AI Use Policy

What AI is used for

AI assists with:

• Research organisation — gathering and structuring source material before writing
• Drafting — producing first drafts that a human engineer then verifies, corrects, and rewrites
• Code and configuration examples — generating candidates that are then reviewed against primary documentation and, where we claim testing, executed
• Editing — tightening prose, catching inconsistencies, improving structure
• Formatting and structural consistency — keeping runbooks and lessons to their templates

What AI is not used for

• Publishing without review. No technical content reaches published with its review flags set on the basis of AI output alone. A human sets those flags, and setting them is an attestation.
• Claiming a procedure was tested. Only a human who ran it can mark something tested.
• Research findings. Numbers in Research come from measurements we took. AI does not produce data points.
• Mass generation. We do not generate content at volume, in any form. The entire publishing architecture is built to prevent it, and our launch target is fewer than sixty indexable URLs.

The workflow

Every technical resource moves through explicit states, recorded in the file itself:

draft  →  technical-review  →  editorial-review  →  approved  →  published

Alongside that state, two separate flags record who has signed off:

• technicalReviewed — an engineer has checked the commands, configuration, and claims
• editorialReviewed — an editor has checked the structure, accuracy of framing, and clarity

Both are false by default. A resource with either flag false cannot be indexed — the publishing gate refuses it — and shows a visible banner telling readers that review is incomplete. This is enforced in code rather than by policy alone.

Why we are strict about this

Two reasons, and only one of them is about search engines.

The first is that our subject matter is operational. If a command on this site is wrong, someone runs it against production infrastructure. The cost of an unreviewed technical error here is not an embarrassing correction, it is somebody’s outage.

The second is that scaled, unreviewed content is now abundant and nearly worthless. The only durable reason to read a technical site is that a person who knows the material stands behind what it says. That is a claim we would rather be able to make truthfully than make loudly.

Disclosure on individual pages

Rather than labelling every page with a generic AI disclosure, we show the specific thing that matters: who wrote it, who reviewed it, when it was verified, and whether the procedure was executed. That information is on every technical resource, near the top, and it tells you more than a blanket notice would.

Our tools

The workbench tools marked as available today do not use AI at all — they are deterministic parsers and rule engines that run in your browser. The tools that would need model inference are clearly marked as not built, with a list of what has to exist first, including redaction and evaluation.

When those ship, this page will be updated to describe exactly what is sent where, what is retained, and for how long.

Related

• Editorial policy — the full publishing standard
• How we test — what testing claims mean here