What was measured
Citation presence, citation count, and publisher source type within 77 collected answers generated from 11 prompts across 7 engine surfaces.
Skayle Research · 2026
What AI answers cite: a 77-answer benchmark.
Skayle analyzed citation behavior across 11 prompts and 7 AI engine surfaces. The benchmark separates whether an answer contained citations from the kinds of sources selected.
77
sampled answers
11
prompts
7
engine surfaces
75%
answers with citations
Primary finding
Citations appeared in three out of four sampled answers.
Across the 77 evaluated answers, 75% contained at least one citation. That percentage describes citation coverage for this benchmark set.
The dataset recorded an average of 8.68 citations per sampled answer. This describes the observed sample; it is not an expected rate for every prompt, market, engine, or collection period.
The observed source mix.
Corporate and editorial sources made up the largest classified shares. Source type describes who published a cited source; it does not establish that every source was equally authoritative or influential.
Corporate sources41%
Editorial sources26%
User-generated sources15%
Competitor-owned sources12%
The four displayed classifications total 94 percentage points. Smaller review, reference, forum, and unclassified categories are omitted here; independently rounded shares may not total exactly 100%.
How to read this benchmark.
What it can show
How citation behavior and source selection appeared in this defined sample, including the relative mix of classified source types.
What it cannot show
A universal citation rate, complete engine coverage, causal ranking factors, or guaranteed future inclusion for any brand or source.
Dataset and reproducibility note
This page publishes aggregate results, evaluated-set dimensions, classification definitions, and limitations. It does not publish response text that may contain third-party or user-specific material. Reproduction requires the same prompt wording, engine surfaces, collection window, and evaluator definitions, and even then generative responses may vary.
Rolling 90-day sample generated August 11, 2026 at 12:48 PM UTC · Published August 11, 2026 · Analysis by Skayle · Author: Edin Abazi