University of Toronto and Google paper: GEO evaluation needs unified editing rules
A paper from the University of Toronto and Google argues that GEO (generative engine optimization) benchmarks need unified rules on permitted webpage edits, otherwise scores cannot be compared. The paper labels each type of GEO edit by where the change is made, how it is made, and the cue it targets (such as authority), and recommends that tests report visibility, answer quality, and rule compliance together. The paper conducted no experiments; its evidence is a review of 11 published GEO methods, showing that the types of edits covered by each method differ.
Body
New University of Toronto and Google paper argues that picking an AI assistant may be the last big decision people make, since that assistant then makes the decisions that follow.
Argues that tests of how websites game AI answers need shared editing rules, or scores can't be compared.
Content owners now edit web pages so AI assistants cite or recommend them, a practice called generative engine optimization, or GEO. Current GEO benchmarks allow different edits, so a method that rewrites whole pages can look better than one that only appends a line.
They propose labeling every GEO edit by where the page changes, how it changes, and which cue it targets, such as authority. Tests would fix the allowed edits and report visibility, answer quality, and rule-following together.
The paper runs no experiments. Its evidence is a mapping of 11 published GEO methods, showing that each covers a different slice of the possible edits.
When comparing GEO methods or AI systems, check which edits each score allowed and whether answer quality was measured.