Ads in chat: what to ask before you buy

Ad placements inside AI assistants are now buyable. Access runs through DSP partners, the targeting is contextual rather than demographic, the placements read like a text ad beside an answer, and first-party audiences — the lists already sitting in Shopify, HubSpot or Klaviyo — can now be connected to them.

That is genuinely new inventory. It is also a channel where nobody has a track record, including the people selling it.

Be honest about what isn't known

Pricing guidance exists, but it arrives without the context that makes a benchmark a benchmark: which account, which category, which creative, optimized toward what. A cost quoted without the account it came from is an anecdote. And in a channel this young, figures from a few months ago are history rather than a forecast.

So the honest framing for a first campaign here is a test, with a budget you can afford to lose, run to learn one specific thing — not a line item that takes money from the channels currently paying the bills.

Questions to settle before the budget moves

1. What is this test meant to teach us? Write down the sentence you want to be able to say in eight weeks. If it is "we tried it," don't run it.
2. What is the all-in cost? Media, platform, fees, and the creative time to make something native to the placement rather than a resized banner.
3. What is the smallest spend that produces a readable answer? Below that number you are buying noise and calling it a learning.
4. Who pays for the learning? Client-funded, agency-funded or shared. Decide in writing, before it starts.
5. What does the placement look like in context? See it next to a real answer. Then ask what happens when the assistant describes your category, or your brand, in words you did not choose.
6. Which audiences are in, and which are out? Bringing your own data is the most interesting part of this. Suppressing recent purchasers is often worth more than reaching new people.
7. How will we attribute it? People do not behave in a conversation the way they behave on a search results page. Agree the measurement before the spend, not after.
8. When do we stop? A date and a threshold. Tests without an end date quietly become budget lines.

Where it should sit

At the bottom of the stack, funded as an experiment, while the channels that work keep working. A brand can be early here at a scale that does not hurt if it fails. That is the whole point of being early.