A new paper proposes a way for websites to get quoted by AI search tools without ever knowing what someone actually searched for.
Traditional search engines rank links, but a newer wave known as Generative Search Engines uses large language models to write direct answers and cite sources within them, shifting the goal for publishers from ranking high to being the page an AI decides to quote - a practice researchers call Generative Engine Optimization, or GEO. Most GEO methods align content to specific search queries, guessed or observed, but the new method, called Query Implied Generative Engine Optimization (QI-GEO), skips that step by scanning a document to estimate the range of questions it could plausibly answer and flagging relevance gaps. Tested on two benchmark suites, GEO-Bench and an extended version, the authors report gains of up to 15.9% on objective scoring metrics and 17.6% on subjective ones, with roughly twice as many citation gains as losses.
That matters because AI-generated answers are increasingly standing between publishers and readers, and most of what a page's actual audience searched for is invisible to the page's owner. A method that infers likely questions from the content itself, rather than from query logs a publisher may not have, could become a more accessible on-ramp to this kind of optimization than existing tools.
Still, these are self-reported numbers from a single paper's own benchmarks - worth watching for independent replication before anyone rewrites their content strategy around it.