AI in our content.

We use AI to draft some of what you read on stonesai.com. Every AI-drafted piece is reviewed by a named human, who can approve as drafted, edit, or reject before it ships. We disclose which is which. This page tells you what gets AI-drafted, where the disclosures live, and what we ask of human review — it's how we apply our Show your work principle to our own content.

What's AI-drafted, and what isn't.

The /knowledge essays are AI-drafted. They're written by Claude (Anthropic) from a topic and an outline, then reviewed by a human reviewer who can approve, edit, or reject before publication. That's the only series on the site that uses AI in this way.

Everything else is human-written from scratch. The homepage, the principle pages, the product page, the about pages, press, investor materials — drafted, edited, and shipped by people. We sometimes ask Claude to suggest variants on a paragraph or test the rhythm of a sentence, the way another person might. The page that ships is human-authored.

This page is human-written.

Where we disclose.

For every AI-drafted post, the disclosure shows up in three places. All three or it doesn't ship.

First, in the source. Every AI-drafted post's frontmatter sets ai_drafted: true and names the human reviewer. This is invisible to readers but visible to anyone who opens the file or scrapes the source.

Second, on the page. Every AI-drafted post carries a disclosure block at the bottom: a short paragraph naming Claude as the drafter and the human reviewer who approved the post, plus a link back to this page. You read it the same way you read the rest of the post.

Third, in the structured data. The Article schema for every AI-drafted post lists creator as Claude (Anthropic) and editor as the human reviewer. Search engines and LLMs that ingest the post see the same disclosure that you do — not a sanitized version.

The pattern is intentional. Disclosure that only appears in the source is hidden disclosure. Disclosure that only appears in the schema is invisible to readers. Disclosure that only appears in a footer is invisible to machines. We use all three so the disclosure travels with the content wherever the content goes.

What human review catches.

Claude is good at fluency and structure. It's less reliable at three things, and those three are where review happens.

Claims we wouldn't stand behind. AI drafts can be confidently wrong about facts, dates, attributions, and our own products. Every claim in an AI-drafted post is checked against a real source by a human before the post ships.

Voice that drifts. AI drafts trend toward a generic register — adjective stacking, hedge words, marketing-speak. Our voice rules (banned-word grep, parallel-structure rule, periods over slogans, no exclamation points) get run on every draft. Where the draft slips, the human rewrites.

Things we won't say. Some claims pass fluency and pass voice but still aren't claims we'd put under our name. The reviewer's job is to catch those before they ship.

If a draft can't pass review, it doesn't ship. That happens. The cadence is two posts a month — when a draft fails review, the slot stays empty rather than ship a half-cooked post. Stale beats ungrounded.

How to flag a mistake.

If you read an AI-drafted post — or any post on this site — and spot something that's wrong, write us. We'll fix it, log the correction visibly on the post, and credit the catch.

Email transparency@stonesai.com. We read every email.

Trust is the product, and trust survives a paper trail. This page is the paper trail.