Editorial methodology

A reference work is only as good as its accuracy. Every GEO Glossary entry runs through a multi-pass editorial workflow before publish, with explicit quality control on factual claims.

The 4-step workflow

1. Drafting. Initial draft prepared with LLM assistance. The draft is treated as a starting point, not a finished entry; every claim is rewritten or removed by later passes.
2. Fact-audit. A separate research agent inspects every numerical claim, date, vendor capability assertion, and primary-source citation. Anything that cannot be verified against a public source is softened (a specific percentage becomes "modest", a confident vendor capability becomes a hedge) or removed entirely.
3. Independent peer review. Substantive entries (cluster anchors, strong-empirical-claim entries, contested-topic entries) go through one or more rounds of independent LLM peer review after the fact-audit. By "peer review" we specifically mean independent LLM/web review (paste the entry into a fresh ChatGPT or Claude session, ask for fact-checking and conceptual critique) plus source verification, not academic-journal peer review. Edits surfaced by peer review are applied with the same discipline: claims that cannot be backed by a primary source get softened or removed. For practitioner-coined anchor entries in empty or emerging territory, we run three review layers (agent fact-audit, ChatGPT web peer review, Claude web peer review) because each layer catches different error classes: factual / URL / date errors, conceptual coherence errors, and framework adoption issues respectively.
4. Schema and structure check. DefinedTerm JSON-LD, FAQ schema (where applicable), Breadcrumb schema, and Open Graph metadata are validated before deploy. Schema is treated as machine-readability hygiene, not as a citation lever; the editorial work above is what actually drives citation accuracy.

Fact-check cadence

Every term page shows a Last fact-checked date in its editorial footer. The cadence:

• Every new term batch runs through an independent research-agent audit before commit, with the agent's old to new edit pairs applied verbatim.
• Weekly source-URL audit via scripts/audit-sources.mjs. Dead links are replaced or removed, not silently left to rot.
• Status freshness sweeps. The "Status in 2026" sections are revisited monthly against current vendor announcements; every substantive edit bumps the entry's lastFactChecked stamp.

The fact-check passes are not optional or batched at end-of-month. Each new entry triggers them as part of the publish workflow, and the workflow itself has been refined across the corpus as repeat failure modes (date-year off-by-1 errors, fabricated footnote methodology, body-vs-footnote drift) get added as specific pre-commit gates.

What the workflow does not guarantee

We do not promise zero errors. We promise: every error gets a public correction with a dated changelog entry on the affected term, the workflow gate that should have caught it gets tightened, and the failure pattern is published as a memory rule that future writes have to honor. Editorial discipline is a moving baseline, not a fixed state.

How this connects to other parts of the site

• The citation-tracking page describes the workflow for moving citation status badges from untested to cited or not-cited.
• The cluster-discipline page describes the consistency rules that apply when an entry belongs to a cluster of 5-10 related concepts.
• The why-this-exists page describes the brand positioning that the editorial discipline backs up.