Kumar Abhinav

Who Gets the Credit When an AI Cites Your Paper?

Abstract:

When you cite a paper, the author gets credit. A citation count goes up, a CV gets stronger, a career moves forward. When an AI assistant cites that same paper to answer someone question, nothing happens. No count, no record, no acknowledgement. Now multiply that by the number of AI answers generated every day.

The licensing side has started to catch up. Publishers can meter and charge for crawling, and a new licensing standard even defines payment per AI answer. Some AI companies have begun signing and watermarking what their models produce. But one piece is still missing, and everything else depends on it. A publisher can see that its page was fetched. It has no way to check whether that page was actually used in an answer.

So payment terms written per answer cannot be enforced. "Cited by AI" cannot be counted. And a company reporting its own citation numbers could quietly under-report them, hide them from one publisher, or inflate them for a partner, with nobody in a position to tell.

This talk lays the problem out in plain terms: what has already been built, what is still missing, and what any citation count would have to satisfy before the people it pays would believe it. You will leave with a clear picture of an open problem sitting between AI systems, publishing, and audit.

Profile:

Kumar Abhinav builds agentic AI systems at AiDash, where he works on retrieval and generation for assistants used by electric utilities and other infrastructure operators. His recent work covers embedded retrieval for agents, risk signals mined from engineering tickets, and verifiable provenance for AI-generated answers. He has released his retrieval research as open source with an archived DOI, filed a provisional patent application on agent retrieval, and has papers under review at IEEE venues. One question drives most of his work: how do you check what an AI system actually did, instead of taking its word for it?