The “default shortlist” problem: how AI tool recommendations concentrate visibility around incumbents

By outranker · Published 2026-11-17 · Version 1

Sample: 55 AI answers to 20 questions · Engines: ChatGPT, Claude, Gemini · Collected 2026-11-17 – 2026-11-17

How the study was run Ask recommendation-style buyer questions to every available engine and record which brands are named and in what order. Mentions are detected by whole-word match; answers are stored verbatim. ## Scope - Questions: 20 - Engines: ChatGPT, Gemini, Claude - Verified answers: 55 - Repeats per question: 1; waves: 1 - Collection dates: — ## Limitations - AI answers vary between runs; each question is asked once per engine. - Plain-chat engines (ChatGPT, Claude, Gemini) report mentions only, not source links. - No source-returning engine is available, so no citation metrics are planned. - Results are descriptive observations of AI answers, not proof of cause.

Limitations

  • Collection finished partially; missing answers were accepted by the owner.
  • ChatGPT, Gemini, Claude answer from model knowledge without live sources; a mention there is not a citation.
  • Small sample (3 answers); treat as directional only.
  • Mentions are not endorsements.
  • No source links were provided in the dataset; drivers of selection cannot be inferred.
  • Collection was partially finished; smaller clusters have limited sample sizes.