Care Home Menu Planner: Four-Week Cycle With AI and Dietary Checks

Give the AI your meal pattern, ordering days, budget per resident per day, the dietary categories you serve and your allergen rules, and ask for a four-week grid, a weekly shopping list and an allergen matrix. Then have the cook check every dish against real labels, and a dietitian approve the dietary variations, before anything is served.

AI is good at the tedious parts of a cycle: balancing the weeks, avoiding repeats, scaling quantities and tabulating. It is poor at the safety parts, because it cannot read your product labels or know a resident's needs. The method below keeps the drafting and the checking apart, so the draft saves time and the checks stay with the people who are accountable for them.

Follow me on Instagram@sagnikteaches

Gather the inputs before you open a tool

A menu draft is only as good as the brief. Collect these into one page, which you will paste into the prompt and keep as a standing document.

Connect on LinkedInSagnik Bhattacharya

• Meal pattern. For example breakfast, mid-morning drink and snack, lunch (main meal), afternoon tea, supper, plus night-time snacks on request.
• Numbers. Residents per day, with counts for each dietary category: standard, vegetarian, diabetic-friendly, fortified, each texture-modified food level you serve (for example soft and bite-sized, minced and moist, pureed), thickened drinks by level, and religious or cultural needs. Counts only, no names, no initials.
• Allergen rules. Your list of declarable allergens, written in full, and any that are excluded from the kitchen altogether. Include how you handle "may contain" labelling.
• Budget. Food cost per resident per day, and where the flexibility is. Use your own figure in USD, for example $6.00 per resident per day for illustration.
• Kitchen limits. Equipment, staffing hours, delivery days and storage, because a menu that needs an oven nobody has is not a menu.
• Preferences. What residents have said they like and dislike, by item rather than by person, and what has been popular on past menus.

Say a 30-resident home with a cook working 07:00 to 14:30 and a kitchen assistant for tea and supper. The page might read: 30 residents; counts: 14 standard, 6 fortified, 5 minced and moist, 3 pureed, 2 vegetarian; four residents on thickened fluids; four allergens excluded from the building; two main deliveries a week and a top-up on Thursday.

Subscribe on YouTube@sagnikteaches

Step 1: ask for the structure, not the recipes

Do not begin with "write me a four-week menu". Start with the skeleton, so you can correct the pattern before any dishes exist.

I run the kitchen in a 30-bed care home. Using the brief below, propose the SKELETON of a four-week menu cycle: for each day of each week, give the type of main meal (for example roast, pie, fish, casserole, pasta, curry) and the type of pudding, without recipes. Rules: no main-meal type twice in one week; fish at least once a week; a roast on Sunday every week; one vegetarian-led day per week; keep the pattern easy for one cook to prepare. Brief: [paste]

Read the skeleton against your kitchen. Does a pie fall on the day with the shortest staffing? Is there a long gap between fish meals? Fix the pattern by instruction ("move the casserole to Wednesday, swap Thursday and Friday") until it suits your rota. Only then ask for dishes.

Step 2: draft dishes and variations, one week at a time

Ask for Week 1 only, with a standard main and pudding for each day, then the variations: a vegetarian option, a fortified version where relevant, and texture-modified alternatives. Request ingredient lists with quantities per 30 portions. Working a week at a time makes errors easier to spot and allows you to feed back.

Now draft Week 1 in detail, using the skeleton. For each day: lunch main, pudding, supper option, and the afternoon tea item. For each main, add (a) a vegetarian version, (b) a fortified version (extra energy and protein added to the same dish, not a different dish), (c) a minced and moist version, (d) a pureed version. Give ingredients for 30 portions in grams and millilitres. Mark each ingredient that is commonly an allergen with [A] and name the allergen. Say clearly when a version cannot be made from the same base recipe. Do not state nutritional values.

The last instruction matters. Nutritional estimates from a chat tool look precise and may be wrong, and nobody should plan a resident's diet from them. If you need nutritional analysis, use your dietitian's software or an approved recipe database.

An illustrative day, with what the cook would fix

A draft for Tuesday of Week 1 might read: "Lunch: chicken and leek pie with mashed potato and carrots. Vegetarian: leek and cheese pie. Fortified: add cream to the sauce and butter to the mash. Minced and moist: finely minced chicken in thick sauce under a soft pastry lid, mashed potato, mashed carrots. Pureed: pureed chicken and sauce, shaped and served separately. Pudding: apple crumble with custard."

The cook reads it with a pen. The pie pastry contains gluten and butter, so cereals and milk are flagged; the custard may contain egg or milk depending on the product; the cream adds milk. The tool has not seen your pastry or custard product, so the allergens are guesses. The cook writes the actual supplier product names beside each ingredient and replaces guessed allergens with those on the real labels. Then there is a practical note: "No pastry on the minced and moist plate. Serve the filling only." The tool had suggested a soft pastry lid, which is not suitable for everyone on a texture-modified diet. The pudding gets the same question: IDDSI's detailed level definitions flag crumbly textures, such as dry cakes and biscuits, as a choking risk, so the crumble topping needs a texture-modified alternative, and the therapist decides what that is. This is why a dietitian or speech and language therapist checks the variations.

Step 3: build the allergen matrix from real labels

An allergen matrix is a table with dishes down one side and your declarable allergens across the top, with a mark where the dish contains the allergen. It should be created from product specifications and labels, not from the AI's guess. Use the tool for the layout and for populating a first pass; then verify every mark.

The matrix above shows five allergens to fit the page. Yours needs a column for every allergen your food law lists. Where the 14-allergen list applies, the other nine are peanuts, tree nuts, sesame, soybeans, fish, crustaceans, molluscs, lupin, and sulphur dioxide and sulphites above ten parts per million. Some markets use a shorter list of major allergens (one widely used list has nine), so build the columns from the rules that apply to your kitchen, then add any allergy a resident has that sits outside the legal list. Put "check" wherever the answer depends on a brand, and make sure those cells are resolved before the menu is issued. A change of supplier or a reformulated product should trigger a re-check of every dish that uses it, so keep a column for the product name and a date.

For printed or displayed menus, the tutorial on allergen icons and dietary labels covers layout and proofing. For the wider question of why an AI should not answer allergen questions from guesswork, the restaurant chatbot tutorial explains the same risk from the other side.

Step 4: turn the cycle into a shopping list

This is where AI saves the most time. Paste the ingredients for a week and ask for a consolidated list grouped by supplier section (fresh, chilled, frozen, dry, store cupboard), with totals across the week and a note for items used on two days that could be bought once.

Combine the ingredient lists for Week 1 into one shopping list for 30 residents, grouped as: fresh produce, meat and fish, dairy and eggs, frozen, dry goods and tins, other. Add up quantities used across the week, convert to the unit suppliers sell in (kilograms, litres, whole trays), and flag any item used on more than one day. Do not add items that are not in the recipes. List any quantity that needs checking with me.

Illustrative output: "Fresh: leeks 2.4 kg (Tuesday pie, Thursday soup); carrots 6 kg across Tuesday, Wednesday and Friday. Dairy: milk 24 litres, butter 3 kg, cream 2 litres (fortified versions only)." The cook checks against what is already in store, adjusts for portion sizes your residents actually eat, and adds a small buffer for waste. Ask for a "stock check first" column so the order is net of what you already have.

Costing works the same way: paste last month's supplier prices, in your own currency, and ask for a cost per portion and per resident per day for each week. Check the arithmetic by hand on three dishes before trusting the rest. A cost model can be wrong in a way that looks tidy.

Step 5: the cook and dietitian review

No menu goes live on AI's say-so. The review has two roles with two different jobs.

Texture-modified food has its own standard. If you work with the IDDSI framework, its eight levels run from 0 to 7: levels 0 to 4 describe drinks, from thin to extremely thick, and levels 3 to 7 describe foods, from liquidised through pureed (4), minced and moist (5) and soft and bite-sized (6) to easy to chew and regular (7). Ask your speech and language therapist how each dish maps to a level, and test the textures in the kitchen with the framework's own checks, such as the fork pressure and spoon tilt tests for food and the flow test for drinks. A tool may describe a dish as "pureed" without it actually meeting a level, and residents who need modified textures are at risk if it does not. The dietitian or therapist confirms suitability; nothing here is clinical advice.

Gathering resident feedback without names

A good menu depends on what people like to eat. Ask at residents' meetings, at the table and through care staff, and write feedback as "three residents liked the fish pie; two did not like the soup" rather than naming anyone. Before the next cycle, paste those notes in and ask the tool to propose swaps that keep the nutrition pattern intact: replace unpopular dishes with something similar in type, keep favourites, and avoid repeating the same meal twice in a week. Then repeat the cook and dietitian review for any new dish. The residents' meeting tutorial shows how to record those conversations.

A second day, with a fish dish and a different fix

Friday of Week 2 in the draft: "Battered fish with chips and peas; vegetarian: cheese and onion flan; minced and moist: flaked fish in parsley sauce with mashed potato; pudding: sponge with jam sauce." The cook finds three things. The batter uses wheat flour and sometimes egg, so gluten and egg are flagged. The peas are fine, but the frozen chips from the current supplier are coated, and the label says "may contain celery", so celery goes in the matrix as "may contain" and the home's policy for such wording applies. Finally, "flaked fish" for the minced and moist version may carry bones, and the cook knows from experience that the kitchen's own fillets are skinless and boneless, so the note changes to "boneless fillets, checked by hand before serving". None of those three corrections came from the tool. All three came from someone who has handled the products.

A costing example with numbers

Illustrative figures. Suppose the target is $6.00 per resident per day for food, so $180 a day for 30 residents. Pasting a week's supplier prices into the tool produces a table of costs per dish. Say Tuesday's pie costs $2.10 per portion at lunch, the standard tea item $0.70, breakfast $0.90, supper $0.80, drinks and snacks $0.95: that is $5.45 a day. Wednesday's roast comes out at $6.80. The week's average is $5.92, which is within budget, but Week 3 comes out at $6.30. The question for the tool is "which two dishes in Week 3 raise the cost most, and what cheaper dish of the same type could replace each?" It might suggest swapping a lamb dish for a chicken one and a salmon dish for a white fish. The cook then judges whether that is acceptable to residents and whether the dietary variations still work. Check three of the tool's arithmetic lines by hand, because a percentage that looks plausible can hide a multiplication slip.

Finger food and eating with dignity

Some residents do better with food they can pick up, particularly people living with dementia who find cutlery difficult. A short extra prompt gives you ideas: "For each main meal in Week 1, suggest a finger-food alternative that carries the same food group and is safe to hold." The reply may suggest sandwiches, fish goujons, sliced omelette or roasted vegetable batons. The cook and dietitian decide which are suitable for whom. Do not name residents in the prompt. Instead, record in your own kitchen sheet which residents have a finger-food plan, and keep it where the cook can see it on the day. The aim is to keep dignity and independence at the table, and a tool can supply ideas faster than a cookery book, but the suitability judgement stays with staff.

Drinks, snacks and the hours between meals

Menu planning tends to centre on the main meal. Between-meal drinks and snacks matter as much for residents who eat little at a time. Ask the tool to produce a rotating snack list: three savoury and three sweet options, one fortified option, one that is suitable for texture-modified diets, and one that is easy to hold. Keep the same allergen rules, and add the items to the allergen matrix, because a biscuit or a yogurt is a product with a label like any other. Put the snack list in the same document so that night staff and the kitchen assistant use the same checked options.

Festival days and special occasions

A fixed four-week cycle leaves gaps for birthdays, festivals and cultural celebrations. Keep a separate one-page list of special-occasion menus with the same checks: a recipe, an allergen row, a dietitian check for texture-modified versions, and a shopping list. Ask residents and families what dishes matter to them; the tool can then draft an authentic-sounding menu and quantities, but the cook should check the ingredients, because a traditional dish can contain nuts, shellfish or alcohol that need to be flagged. Add the dates to your activity plan so the kitchen is not surprised.

A three-week timeline to launch the first cycle

Week 1: gather the brief, run the skeleton step and correct the pattern with the cook. Week 2: draft Weeks 1 and 2 of the cycle in detail, build the allergen matrix from real labels and send the variations to the dietitian. Week 3: draft Weeks 3 and 4, run the costing, and hold a short tasting or feedback session with residents on two new dishes. Then approve, version and start the cycle. Plan a review after the first full rotation, and note which dishes were left on the plate, which is the most honest feedback you will get.

Where AI menu drafts go wrong

Unsafe allergen assumptions. The tool says a dish is "dairy-free" because the recipe did not name milk, but the stock cube or sausage did contain it. This is the single biggest risk. Never serve on the strength of the draft.

The tool forgets an exclusion. A prompt line saying "no peanuts" can disappear by Week 3 of a long chat. Repeat the exclusion list in every request, and read each week for the excluded items.

Dishes that are too ambitious. Chat tools like variety; your cook may have three hours and one oven. Add "one cook, one oven, 07:00 to 14:30" to every prompt.

Repetition across weeks. Compare main-meal types across the four weeks in a simple grid. If "pie" appears nine times, ask for swaps.

Nutrition claims. If a draft mentions protein grams, vitamins or "suitable for diabetics", delete it. Those statements need your dietitian.

Substituted ingredients on delivery. When a supplier substitutes a product, your allergen matrix is out of date the moment it arrives. Add a rule: any substitution is checked by the cook against the label before use, and the matrix is updated.

Keeping the cycle alive

Once approved, keep one master document per cycle with a version number, an approval date and the names of the reviewers. When you tweak a dish, create a new version and re-run the allergen check for that dish. Review the cycle at least twice a year for seasonal produce, and after any change of supplier. A winter cycle can lean on root vegetables, stews and puddings, and a summer cycle on salads, cold plates and fruit, but each needs its own check.

If you want a general version of the cycle, with costing for a catering kitchen, the four-week menu cycle with costs and allergens goes into the arithmetic in more depth. To keep mealtimes and activities in step, the activities calendar tutorial shows how to plan the month around them. And if you want someone to look at your kitchen process with you, my 1:1 consultation can help you decide what is safe to hand to AI and what is not.

Menu planning questions from care home cooks and managers

Can AI tell us which dishes contain allergens?

It can suggest likely allergens from a recipe, but it cannot see your actual ingredients and brands. Allergen information must come from product labels and supplier specifications, checked by the cook. Treat any AI-generated allergen column as a first draft to be verified against labels, never as the final record.

How do we cover texture-modified and fortified diets?

Give the tool the categories your home uses, such as the IDDSI levels or your own names, and ask for a variation per main dish. Then have a dietitian or speech and language therapist confirm each variation. AI can draft options, but the suitability decision for each resident is theirs.

Why a four-week cycle rather than a weekly menu?

A four-week rotation avoids repeating the same meals too often while keeping ordering, costing and cooking routines predictable. It also gives you a fixed set of dishes to check once for allergens, then reuse. Some homes also run separate summer and winter cycles.

Should residents have a say in the menu?

Yes. Resident feedback is the best test of a menu, and cooks often learn more from a quick conversation or a residents' meeting than from a survey. Record likes and dislikes without names in the planning notes, and rotate favourites through the cycle.

Further reads

• Meal Prep Business Software: Weekly Menus, Orders and AI Labels — Software that holds menus, orders and labels together.
• AI for Delis: Daily Specials, Platter Orders and Allergen Labels — Allergen labels on food made to order.
• Care Home Medication Audit: A Monthly Template and AI Action Log — Another resident-safety record with a signed action log.
• AI for Dementia Care Providers: Life-Story Profiles and Activity Plans — Food preferences and memories that shape mealtimes.
• How to Write Menu Descriptions With AI That Still Sound Like You — Plain, appetising wording for the menu board.
• AI for Care Homes: A Registered Manager's 90-Day Implementation Plan — Where menu planning sits in a wider AI rollout.
• AI Use in Long-Term Care Facilities: Where Resident Data Must Not Go — A long-term care operator's guide to AI: which jobs are safe, which resident details never leave your system, and the staff rules to print.
• AI and Senior Living: Where a Small Care Home Should Start — A small care home's starting order for AI: family updates, activities, rotas and enquiry replies first, safeguarding and resident records kept human.
• Can Dietitians Use AI for Admin and Content? What Stays Human — Which dietitian tasks AI can draft (admin, handouts, posts) and which need the dietitian's own judgement, with prompts and data rules.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the iddsi.org framework page and the IDDSI Framework and Detailed Level Definitions 2.2 (level names, testing methods, choking-risk textures); food-regulator allergen guidance listing the 14 allergens, and a food-regulator list of nine major food allergens; all checked 4 November 2026. No clinical or dietary advice is given.