Pay-per-inference
Pay-per-inference is the idea of paying a website each time its content is used to write an AI answer, instead of each time it's crawled.
Pay-per-inference is the idea that an AI company should pay a website whenever an answer draws on that site's pages. The crawl itself would be free. Inference is the moment a model produces an answer. So under this idea, the payment follows the use.
It's an idea more than a working system. There's no common standard for it yet. Marketers should still know the term, because it's part of the wider argument about who pays for the pages AI answers are built from.
How it differs from pay-per-crawl
Pay-per-crawl charges at the door. A bot reads your page once and pays once, whether the page is then used in ten thousand answers or none.
Pay-per-inference charges for what happens after. A page read once but used in many answers would earn many times. A page read but never used would earn nothing.
For example, say a coffee review site's guide to grinders is read by a crawler once in March. Under pay-per-crawl, the site is paid for that one read. Under pay-per-inference, it's paid each time an answer about grinders draws on the guide, which might be hundreds of times over the following months.
Why it's hard
To pay per use, someone has to know when a page was used. That's easy when an answer cites a page. It's much harder when the content shaped an answer without a link, or came from what the model learned in training. Every party would also have to agree on how much one use is worth, and trust the count.
That's why most of the working systems today charge at the crawl.
Why it matters to you
If you're a brand, you mostly want to be read and named, so you're unlikely to charge either way. It still matters, because the publishers, review sites and directories that name you may start charging. If they do, some platforms may read them less, and the sources behind your answers could change.