This reference page collects what is publicly known about AI citation behaviour. It deliberately contains no invented statistics: where research exists it is named, and where the evidence is anecdotal that is stated plainly. The fastest-changing parts of this picture are updated as models release.

What the academic literature shows

The term “generative engine optimisation” entered public usage with a 2023 academic paper from a University of Oxford research group (available on arXiv as 2311.09735), which demonstrated that content perturbation techniques could measurably increase how often AI engines mentioned a source. The paper’s significance is methodological: it showed citation behaviour responds to content changes, establishing the empirical basis of the discipline.

What practical testing consistently shows

Across independent practitioner testing reported publicly since 2024, a consistent pattern emerges: structured, answer-first content is retrieved and quoted more reliably than unstructured prose; engines differ materially in how often they display visible citations; and Google’s AI Overviews cite sources far more frequently than chat-style engines, which often answer without attribution. These are directional observations from public testing, not controlled benchmarks — treat them as hypotheses to verify in your own market.

What remains unknown

There is no public, independently verified benchmark of citation rates by industry or by platform. Vendors that publish impressive-sounding “industry averages” without methodology should be treated with suspicion. The truthful position is that your baseline must be measured in your own market with your own prompt portfolio — which is exactly what an AI visibility audit does.

How to build your own benchmark

  1. Fix a prompt portfolio of 20-50 queries.
  2. Define what counts as a citation for each engine (visible link, named source, both).
  3. Run a baseline, then measure monthly under identical conditions.
  4. Record model versions — answers change when models change.

This is the only benchmark that matters for your decisions, and GEO reporting operationalises it. The named research and the directional patterns above provide context for interpreting your own numbers.