← Back to recent questions

Yes, on both counts, and the answer is more interesting than you might expect. The photo-of-the-fridge trick is one of the most quietly useful things any chatbot will do for you, and pulling a scattered pile of recipes into a single place that you can actually ask questions of is now a one-afternoon job. The ongoing tracking is a bit harder than the marketing suggests, and the honest answer there is to keep it simple. Here is how I would do each part.

How I'd approach it

The photo-to-dinner moment. Open Claude, ChatGPT, or Gemini on your phone. Take a photo of the open fridge. Add the staples sitting in the pantry that the photo cannot see (rice, pasta, oil, tinned tomatoes). Then type something like: "Suggest five dinners I can make in under 40 minutes using mainly what is in this photo. Tell me what extras I would need to buy and roughly how much each would cost. Flag anything that looks like it needs eating tonight." The answer will be specific recipes with quantities, not generic suggestions. There is a fuller walk-through of this on the Build a Meal Planner page, which started from exactly this scenario.

Building a recipe collection. For ongoing recipe storage, set up a ChatGPT Custom GPT or a Claude Project. Upload your favourite recipes (PDFs, photos of cookbook pages, or just typed notes), tag them by cuisine and difficulty, and ask it to help you cook them, scale them, or substitute ingredients. The first set-up takes an afternoon. Once it is done, "what can I cook tonight from those recipes that uses what I have on hand" becomes a one-line question. NotebookLM works for the same purpose and is free, although it is less good at the active "help me cook this" part of the job.

Tracking what is actually in the fridge. Be honest with yourself before going down this rabbit hole. Genuine fridge inventory needs you to log every shop and tick things off as you cook them, and most people will not stick to that for more than a fortnight. The dedicated apps that try this (Whisk, SuperCook, Mealime) all have decent free tiers and are worth a quick look, but I have not met anyone who keeps their fridge state perfectly current in one. The pragmatic version is: photo of the fridge once a week before the shop, photo of the receipt after, and let the chatbot keep the rough running list. It is less precise than a real database but it is the version you will actually use.

If you want to go further. The same site has a tutorial for building your own meal-planner web app without writing code, using Bolt.new. That gives you a shopping list grouped by supermarket section, the ability to swap meals, and a fridge-photo upload step that the standard apps do not handle as well as a chatbot.

And once dinner is on the table, the same trick works on the liquor cabinet. Point the phone at the bottles on the shelf and ask "based on these, plus ice and the basics from the fridge, what cocktails can I make? Give me three with full ratios and the method." You will get classics that match what is actually on the shelf rather than a generic list. This version of the trick is more reliable than the fridge photo for one straightforward reason: every bottle is labelled. The AI is reading writing on a label, not guessing what might be in a half-used jar at the back of the second shelf. Identification errors drop close to zero. You still want to sense-check the ratios before pouring, but you do not have to wonder whether the AI thought a bottle of soy sauce was something else.

What I'd avoid

The chatbots are good at recognising obvious items in a fridge photo, but they make mistakes. Half-empty jars at the back, leftovers in a takeaway container, anything in a sealed box: the AI will sometimes invent what is in there, or miss it entirely. Always read the recipe back before you start cooking and ask "did you actually see all of these in the photo, or did you assume?". For anything to do with allergies or strict dietary requirements, do not rely on the AI's identification at all. Read the labels yourself.

The other thing to avoid is paying for a premium recipe app on day one. The free tiers of all the main chatbots do this job well enough that you can wait a month before deciding whether you want a dedicated tool, a Custom GPT, or just the photo-and-ask habit. Most people end up using the photo-and-ask habit and nothing else.