How We Use AI

Last updated April 21, 2026

This page is the working policy on AI tool use at Calorie Tracker Lab. We publish it because the calorie tracking app category is one we cover, including AI-photo-recognition apps, and it would be a strange contradiction to leave our own AI use undocumented while testing other people's. The short version: we use AI tools as research and editing assistants, not as authors. Every page on this site is written and signed off by a named human contributor.

What we use AI for

Three categories of task. We use AI tools (primarily Anthropic Claude and OpenAI ChatGPT) for the following, and we use them only for these:

1. Research summarization. When a contributor needs to read a 60-page validation study, an AI tool may be used to produce an initial summary that the contributor then reads against the original PDF. The AI summary is a navigation aid; it is never the cited source. Every claim in a published article traces to a primary source the contributor has read directly.
2. Citation finding. When a contributor knows a claim needs supporting literature but does not have a citation in hand, an AI tool may be asked to surface candidate citations. Every candidate citation is then independently verified by the contributor — we open the paper, read the methods, confirm the claim is supported, and confirm the publication metadata. AI-suggested citations that cannot be independently verified are dropped. We have caught hallucinated citations from AI tools and we treat the catch rate as a quality metric on contributors.
3. Copy editing. An AI tool may be used to suggest line-level prose improvements (clarity, sentence flow, tightening) on a draft the contributor has already written. The contributor accepts or rejects each suggestion individually. AI tools are not used to expand outlines into prose, generate new paragraphs of content, or rewrite drafts in someone else's voice.

What we do not use AI for

The following uses are prohibited under our editorial policy:

• Primary writing. No paragraph on Calorie Tracker Lab was generated by an AI tool. Every word published here was written by a named human contributor.
• Score generation. No app score, MAPE value, or ranking position on this site was generated by an AI tool. Scores trace to primary testing run by Vincent and Riley; statistics are computed by Yuki on raw data; rankings reflect the published rubric, not an LLM judgment.
• Test-protocol substitution. We do not ask an AI tool to estimate the calorie content of a meal in lieu of weighing it. We do not ask an AI tool to "predict what an app would log" in lieu of installing the app and logging the meal.
• Synthetic citations. We never publish a citation we have not independently verified, regardless of where the citation came from.
• Contributor impersonation. We do not ghostwrite a published byline using AI under the contributor's name. The byline reflects who wrote the piece.

Why we draw the line where we do

Three reasons. First, AI tools hallucinate citations and facts at non-trivial rates. The cost of publishing a hallucinated citation in a piece that touches clinical claims (protein targets, GLP-1 nutrition, body-recomposition framing) is high; the benefit of saving 20 minutes of contributor time is low. Second, the editorial value Calorie Tracker Lab is trying to deliver is, fundamentally, the judgment of named credentialed humans who have done the testing themselves. AI-generated prose erodes that proposition even where the prose is technically correct. Third, the long-term reputation of the publication depends on readers' trust that a byline corresponds to a real person who read the same studies they're citing; we are not interested in trading that reputation for short-term throughput.

Disclosure on individual pieces

Where an AI tool was used in any of the three permitted ways for a particular piece, we do not currently disclose the per-piece use, because the use is so general (every contributor uses these tools as research aids) that a per-piece disclosure would be uninformative. If we ever change this policy — for instance, if we publish AI-generated draft content, even with heavy editing — we will disclose the change here and we will mark affected pieces individually.

This policy is dated

Last updated April 21, 2026. AI tools change quickly; we expect to revisit this page at least quarterly and to versionnotable changes in our update log.