AI Disclosure

Sphere uses AI tools throughout its writing, research, and design work. We think being explicit about which parts were produced which way matters — both because Sphere’s pitch leans on epistemic credibility, and because readers deserve to know what they’re reading.

The four-tier scheme

Every page on this site (and on the Sphere docs site) carries a label at the bottom that falls into one of four tiers:

✍️ Authored

Written by a named human from scratch. AI may have been used as a spell-checker, search tool, or thought partner — but the substantive prose is human work.

🧑‍🔧 AI-assisted

A substantive collaboration between a human and an AI model. Either party may have started the draft — the human may have written prose that AI helped revise, restructure, or fact-check, or AI may have drafted prose that a human substantially rewrote. The defining characteristic is that both shaped the final text materially.

⚙️ AI-drafted

A draft from an AI model that a human reviewed and approved without substantial rewrites. Acceptable for low-risk content (summaries, glossary entries) but not for foundational claims.

🤖 AI-generated

Produced by an AI model with no substantive human review yet. We use this tier for automated research outputs — surveys of related work, literature reviews, comparison studies. Treat these as raw material: useful, traceable, but not endorsed claims.

Per-section overrides

A page may carry one default label but include a section that’s labeled differently. When that happens, you’ll see a smaller inline note at the top of that section.

What’s recorded

For every labeled page, where applicable, the footer also records:

Model — which AI model produced the draft (e.g. claude-opus-4-7)
Drafted — when the AI draft was produced
Edited by — the human who edited it
Edited — when the human edit happened

This data is also embedded as <meta> tags and schema.org/CreativeWork JSON-LD, so any reader, archive, or aggregator can extract it without having to scrape the visible footer.

What this scheme does not claim

It does not measure quality. An authored page can still be wrong; a generated page can still be useful.
It does not track edit-by-edit provenance. If you want commit-level history, the underlying repository is the source of truth.
It does not distinguish between AI tools used as research aids (search, summarization, suggestion) during human authoring versus AI models used as primary drafters. Tier authored covers both — the distinction is whether the prose came out of the model or out of the human.

Why we bother

Sphere is a project about making collective epistemics legible. Hiding the provenance of our own writing while we propose tools to make everyone else’s writing traceable would be incoherent. So we label.