Research and direction
Every cluster of pages starts with a reader need and a review of what already exists: current search results for the queries involved, primary sources where the topic has them, and the Learn lessons already published here so that new pages extend the site instead of repeating it. The angle is decided by a person before any drafting begins. Where a page cites a source, someone opened the source.
Drafting with AI assistance
First drafts of lessons, prompts, workflows, examples, and guides are produced with AI assistance from a written brief: the reader, the job, the claim, the material, the voice, and the constraints. The brief forbids invented facts, sources, people, and statistics. Drafts are then reviewed for perspective, accuracy, originality against existing pages, and tone, and are edited by hand before publication. Example runs on prompt pages are representative responses, labeled as such. They are not presented as outputs of a named model version on a given date.
Prompt pages
Every indexable prompt page must be a complete resource: the job it does, who it is for and not for, every variable with an example, the full prompt, an explanation of why it is built that way, a representative run, follow-up prompts, failure modes with fixes, and the human review it still needs. Each page records its last review date and says that model behavior varies and changes. Prompts that cannot meet this standard are not published as pages.
Examples
The annotated examples show four layers: the writer's notes, the direction given, the assistant's output, and the human revision. The scenarios are constructed for teaching and the people in them are fictional. No example quotes a real private person or a real company's internal material. The AI-draft layers are written to be representative of what a capable assistant returns for that brief, including the generic tells, so that the margin notes have something true to point at.
Frameworks and labels
The site uses several frameworks of its own: the THREAD sequence for iterative prompting, the accepted-decisions ledger, the source note, the bounded change, and Direct, Propose, Judge, Own. They are labeled as this site's editorial frameworks wherever they appear and are not presented as external standards. Technical, legal, and institutional terms are distinguished from them in the glossary.
Verification and legal caution
Factual claims are checked against the sources they cite. Legal and policy statements, such as copyright guidance or disclosure rules, are kept jurisdiction-specific, dated, and narrow, and the pages that carry them say they are information rather than advice. Claims about specific tools are kept general, because features and prices change faster than pages do, and readers are told to check vendor documentation.
The Orwellix relationship
Vibe Writing Hub is published by Orwellix. Orwellix is recommended only where a task on the page could continue there, the recommendation is labeled, and the links carry campaign parameters so the publisher can see whether they are useful. The tools guide describes competing categories of tool on their merits and states the relationship at the top. Content data for lessons, prompts, workflows, examples, and guides contains no product promotion. The disclosed notes are added by the page templates.
Updates and corrections
Pages carry a published date and, when the substance changes, an updated date. Small typographic fixes do not refresh the date. Model-sensitive material such as prompt behavior and tool descriptions is reviewed more often than evergreen writing principles. Errors are handled as described on the corrections page.
Newsletter
The Vibe Writing letter is produced the same way as the site: an observation or technique chosen by a person, drafted with assistance, revised by hand, and labeled where AI did substantial work. Issues that recommend Orwellix say so in the issue.