How FixDinner decides what to suggest

People sometimes ask how FixDinner works — specifically, how it goes from "chicken, zucchini, garlic" to a list of actual dinner ideas in a few seconds. The short answer is that it uses AI. The longer answer is more interesting, and explains why the suggestions feel different from what a recipe search engine would give you.

It doesn't search for recipes

Most food apps work by matching your ingredients against a database of recipes. You enter "chicken and zucchini" and it finds recipes that contain those ingredients. This works, but it has a fundamental problem: you end up with recipes that need 12 other ingredients you don't have. The match is shallow.

FixDinner doesn't have a recipe database. It doesn't search. Instead, it uses a large language model — the same kind of AI that powers tools like ChatGPT — to reason about your specific situation and generate suggestions tailored to exactly what you have.

What the AI actually considers

When you submit your ingredients and preferences, FixDinner's AI thinks about several things simultaneously:

• 🥕
  Ingredient compatibility — which of your ingredients work well together and which combinations would taste strange. It knows that chicken, garlic, and lemon are a natural group. It knows that strong fish and delicate herbs don't mix.
• 🌍
  Cuisine coherence — suggestions draw from a coherent flavour tradition rather than mashing up ingredients randomly. Soy sauce, ginger, and sesame oil point toward Asian cooking. Tomatoes, garlic, and basil point toward Italian. The AI follows these logics.
• ⏱️
  Practical cookability — suggestions need to be genuinely makeable on a weeknight, not aspirational. The AI is biased toward techniques that most home cooks can execute: roasting, sautéing, simmering, assembling.
• 🎯
  Variety across the list — if you get five suggestions, they shouldn't all be variations on the same dish. The AI is instructed to generate genuinely different options — different cuisines, different techniques, different ingredients featured as the star.
• 🚫
  Your dietary preferences — if you've told FixDinner about restrictions or preferences, every suggestion respects them. The AI doesn't suggest shellfish to someone who's avoiding it, even if shellfish would otherwise be a natural fit.

Why it says "dinner ideas" not "recipes"

This is intentional and important. A recipe is a specific set of instructions with exact quantities and a defined outcome. A dinner idea is a direction — something you execute using your own knowledge and judgment.

FixDinner gives you dinner ideas with assembly instructions because the goal is to help you cook confidently, not to make you dependent on step-by-step guidance. "Chicken thighs roasted with zucchini and garlic in olive oil, served over orzo" is enough information for any competent home cook to make a great dinner. Adding precise quantities and timing would just add noise.

Where it gets things right — and wrong

The AI is genuinely good at finding non-obvious combinations. Given a strange assortment of ingredients, it will often produce a suggestion that makes you think "I wouldn't have thought of that but it actually makes sense." This is where it earns its keep.

Where it sometimes falls short is in highly specific preferences or very unusual ingredient combinations. If you have an extremely eclectic set of ingredients, the suggestions may be less confident. If you've noted very specific dietary requirements, occasionally a suggestion will slip through that doesn't quite respect them.

This is why the thumbs up/down feedback in the app exists — it helps us understand where the AI needs improvement. And it keeps getting better.

The goal: dinner in under 10 seconds

From opening the app to having a dinner direction should take less than 10 seconds. That's the north star. Everything about how FixDinner works — the ingredient input, the AI prompting, the way suggestions are displayed — is optimized for that single moment when you're standing in front of the fridge and need an answer fast.