The £40 Aldi Challenge: What Happened When Free AI Did the Meal Plan

Forty pounds. Seven days. Four mouths. That was the brief I gave a handful of free AI tools: build a week of family meals that could be shopped at Aldi for £40, using ordinary UK ingredients and not assuming we already had a second freezer full of bargains. I expected nonsense. I got a mixture of genuinely useful meal ideas, wildly optimistic prices and one plan that appeared to believe a whole salmon was a store-cupboard staple.

The experiment was not a laboratory test. It was a real weekly shop, done in November 2026, with a receipt at the end. Prices move, Aldi ranges change, and what is on the shelf in one town may not be in another. That is part of the point. Free AI can help you plan; it cannot walk the aisles for you.

The brief I gave the bots

I used the free tiers of a few popular chatbots and one free meal-planning app. My prompt was deliberately specific: family of four, two adults and two primary-school-age children, seven evening meals, plus breakfast and lunch basics, £40 total, Aldi own-brand where possible, no fancy ingredients, and assume only oil, salt, pepper and a few dried herbs are already in the cupboard. I asked for a shopping list grouped by aisle, quantities, estimated prices and a total.

I also asked each tool to flag anything it was unsure about. That last request was revealing. Some tools ignored it completely. One listed 'prices are approximate' and then produced a total that was £12 under the sum of its own items.

The plans: from sensible to sci-fi

The best plan was boring in the right way. It leaned on eggs, oats, potatoes, pasta, rice, tinned tomatoes, beans, lentils, carrots, onions, frozen peas, chicken thighs and mince. Dinners included a bean chilli, lentil bolognese, pasta bake, roast chicken thighs with potatoes and cabbage, a chicken and vegetable rice, a frittata and jacket potatoes with cheese and beans. It was not glamorous. It was realistic.

The worst plan was a masterpiece of wishful thinking. It suggested salmon fillets, avocados, fresh berries and a joint of beef, then claimed the whole thing came to £38.72. It had also forgotten cooking oil, stock cubes and any lunch for the children. Another bot suggested 'use leftover roast chicken' without putting a roast chicken on the list. A third invented a 1kg pack of chicken breast for £2.50. If only.

The price problem

This is where free AI falls over. Chatbots are language models, not live price feeds. They can produce a plausible-looking number because that is what the pattern suggests, not because they have checked the Aldi shelf. Some can browse the web, but even then they may mix up pack sizes, promotions and different retailers. Asking for exact prices invites false precision.

The fix is to ask for price bands and a cushion. I changed the prompt to: 'Use estimated UK prices, label every price as an estimate, keep the total under £35 so there is room for price rises, and list cupboard assumptions separately.' That produced a far better list. It was still not a till receipt, but it was close enough to shop from.

The actual Aldi run

I went in with the revised list and a calculator. The basket included oats, eggs, milk, bread, rice, pasta, passata, tinned tomatoes, baked beans, red lentils, carrots, onions, potatoes, frozen peas, a cabbage, chicken thighs, mince, cheese, yoghurt, bananas, apples, stock cubes and a bag of frozen mixed vegetables. The final bill came to just over £41. That is over the £40 target, but not by much, and it covered seven dinners plus breakfast and lunch basics. A like-for-like run at Tesco would have been a few pounds more, though Tesco had better availability on one or two items.

The children did not eat everything without complaint. The lentil bolognese was rebranded as 'special pasta sauce' and accepted. The frittata was a hit. The cabbage was less popular, but it stretched across two meals. The plan worked because it was flexible: when the shop ran out of the exact yoghurt, we bought own-brand natural yoghurt and moved on.

What I would prompt differently

• Name the supermarket and the budget, not just 'cheap meals'.
• State the family size and ages, because a teenager eats differently from a five-year-old.
• List what is already in the cupboard. Otherwise the bot will assume a spice rack that does not exist.
• Ask for seven dinners, not seven days of every meal, unless you want a list that runs to three pages.
• Demand quantities and a recalculated total. Then check the addition yourself.
• Ask for leftovers to be used in the next day's lunch. That is where the real savings are.
• Tell it to avoid fragile assumptions about allergies, medical diets or school rules. Those need a human decision.

The verdict

Free AI is genuinely useful for meal planning. It is good at ideas, structure and turning a vague ambition into a list. It is bad at prices, pack sizes and the difference between 'cheap' and 'actually in Aldi this week'. The £40 challenge was possible, but only because the plan was simple, the cooking was from scratch and the human did the maths at the till.

Would I use it again? Yes, with a tighter prompt and a calculator. Would I trust it to know the price of mince? Not a chance. The bots can write the menu. The supermarket still gets the final word.