Chat Picker

AI Assistants for Recipes and Nutrition Analysis: Personalized Diet Advice Assessed

Vendor pages document different file, personalization, and usage settings, while this afternoon test plan helps you compare ChatGPT, Claude, and Gemini without claiming a winner.

Sources checked 2 Oct 2026

ChatGPT, Claude, and Gemini can be compared for recipe changes, nutrition calculations, diet-constraint flags, personalization, and file-based analysis, but vendor pages do not establish that one gives better diet advice. OpenAI's data-analysis page says ChatGPT's chosen analysis method may not match your intent, Anthropic's incorrect-response guidance advises checking original sources, and Google's Gemini source help says related sources may appear with a response; Chat Picker has not tested the assistants for these tasks, so this page gives you an afternoon test plan instead of a ranking. This rewrite removes the earlier page's unsourced statistics and test results.

Know these limits before you start

Treat nutrition and diet-constraint output as a draft, not medical advice. OpenAI's truthfulness guidance says ChatGPT may sound confident when wrong; Anthropic's guidance calls misleading output hallucination and says not to rely on Claude as the only source of truth; Gemini's safety guidelines say outputs reflect training-data limits and list medically conflicting information as harmful. OpenAI's Usage Policies restrict tailored licensed medical advice without professional involvement, while Anthropic's Usage Policy requires qualified review for covered high-risk advice. Have a qualified professional review renal restrictions, allergies, and other diagnosis-specific advice.

File limits may decide your test design. As read on October 1, 2026, OpenAI's File Uploads FAQ lists a 512MB hard limit per file, a 2M-token cap per text or document file, and 3 uploads per day on Free. Anthropic's file-upload page gives 500MB per chat file and 30MB per project file. Google's file-upload help gives up to 10 files per prompt and 100MB per supported non-video file, subject to availability. Use a file that every selected plan can accept.

Check settings before testing. OpenAI's memory guidance separates memory from model-improvement controls and says relevant preferences may be remembered. Claude can search previous conversations; Anthropic says chats improve models only when allowed and incognito chats are excluded. Gemini personalization can use past chats, connected app content, and response preferences, but is not available to everyone. The Gemini plans page and Privacy Hub material read do not settle a model-training opt-out.

The test plan

Run each task in a fresh chat, keep inputs identical, and log the assistant, plan, settings, file, prompt, output, source checked, error, and access issue. Do not score until every run is complete.

  1. Substitution and scaling. Prepare a recipe card with quantities, servings, pan type, a scaling target, and dietary notes. Say: “Use the attached card. Return the original ingredients, a proportional list for the stated target, and arithmetic for every changed quantity. Preserve the cuisine, state assumptions, and do not invent missing values.” Record omissions, unexplained substitutions, broken ratios, and arithmetic mismatches.

  2. Macronutrient breakdown. Prepare a quantified recipe with its package label or another source you trust. Say: “Using the attached recipe and source, calculate protein, carbohydrate, fat, and energy for the entire recipe and per serving. Show ingredient-level arithmetic, identify label values and estimates, and mark anything you cannot verify.” Compare each result with the source and a calculator. Record unit, serving, arithmetic, and false-precision errors.

  3. Dietary constraints. Prepare an ingredient list to check against current low-FODMAP guidance and another based on clinician-supplied renal guidance. Say: “Review each attached ingredient list against its attached guidance. Quote the relevant passage, distinguish culinary substitutions from clinical care, and do not declare any food safe or unsafe.” Record mismatched citations, missing caveats, and advice needing professional review.

  4. Personalization. Prepare a recipe card and preference profile. In a fresh chat, say: “Use the attached files. Remember that I prefer metric measurements and concise numbered steps. Suggest a substitute for the recipe ingredient named in the card, then state which preference you used.” In a later chat, ask: “What measurement system and response format did I ask you to prefer? Do not guess.” Record verified recall, unsupported memory claims, and any setting changes.

  5. Files and output format. Prepare the recipe as a spreadsheet, PDF, and photo, then use the format you expect to rely on. Say: “Analyze the attached recipe file. Return a table with ingredient, quantity, macronutrients per serving, confidence, and source location; identify unreadable cells instead of guessing.” Record extraction errors, table readability, source locations, upload friction, and ease of copying into your notes.

  6. Regular use and cost. Prepare a workload note listing expected prompts, files, and repetition. Say: “Review this workload note. List which requested operations may depend on plan, file limit, or usage allowance. Do not state a current price or limit unless the attached vendor material shows it; mark gaps.” Record the verified price, included usage, relevant limits, needed paid features, and blocked tests.

How to read your results

Start with rules you will not accept: no invented quantities, unsupported medical certainty, required files rejected, or critical constraints omitted. Count only errors you observed against prepared sources, separating calculation, extraction, citation, dietary, personalization, interface, and access issues. Repeat decisive failures in a fresh chat and record both outputs. Apply your rules first, then recurring cost and workflow fit. If nothing passes, change tools or seek qualified help rather than treating fluent prose as validation.

Where the plans differ

The three-way comparison matrix tracks the relevant rows and leaves unconfirmed figures blank. Every price, storage figure, multiplier, context figure, and usage figure below was read on October 1, 2026; each linked name identifies its vendor source. Unless marked otherwise, dollar figures are US dollars.

  • Free plan: ChatGPT Free, Claude Free, and Gemini all have free access; Gemini requires a Google Account.
  • Low-cost tier: ChatGPT Go shows A$13/month on the Australian page; its US-dollar price is not verified. Claude lists no tier between Free and Pro. Gemini AI Plus is $4.99/month with 2x Free usage limits and 400GB storage.
  • Main paid plan: ChatGPT Plus is $20/month, billed monthly. Claude Pro is $20/month monthly, or $17/month on annual billing with $200 upfront. Gemini AI Pro is $19.99/month with 4x Free limits and 5TB storage.
  • Heavy-use plans: ChatGPT's Pro tiers are $100, $200, and $500/month. Claude Max starts at $100/month with 5x or 20x Pro usage; the 20x price was not shown. Gemini Ultra is $99.99/month for 5x AI Pro limits or $199.99/month for 20x.
  • Context and usage: ChatGPT's app context varies by model and plan; Claude says up to 1M on every plan; Google's plans page gives no verified app token figure. ChatGPT and Claude publish no exact message counts on their pricing pages; Gemini describes compute-based limits that refresh every 5 hours, up to a weekly limit.
  • Training on chats: ChatGPT offers opt-out on Free, Go, Plus, and Pro. Claude lists opt-out on Free, Pro, and Max and says Team is not trained on by default. Gemini's training setting is not verified.

Confirm the current vendor page before subscribing.

Sources

For current prices and limits, see the dated comparison pages.

Compare assistants