Turn a de-identified food log into assessment prep
Before an assessment, dietitians read through food diaries and 24-hour recalls looking for patterns — meal timing, missing food groups, snacking habits. AI is genuinely good at structuring the qualitative story from text you paste in. What it must not do is the nutrient math or the clinical interpretation, both of which it gets wrong in ways that are unsafe for someone with a medical condition.
You are helping a dietitian prep for an assessment. Using ONLY the de-identified food log I paste below, summarize the eating patterns. Context: the client's general goal is {{goal_context}}; this covers {{timeframe}}. Food log (no names, no health details): {{food_log}} Produce a qualitative summary only: 1. Meal and snack timing and any long gaps. 2. Food groups that show up often and ones that are largely missing. 3. Patterns worth exploring in the session (e.g., low fiber sources, most eating late in the day). Hard rules: - Do NOT estimate calories, macros, or micronutrients unless I explicitly ask. If I ask, label every figure [ESTIMATE — RD TO VERIFY against food-composition data] and never present it as clinical fact. - Do not interpret this for any diagnosis, medication, or lab value. - If anything looks medically concerning (very low intake, possible disordered eating), write [FLAG FOR RD REVIEW] and stop rather than advising.
Fill in your details and the prompt updates live — then copy.
Pattern summary (qualitative only): Timing: Breakfast is skipped both days; a long morning gap is followed by heavy late-evening meals (9-10pm). Food groups: Refined carbs appear often (bread, pasta, chips, cereal). Vegetables and fruit are largely absent; protein is present but concentrated late in the day. Worth exploring in session: the skipped-breakfast to late-heavy-dinner cycle, low fiber and produce, and reliance on convenience foods midday. [FLAG FOR RD REVIEW] if the very low daytime intake reflects intentional restriction.
The full workflow
- Strip the log of names, dates of birth, conditions, and lab values before pasting anything
- Run the prompt to get the qualitative pattern read, not numbers
- Recalculate any nutrient figure you need in a validated tool — never trust the model's math
- Bring the summary into your own assessment; the clinical interpretation stays yours
Watch out for
Verify the math: AI nutrient estimates fluctuate wildly (up to ~45% swings in calorie estimates) and underestimate micronutrients. Its numbers are unsafe to act on for diabetes, CKD, or any condition needing precision — recompute everything.
Interpretation is medical nutrition therapy: reading a food log against a diagnosis or labs is an RD's job, not the chatbot's. Use AI to organize patterns, not to make the clinical call.
Privacy: de-identify first. No names, dates of birth, conditions, or lab results in a consumer chatbot, which retains and trains on your inputs by default.
Where this comes from
Every use case on this site is grounded in real reports from working nutritionists — not invented by us.