Bought with context, not audiences.

A placement next to an answer, matched to what the person is describing. You aren't picking an audience and you aren't bidding on a keyword — the fit is decided by what your product is understood to be. That is why the setup work reads more like documentation than like an ad: you're making sure the description of your product that exists in the world is the one that fits the moment you want.

The inventory is small and always will be, so it will lose any impression-volume comparison against Meta or TikTok. What it returns is placement at the moment of decision and unusually clean intent data — you learn the words your buyers actually use. Judge it on the cost of an acquisition and on what it teaches you, not on reach.

Because its failures are silent. A purchase value sent as 49.99 rather than 4999 is not rounded — the event is rejected outright, so the sale never reaches your reporting. An event called purchase — borrowed from Meta — isn't in OpenAI's taxonomy and is dropped on arrival. One undocumented key inside an item makes the SDK discard the whole event rather than the field. A plaintext email where a SHA-256 hash belongs isn't rejected, it's forwarded raw. Nothing errors. You would simply read the numbers and believe them.

The SDK. What decides whether an event is kept is the event table and field whitelist inside the shipped oaiq.min.js, not the documentation — and they diverge in both directions. The docs list app_installed as supported when the table has no such entry; the docs say hashes must be lowercase when the regex accepts uppercase and folds it. Following the docs alone means reporting working installs as broken and broken ones as fine.

It's the same work, so yes — but not as a separate retainer with a content calendar attached. What earns a paid placement is how your product is described where a model can read it: your store listing and its reviews, comparison pages, third-party write-ups, the forums your buyers already use. That is also what decides the unpaid answer. We don't sell blog posts on your own domain; models discount obvious vendor material, and if you're already being read, more of it changes nothing.

Almost not at all. TikTok and Instagram are walled off from the crawlable web — no reliable public transcripts, no indexable text. Answer engines build their picture of a category from text they can read. A video can be the best money you spend and still leave no trace here.