Sponsored answer placements in ChatGPT and Perplexity

What is actually purchasable: Ad inventory inside AI assistants is sold against a placement on an impression basis, not as a query-level auction, so there is no keyword report to optimise and no bid to move on a Tuesday. That changes what the money is for. I treat this as a brand test with a hard cap, and I set the cap where writing the whole amount off costs you nothing you need: for the clients I have run this for, between 3 and 8 percent of the paid search budget. If someone is selling this to you as a performance channel with a target cost per acquisition, ask them where the query-level report comes from. Where the money goes: Placement selection is the only real lever, so it gets the attention. Sponsored follow-up and related-question units sit beside an answer the assistant has already given, which means the visitor arrives mid-consideration with half their question resolved. A homepage is the wrong destination for that person and a pricing or comparison page is usually the right one. Assistant-side units inside Microsoft's stack can be bought through the same account as your Bing search line, which makes budget separation an accounting decision you have to take deliberately or the two channels will be read as one and the search line will absorb the blame. I write the destination map: which placement, which page, and which question that page answers in its first screen. How this gets measured, since the platform will not tell you: Three instruments, all outside the ad account. A geo holdout, two matched country or region sets, one exposed and one not, running a minimum of six weeks and read on branded query volume and direct sessions. A single post-purchase question in checkout or onboarding asking where the buyer first heard of you, with a sample floor I state before we start, because under it the read is decoration. And branded query volume in Google and Bing, which moves before revenue does and is the earliest honest signal available. Excluded, and this matters: I do not do content work aimed at getting cited organically inside assistant answers. That is a different discipline with a different lever set, I will not sell it alongside this, and I will not imply the ad buy influences it, because it does not. No llms.txt, no schema, no digital PR. No promise that the assistant will describe your product accurately: you are buying a placement, not editorial control. No monthly retainer for this line on its own, because there is not yet enough to decide each month to justify one. Say no if: Your product needs a demo, a contract and six weeks of legal review to buy. Assistant placements are poor at that shape of sale and an honest read takes longer than the budget survives. Say no if your paid search line is under roughly 40,000 euros a month, because the cap that keeps this safe becomes too small to produce a readable result. Say no if a negative outcome cannot be reported internally. A test you are not permitted to fail is not a test, and I would rather not be the person who writes the number that then gets buried. What you get: A written buy: placement, market, flight dates, cap and destination per placement. A measurement plan with the holdout definition and the sample floor agreed before anything runs. At the end, a document stating whether this deserves budget next quarter, with the evidence, in language your CFO can read without me in the room. Roughly half the tests I have run ended with a recommendation to stop. That is the outcome the fee buys, not the one it fails to produce.