Where AI is and isn't involved
AI assistance is used to draft content from federal data we have already loaded — never to invent statistics, simulate clinician opinions, or paraphrase third-party copy.
Allowed AI tasks
- Initial draft for blog articles in insurance / cost / care clusters.
- Programmatic state and state×specialty page intro paragraphs.
- FAQ generation grounded in actual search query patterns.
- Schema markup field extraction from structured provider data.
- Slug, summary, and meta description generation for provider records.
Forbidden AI tasks
- Generating medical advice, treatment recommendations, or clinical claims of any kind.
- Inventing or extrapolating statistics not present in the source input.
- Paraphrasing competitor finder-site copy. (We don't scrape or feed it to a model.)
- Auto-publishing without a human review pass.
Required gates before publishing
- Source-grounded. The prompt includes the raw government data the article must reference. The model is instructed not to introduce statistics that aren't in the input.
- Human edit. An editor reads and revises every AI draft. For YMYL articles (medical / dental / insurance), a clinical reviewer with relevant credentials (DDS, RDH, MPH) signs off before publication.
- Frontmatter disclosure. Every article carries
aiAssisted: true|false, the named author and reviewer, last-reviewed date, and a sources array. - Public disclosure. This page is linked from every article's footer.
Models we use
- Longform drafting — Claude Sonnet 4.6 / Opus 4.7.
- Short transforms (slug, summary, schema fill) — Claude Haiku 4.5.
Why we disclose this
Google's position on AI-assisted content is that it's acceptable when it's helpful and transparent. Disclosure is also the right thing to do for a YMYL site — readers deserve to know how their information was assembled.
Reporting concerns
If you spot an AI-drafted article that looks wrong or unsourced, email contact@dentalnpi.com and we'll review it. See our editorial policy for the corrections process.