What AI assistants cite: citations in ChatGPT, Claude, Gemini and Perplexity answers
Tonecast, https://tonecast.io/research/what-ai-assistants-cite
Data: 7 September to 4 October 2026 (pilot on 7 September, weekly cycles on 13, 21, 27 September and 4 October).
Published 7 October 2026.

Licence: Creative Commons Attribution 4.0 International (CC BY 4.0), https://creativecommons.org/licenses/by/4.0/
Cite as: Tonecast (2026). What AI assistants cite: citations in ChatGPT, Claude, Gemini and Perplexity answers.
https://tonecast.io/research/what-ai-assistants-cite

WHAT THE DATA IS
1,730 answers (status OK) to 81 questions about 7 subjects (2 brands, 1 product, 1 public figure, 3 topics) in
en, fr, it, de, es, sampled through the official APIs with web search or grounding enabled:
OpenAI gpt-4o-mini (web search tool), Anthropic claude-haiku-4-5 (web search), Google gemini-2.5-flash
(Google Search grounding), Perplexity sonar. 19,713 citations: the links each API returned with the answer,
reduced to the registered domain. Counts only: no answer text, no question text, no URL. The subjects' own
websites are replaced by "[subject's own site: <type>]". domains.csv lists domains cited at least twice.

COLUMNS (common)
engine                       ChatGPT, Claude, Gemini or Perplexity
answers                      answers with status OK in the group
answers_with_citation        answers that carried at least one citation
citation_rate_pct            answers_with_citation / answers, in percent
citation_rate_low, _high     95% Wilson interval of citation_rate_pct
citations                    citations in the group
citations_per_answer_mean    citations / answers
citations_per_answer_low, _high   95% interval of the mean (normal approximation)
citations_per_answer_median  median citations per answer
citations_per_cited_answer_mean   mean over answers that carried at least one citation
distinct_domains             distinct registered domains cited in the group

FILES
by_engine.csv                    one row per engine
by_engine_language.csv           + language (ISO 639-1)
by_engine_campaign_type.csv      + campaign_type: brand, product, person (public figure), topic
by_engine_prompt_category.csv    + prompt_category: discovery, comparison, evaluation, pricing, support, news
by_engine_cycle.csv              + cycle: date of the weekly sampling cycle ("(pilot)" for 7 September)
source_types_by_engine.csv       engine, source_type, citations, citation_share_pct (share of the engine's
                                 citations), answers_citing, answers_citing_pct with 95% Wilson interval
                                 (_low, _high). source_type: news (news & blog), wikipedia, owned (subject's
                                 own site), review, reddit, youtube, forum (forum / Q&A), editorial
                                 (high-citation editorial domain), web (any other page)
first_citation_type_by_engine.csv engine, source_type, first_citations (answers whose first citation is of
                                 that type), pct
overlap.csv                      engines_citing_the_domain (1 to 4), domains (question-and-domain pairs),
                                 pct. A pair is one question and one domain cited for it by any engine.
domains.csv                      domain, source_type, citations (all engines), answers_citing, campaigns_citing
                                 (subjects, 1 to 7), citations_<Engine>, answers_citing_<Engine>_pct (share
                                 of that engine's answers that cited the domain)