Methodology

Every quote on this site passes through four steps. The labels on each page are the output of those steps — they are not decoration.

1. Collection

We read a fixed list of technology and business publications, plus search results for each tracked person. Primary sources are preferred: earnings-call coverage, on-record interviews, company announcements, conference keynotes. Syndicated copies are used only when no original report is reachable, and the outlet is always named.

2. Extraction

A language model reads the article and identifies what the tracked person actually said, ignoring the reporter's framing and any claims about other people. Statements are kept only if they concern technology, products, markets or company strategy.

3. Verification

Direct quotes must appear verbatim in the source text — the sentence is matched character-by-character against the article. If it does not match, the item is downgraded to a Reported paraphrase, or dropped. This is why the two labels exist and why they are never merged: on this site, “Quote” means the words are the speaker's, “Reported” means the outlet's summary of their position.

We also drop statements that are not from the news event itself (old interviews being recalled, evergreen advice in profile pieces), anything older than roughly two months, political and partisan remarks, and pure product announcements.

4. Attribution

Each entry stores the outlet, the article headline, the publication date and a link. The link is the point: if a reader cannot verify a sentence, the entry has failed.

Corrections

If a statement is misattributed, mistranslated, stripped of context, or dated wrongly, email [email protected] with the page URL. We fix or unpublish within two working days and note the change on the page. We would rather delete a quote than defend a wrong one.