How we research, write, and correct our work

Every guide on brandcited.ai is grounded in real scan data, primary sources, and direct testing. We run scans against our own platform daily and use those logs as the ground truth for claims about how AI engines behave.

Primary-source preference: when we claim that Google's Gemini cites a brand in a certain way, we link to Google Search Central documentation, not to a third-party blog summarising it. When we claim that Anthropic added Claude-User as a separate bot, we link to Anthropic's official bot documentation. We avoid citing secondary sources unless they add unique original research.

Inline citations: factual claims link to the primary source next to the claim, not at the end of the post. Readers can verify any specific assertion without scrolling to a references section.

Acceptable sources: vendor documentation (Google, OpenAI, Anthropic, Microsoft, Perplexity, Apple), peer-reviewed research, .edu / .gov publications, recognised industry research organisations (DataReportal, Pew, McKinsey), and our own scan data labelled clearly.

We do not cite: AI-generated summaries of other sources, content farms, marketing blogs without verifiable authors, or screenshots without original URLs.

BrandCited stays anonymous as a team until the platform reaches significant revenue. The canonical author byline is "BrandCited Editorial Team" with the role-only Person schema published at /authors/editorial.

Technical posts (schema, robots.txt, methodology deep-dives) carry an additional reviewer — the engineer who shipped the underlying feature is named in the post's "Reviewed by" line.

When the team's anonymity preference changes, real individual bylines and verifiable identity surfaces (LinkedIn, Wikidata Q-IDs) will appear on every post.

BrandCited uses AI assistance in parts of the content pipeline. Where AI helps with drafting, every published post is reviewed and edited by a human team member before publication. The reviewer is named at the bottom of any post that used AI drafting.

We do not publish raw AI output. We do not publish posts that lack a specific data point, a named example, or a first-person observation grounded in our scan data.

Posts that are fully AI-written without substantive human editing are not published. The /ai-policy page describes our broader stance on AI use across the product.

Spot a factual error or stale claim? Email editorial@brandcited.ai and we will investigate the same week. Corrections are versioned: the post's dateModified is bumped, an editor's note is added at the top describing what changed and when, and the original claim is preserved in the post history.

We do not silently rewrite posts. If a methodology change makes a previous post's recommendation wrong, we publish a new post linking back and explain what we got wrong.

Every claim about how AI engines work that depends on our own measurement is traceable to a methodology version. The full BrandCited methodology, its weights, its check list, and its changelog are public at /methodology.

When a methodology change shifts our recommendations, we publish a methodology changelog entry and link from the affected posts back to the entry. Our scoring model is not a black box.