How We Use AI

The short version: we use AI tools as research and editing assistants, not as authors. Every page is written or signed off by a human contributor. No score, ranking, or factual claim on this site originates from AI without a human verifying it against test data or a primary source.

Tools we use

• Anthropic Claude (research summarization, drafting, copy editing)
• OpenAI ChatGPT (research summarization, draft revisions)
• Standard code-completion tools when writing internal scoring code

What we permit AI to do

• Summarize a body of source material we have already gathered.
• Suggest reorganization of a draft a human wrote.
• Copy-edit prose for clarity and grammar.
• Find candidate citations for a claim, which a human then verifies and links.
• Generate test scaffolding or boilerplate code that a human then reviews.

What we prohibit AI from doing

• Writing reviews end-to-end without human authorship.
• Generating scores or rankings.
• Producing factual claims that aren't verified against a primary source.
• Generating fabricated authors, credentials, or citations.
• Drafting medical or nutrition guidance without human RDN review (once contributors are onboarded).

Why we disclose this

AI tools hallucinate citations and facts at non-trivial rates. The editorial value of Tracker Benchmark is the judgment of humans who have done the testing — disclosing where AI is and isn't in the pipeline lets readers calibrate trust appropriately.