. What I refuse: Compensation in the publisher's token or equity, which makes you a stakeholder in the numbers you are supposed to be auditing. Deals where placement is decided by the model with no rule anybody can read. Renewals negotiated off the publisher's own dashboard. Any arrangement where the sponsored response is not disclosed to the end user as sponsored. What you receive: A qualified seller list with scores and the stated reason for each rejection, benchmark rate ranges for the formats you are buying, a term sheet drafted to be handed to your counsel rather than rewritten by them, the verification protocol with its log schema, and me in the room until signature. Who this is not for: Anybody who wants volume across many sellers. Three to six well-instrumented deals is what this channel supports at present, and a plan with twenty logos in it is a plan nobody will verify. Also not for teams that need the deal closed this month regardless of terms; the log clause is exactly the one that takes time. How it runs: Two weeks to a qualified list and rate benchmarks, then negotiation at the pace of the counterparties, typically three to eight weeks to signature per deal. I do not take a percentage of the media, so recommending that you walk away costs me nothing.