The AI Answer Corpus

Name: The AI Answer Corpus
Published: 2026-10-25

Every month we send ChatGPT and Perplexity the same 331 buyer questions, 12 times each, through their APIs with web search on, and keep every answer and every source. This is the public record of what that archive shows. The figures are pooled across 12 company sites in 7 industries and name no client.

The short answer

ChatGPT and Perplexity disagree about what to cite, and ChatGPT disagrees with itself. Twelve identical ChatGPT runs draw on about four times the hosts one answer shows; Perplexity's twelve draw on nearly the same list. Four in five cited pages carry no date at all. And one ChatGPT check finds a brand only 45.9% of the time that twelve checks would.

12,408answers

331prompts

12repeats per prompt per month

7industries

122,018citation rows

6,158distinct hosts cited

2monthly fills so far

The studies

4.0 against 1.12Where the sources come fromDistinct hosts across 12 runs divided by hosts in one answer, ChatGPT against Perplexity. Perplexity cites LinkedIn in 16.0% of answers, ChatGPT in 0.7%; ChatGPT cites Wikipedia in 12.6%, Perplexity in 0.4%.4 in 5How old the cited pages areFour in five cited pages carry no date in the URL or title. Where one exists it is the current year 62.6% of the time on ChatGPT and 81.1% on Perplexity. The old pages are files.45.9%One check against twelveWhere 12 ChatGPT runs ever name a brand, one run names it 45.9% of the time; on Perplexity, 87.1%. All 12 ChatGPT runs agree in 4.7% of cases.6,204 × 2How the corpus is builtThe instrument, the sample, the three citation bases, the statistics, and what is withheld.

How to cite it

Link the study page that carries the figure. Each page states its own as-of date and the fills it covers, and each figure names its basis: ChatGPT counts every URL it returns; Perplexity figures use its first 10 sources unless the text says all 20.
Quote the interval with the rate where one is given. The brackets are 95% intervals clustered by prompt, so twelve repeats of one prompt do not count as twelve prompts.
Do not infer a client. Every figure is pooled across the 12 sites. Per-site and per-industry figures exist in the internal record and are not published.
There is no download. The raw archive holds the prompts clients pay to track, so the page is the citation. The method page describes the tables well enough to rebuild the instrument on your own prompts.

What it cannot tell you

The engines are the API models, gpt-5-mini and sonar, not the consumer apps, and the calls carry no location, account or chat history. Rates are directional visibility, not what any one person sees.
Prompts were written per tracked client, not sampled from search demand. 230 of the 331 ask for a recommendation; a 66-prompt energy panel was written for no tracked brand.
Two fills so far (12 to 26 September 2026 and 1 to 5 October 2026). Nothing here is a trend.
Perplexity's API returned 20 sources in September and 10 in October, so Perplexity comparisons across the two months use the first 10 of each.
A citation rate assumes the engine's crawlers could read the site. Part of the sample sat behind a crawler block during both fills; the affected findings were held back.

What is next

The November fill adds a 15-prompt research panel chosen for the questions these pages raised, not for any client. The pages refresh after the November 2026 fill closes, about 15 November. If there is a question you want the corpus to answer, suggest a prompt for that panel.

Quoting a figure? Link to this page as the source: https://www.bestseopodcast.com/research/ai-answer-corpus. Figures are dated 2026-10-25 and cover answers collected 12 to 26 September 2026 and 1 to 5 October 2026; the next refresh is after the November 2026 fill closes, about 15 November. The page is the citation: there is no download.

Who did this

Matthew Bertram

Author · Host of The Best SEO Podcast since 2009 · Owner and CEO, EWR Digital

Matthew Bertram runs EWR Digital, the SEO and AI-visibility agency whose monthly tracker produced this corpus, and he is the author of LLM Visibility: A Decision-Grade System for Winning AI-Mediated Discovery. The studies were built, re-derived from counts and reviewed before anything was written up; every figure on this page traces to a script in the project record. Questions about the method go to him through the contact page. More about Matthew Bertram.

Want these numbers for your own brand?

Find out which engines name you, which cite you, and what to fix first.

Get a free AI visibility audit