
There's a very specific kind of tired that happens at 6pm on a weekday. The kind where someone asks "what do you want for dinner" and your brain just... stops. Not because you don't care about food. Because you've already made approximately four hundred decisions today and this one, somehow, is the one that finishes you off.
I've been there enough times that I started treating it as a solvable problem. And it is — once you know what to actually tell the AI.
Decision fatigue is a real and well-documented cognitive phenomenon: the quality of decisions declines as the number of decisions made earlier in the day increases. Dinner happens at the worst possible time — after work, after commute, after whatever the afternoon threw at you.
The problem isn't that you don't have opinions about food. It's that generating options from scratch requires creative energy you've already spent. That's why staring into the fridge for five minutes produces nothing useful, even when the fridge is full.
What AI does here isn't magic — it's pattern interruption. You don't have to think of something. You just have to pick from something. That's a fundamentally lighter cognitive task.
A useful suggestion is one you can actually execute tonight. That means it passes three tests: you have the ingredients (or close enough), it fits whatever time you have, and it works for the people eating.
Generic suggestions fail the first test immediately. "Make a stir fry" isn't useful if you don't have the vegetables for a stir fry. "Try this Thai recipe" isn't useful if you're gluten-free and the sauce has soy in it. The more specific your input, the more actionable the output.

This is the input that matters most. You don't need to inventory your whole kitchen — just name the proteins and vegetables that actually need using. Two or three ingredients is enough to anchor the suggestion to your real situation.
The magic phrase: "I need to use up..." This shifts the AI from generating generic suggestions to solving a specific problem.
Works: "I need to use up chicken thighs, half a red onion, and some wilting spinach." Doesn't work: "I have some food in my fridge."
If you're genuinely low on ingredients, say that too: "My fridge is basically empty, I have eggs, butter, and some pantry staples." The AI will work with that.
Time is the constraint most people forget to include, and it's the one that makes suggestions actually cookable tonight. Without it, you might get a suggestion that takes 45 minutes when you have 20.
Add: how many people are eating, any dietary restrictions, and how much active cooking time you have. You don't need to write a paragraph — a single sentence covers it.
"Under 30 minutes, cooking for two, no dairy."

Ask for three to five options, not one. When you get a list, picking feels easy. When you get a single answer, your brain starts questioning it.
Once you've picked a direction you like, ask for a simple recipe separately. Combining "give me dinner ideas" and "write me a full recipe" in one prompt tends to produce more output than you need at 6pm. Break it into two steps.
These are ready to use. Copy them directly, swap the bracketed parts for whatever's actually in your kitchen.
I need to use up [ingredient 1], [ingredient 2], and [ingredient 3 that's going off soon]. I also have basic pantry staples. Suggest 3 dinners I can make without a shop run. Keep each suggestion to one sentence — I just need the idea, not the recipe yet.
Example filled in:
I need to use up ground beef, half a zucchini, and some cherry tomatoes that are getting wrinkly. I also have basic pantry staples. Suggest 3 dinners I can make without a shop run. Keep each suggestion to one sentence — I just need the idea, not the recipe yet.
Why this works: "without a shop run" is the constraint that keeps suggestions grounded. "One sentence" prevents an overwhelming wall of text when you're already depleted.
Suggest 3 quick dinner ideas that take under 30 minutes from start to plate. Cooking for two people. [Any dietary restrictions or preferences]. I'm a decent home cook but it's a weeknight — nothing complicated. One sentence each.
Example filled in:
Suggest 3 quick dinner ideas that take under 30 minutes from start to plate. Cooking for two people. One of us doesn't eat red meat. I'm a decent home cook but it's a weeknight — nothing complicated. One sentence each.
Why this works: "start to plate" closes the loophole where AI gives you a 10-minute recipe that assumes 20 minutes of prep. "Nothing complicated" signals weeknight mode.
I want something healthy for dinner tonight but I'm not in a salad mood. Constraints: [time], [dietary restrictions if any], cooking for [number]. Give me 3 options that feel like real food, not diet food. One sentence each.
Example filled in:
I want something healthy for dinner tonight but I'm not in a salad mood. Constraints: 25 minutes, gluten-free, cooking for one. Give me 3 options that feel like real food, not diet food. One sentence each.
Why this works: "feel like real food, not diet food" is surprisingly effective language — it steers away from the punishing, joyless suggestions that health-framing can trigger.
All three handle dinner suggestions well on their free tiers. The difference isn't output quality so much as usage limits and interface.
ChatGPT (free, GPT-5.2 Instant with rate limits) — the most familiar. Good at following the "one sentence each" instruction and returning a clean list. Rate-limited on the free tier during peak hours.
Claude (free) — tends to produce more natural, conversational suggestions and is particularly good at following complex constraints without dropping any of them mid-response. No image input on the free tier.
Gemini (free) — useful if you want it to check what's currently in season or pull from real web data. Has Google integration that the others don't.
For a quick weeknight dinner decision, any of the three work. The prompt matters more than the model. Use whichever one you already have open.

Flavorish — free to download, iOS and Android. Enter your ingredients or describe what you want; get a full recipe back. Saves recipes to a personal cookbook and imports from Instagram, TikTok, or any website. The 5-recipe AI generation limit on the free tier is enough for occasional use; premium is $4.99/month for unlimited.
DishGen — free Basic account with 15 credits per week. Enter an ingredient list or describe the meal you want, get multiple recipe options, iterate via chat. Good for people who want to see several options at once rather than a single suggestion.
Samsung Food — free, iOS and Android. Strong if you're also meal planning for the week. The dinner suggestion feature works well inside the weekly planner, but it's more useful as a planning tool than a "what should I eat right now" tool.
Every tool on this list — including the dedicated apps with pantry features — only knows what you tell it. If you say "I have chicken thighs" but the ones in your fridge have been there since Sunday and you're not sure about them, the AI doesn't know that. If you forgot to mention you're out of olive oil, the recipe will call for olive oil.
The fix is simple: do a 60-second scan of what you actually have before prompting, and be specific. The AI can only work with what you give it.
If you use the same tool with similar inputs every night, you'll start seeing the same suggestions cycle around. Chicken + vegetables → stir fry. Ground beef + pantry → pasta. The model isn't being lazy — it's giving you the statistically common answer for those inputs.
Two things help: rotate the cuisine style in your prompt ("something Japanese-inspired" vs "something Mediterranean"), or describe a mood rather than just ingredients ("I want something that feels cozy and low effort" gives different results than a plain ingredient list).
The prompts above work for tonight. But tomorrow evening you'll be back at the same blank screen. At Macaron, we built something that actually remembers — what you liked, what you skipped, what fits your evenings. One conversation, and the next time you ask "what should I make," it's not starting from nothing. Try it free.

For pure flexibility and constraint-handling: ChatGPT or Claude on the free tier, with a well-structured prompt. For a dedicated mobile experience where you also want to save and organize recipes: Flavorish. For seeing multiple options at once and iterating through chat: DishGen. The prompt matters more than the tool — a specific, well-structured input on any of these produces significantly better output than a vague one on the "best" tool.
Yes — this is one of the most reliable use cases for general-purpose AI. Tell it specifically what you have (especially anything that needs to be used up), add any constraints, and ask for suggestions that don't require a shop run. The more specific the ingredient list, the more relevant the suggestions. Apps like Flavorish and DishGen are built specifically for this use case if you prefer a dedicated interface over a general chatbot.
Related Articles: