How SplitAtlas uses AI
We use AI heavily and we're not coy about it. This page explains exactly where AI helps, where it's forbidden, and why you can trust the result.
Where AI helps
- Research assembly: collecting manufacturer documents, extracting specifications into our database, and flagging conflicts between sources.
- Drafting: turning verified data and structured briefs into readable pages.
- Consistency checks: catching unsupported numbers, missing disclosures, contradictory recommendations, and stale dates before publication.
Where AI is forbidden
Our production system enforces these rules; they aren't aspirations:
- AI never invents a specification, price, review, test result, or credential. Facts enter pages only from database records with source attribution.
- AI never claims hands-on testing that didn't happen.
- AI never states a rebate or tax credit is available without a verified, dated program record.
- AI never writes electrical or refrigerant instructions beyond what our safety-reviewed guidelines allow.
- No page is published just to exist. Every page must answer a real question usefully, or it stays unpublished - our owner's standing rule is blunt: no thin content, ever.
The human layer
Safety-relevant content (electrical guidance, sizing claims), product recommendations, and incentive claims require human review before they can be indexed by search engines. Pages display their last-reviewed date. When we add credited expert reviewers, their names will appear only on pages they actually reviewed - we don't invent experts, and author photos on this site will never be stock images of people who don't exist.
Why this approach
The failure mode of AI content is confident nonsense at scale. Our answer is structural: the database is the only door facts can enter through, every fact carries its source, and the things that can hurt someone - wiring, gas, refrigerant, warranties - get human eyes before they get an audience.
Questions or concerns: hello@splitatlas.com. Errors: /corrections.