Claude reads what it can see, so ninety percent of a bad answer is a bad photo. Sixty seconds here changes the whole output.
The important part of this prompt is the confidence tiers. Left to its own devices a model will confidently name things it half saw, and you find out when you are standing at the stove. Making it separate what it is sure about from what it is guessing fixes that before you cook.
These photos are my fridge, my crisper drawer and my pantry. Read them carefully and then build me a set of meals from what is actually in them.
First, inventory what you can see in three separate lists. List one is what you are confident about, with a rough quantity where you can tell (half a block of tofu, four eggs, around 200g of spinach). List two is your best guesses, where you think you know what something is but are not certain, and for each of those ask me to confirm before you rely on it. List three is anything that looks like it should be used soon, such as opened dairy, wilted herbs or cut fruit, and treat those as the ingredients to build around first.
Then give me eight meals I could genuinely make, sorted from fastest to slowest, and be honest about the split: mark each one as either everything is here already, or needs one or two things I would have to buy. For each meal give me the name and cuisine, total time from standing up to sitting down, how many it serves, the ingredients used from my photos, a short numbered method under eight steps, and one swap if I am missing something.
Do not invent ingredients you cannot see and I have not mentioned, and skip any disclaimers about consulting a professional. If my photos are too dark or too crowded to read properly, tell me that first and tell me exactly what to re-shoot.A generic meal plan is easy to ignore. The version you cook from is the one that already knows you eat dinner at 8pm, hate coriander, and have twenty minutes on a weeknight.
Redo that plan against my real constraints, and keep it to what my photos actually support.
My rules: I am cooking for [number] people, I have [how long] on a weeknight and more on the weekend, and my kitchen has [stovetop, oven, air fryer, rice cooker, microwave, whatever you have]. I am aiming for [for example, roughly 35g of protein per main meal, or under 600 calories, or high fibre, or nothing in particular], I do not eat [allergies and hard nos], and I am [beginner, comfortable, confident] with cooking.
Cut anything that breaks those rules rather than adapting it into something unrecognisable, and tell me which meals you dropped and why. For the ones that survive, add the approximate protein per serving and mark the two that are best for cooking once and eating twice.
Also flag the point in each recipe where a beginner usually goes wrong, in one line, because that is more useful to me than another ingredient list.This is where the money actually gets saved. Rather than a full week's shop, you are asking for the smallest number of items that unlock the most extra meals from food you already own.
Now write me one shopping list for the rest of the week, built on top of what is already in my photos.
Give me the shortest list of items that unlocks the most additional meals from what I already have, no more than eight things, and next to each item say which meals it makes possible. Order the list by how many meals each item unlocks so I can stop reading partway down if I am in a hurry, and group it by aisle: produce, protein, dairy, dry goods, freezer.
Then give me a rough total cost estimate in [your currency] with a note that it is an estimate, and tell me which two items I could skip if I wanted to spend less and what I would lose by skipping them.
Finish with a three line plan for the week: what to cook tonight, what to batch cook on my quietest evening, and what that batch turns into for two more meals.Doing this once is a nice trick. The version worth keeping is a project that already knows your budget, your macros, your equipment and the meals you actually liked, so next Sunday you send a photo and nothing else. Projects are on every plan, and free accounts get up to five, per Anthropic's projects documentation.
You are my meal planner. Whenever I send a photo of my fridge, pantry or crisper drawer, run the full routine without me having to explain it again.
Always start by inventorying what you can see in three lists: confident, best guess (ask me to confirm these), and use first because it is close to turning. Never invent an ingredient you cannot see and I have not mentioned. If a photo is too dark or too crowded to read, say so and tell me what to re-shoot rather than guessing.
My standing rules: I cook for [number] people, I have about [minutes] on a weeknight, my kitchen has [equipment], I am aiming for [macro or health goal], I never eat [allergies and hard nos], and my weekly grocery budget is around [amount]. Meals I already loved are in the project files, so lean towards that style and do not repeat the same cuisine two weeks running.
Every time, give me eight meals sorted fastest to slowest with time, servings, approximate protein and a method under eight steps, then a gap-filling shopping list of no more than eight items ordered by how many meals each one unlocks, then a three line plan covering tonight, the batch cook, and what the batch becomes. Keep the whole reply scannable on a phone with clear headings, and no disclaimers.What to put in the project files, so the plans get sharper every week: