Measure the shop, give an AI assistant the dimensions and fixture list as text, and ask for three layout options. Test the best one on the floor with masking tape, then check it against fire exits and accessibility before moving anything heavy. AI supplies options; the tape and the checklist do the proving.
A chat assistant is a good brainstorming partner for arrangement and a poor surveyor. It will cheerfully put a rack across a fire exit, and it reads scale badly from photographs: Anthropic's vision documentation says the positions and coordinates Claude reads from an image are approximate and should be verified before you rely on them, and the same caution suits any assistant. So the dimensions go in as text, the limits go in as a written rules sheet, and nothing moves until the layout has passed on the floor.
Measure first: the one-page shop sheet the AI works from
Everything the assistant produces is only as good as the sheet you give it. Measure with a tape or a laser measure and write down the walls, the openings and everything fixed, then list each fixture with its footprint. The running example here is a small gift shop, 5.0 metres wide and 4.0 metres deep (20 square metres). The figures are illustrative.
| Item | What to record | Example |
|---|---|---|
| Room | Width, depth, ceiling height | 5.0 m wide (front wall), 4.0 m deep, 2.6 m high |
| Entrance | Position, width, swing direction, hinge side | Front wall, 1.0 m from the left corner, 1.0 m wide, swings in, hinged on the left |
| Fire exit | Position, width, swing direction | Right wall, 1.0 m wide, 0.5 m from the back corner, swings out |
| Stockroom door | Position, width, swing | Back wall, 0.8 m wide, 3.5 m from the left corner, swings into the stockroom |
| Fixed items | Radiators, meters, sockets, fire equipment, signs | Extinguisher beside the fire exit; sockets at the till and mid-back wall |
| Fixtures | Length by depth by height, and how many | Wall units 1.5 by 0.5 by 1.8 m, six of them; till counter 1.2 by 0.6 by 0.9 m; one round table 1.2 m across; two gondolas 1.2 by 0.5 by 1.4 m |
Photograph the hand-drawn plan as well if you like, but as a second source, never the only one. The next step tests how well the assistant has understood it.
Have the assistant read the plan back before it designs anything
Ask for a numbered restatement of the walls, openings and fixtures, with the usable floor area worked out. You are checking comprehension, and mistakes here are far cheaper than mistakes in a layout.
Read the SHOP SHEET below. Before suggesting any layout, restate it as a numbered list: each wall with its length, each opening with position, width and swing direction, each fixed item, and each fixture with its size and quantity. Then calculate the clear floor area if the six wall units run along both side walls and the back wall, leaving every door clear (the units are 0.5 m deep). Do not design anything yet.
SHOP SHEET:
[paste the sheet]A typical read-back, illustrative, contains two or three slips like these.
- The fire exit on the wrong wall. The assistant placed it on the left, probably from a photographed sketch drawn from the other side. Correct it before going on.
- The wrong swing. It said the entrance opens outwards. It swings in, which means the first metre inside the door must stay free.
- Forgotten depth. It calculated the clear floor as 5.0 by 3.5 m, ignoring the wall units on the sides. The correct clear floor is 4.0 m wide (5.0 less two 0.5 m units) and 3.5 m deep (4.0 less the back run), which is 14.0 square metres.
Correct each error in writing and ask for the restatement again until every line matches your sheet. Only then ask for ideas.
Write the rules sheet before asking for ideas
The limits come from people and documents, not from the AI. Write them at the top of the prompt so the assistant has to design inside them. The table lists where each number comes from; the stand-in values are only for the sums in this tutorial.
| Rule | Where the real figure comes from | Stand-in used here |
|---|---|---|
| Clear route to each fire exit | Your fire risk assessment, or whoever carried it out | 1.0 m wide, nothing in the way |
| Door swing zones | Measured arcs of every door | Entrance arc 1.0 m; keep it clear |
| Minimum clear aisle | Your accessibility guidance, adviser or landlord | 0.9 m |
| Turning space at the till and any fitting room | The same source | Your figure |
| Extinguisher, call point and exit sign access | The fire risk assessment | Visible and reachable from the route |
| Fixings and window rules | Your lease, landlord or insurer | No new wall fixings |
| Sockets and cables | Your electrician | Nothing across an aisle |
Do not accept a figure from the assistant for any of these. It will give you one confidently, and it may not apply to your premises. If you do not have a fire risk assessment or an accessibility figure, getting them is the first job, not an AI task.
Decide what the layout has to achieve, in order
An assistant asked for "a better layout" returns something generic. Give it goals, ranked, because every layout trades one goal against another. For the gift shop, the owner writes the list below in priority order and pastes it under the rules sheet.
- Every customer sees the new-in table within three steps of the door.
- Cards and gift wrap sit together, because the shop sells them together.
- Small, easy impulse items sit at the till, where the queue stands.
- The back wall carries one focal display, visible from the entrance.
- Staff can see the door and the till queue from the till.
- Restocking from the stockroom does not cross the customer route.
Each goal can be tested after the move: the first by standing at the door and counting steps, the fifth by sitting at the till. The ranking matters because an island in the centre serves goal 1 well and goal 6 badly, while a table to one side does the opposite. Without a ranking, the assistant tries to satisfy everything and quietly satisfies nothing.
Three layout options from one prompt
Now ask for options that differ in logic, not only in details, and ask the assistant to say what each one gains and loses.
Using the corrected SHOP SHEET and the RULES SHEET below, propose three different layouts for the clear floor area. Each must differ in its organising idea (for example an island, a spine, a feature table).
For each: where every fixture goes (position in metres from the front-left corner), the customer route from the entrance, the narrowest gap between fixtures and between a fixture and a wall unit, what it gains and what it loses.
Do not tell me whether a layout complies with any rule; list the numbers I need to check and leave the checking to me.
SHOP SHEET: [paste]
RULES SHEET: [paste]The last instruction matters, because left to itself the assistant writes "this layout meets standard safety requirements" and means nothing by it. With the clear floor at 4.0 by 3.5 metres, here are three options in the form the AI returns them, with the gaps worked out.
| Option | Central feature | Gaps at the sides and front and back | Narrowest | Against 0.9 m | Route to fire exit |
|---|---|---|---|---|---|
| A: island | 2.0 by 1.0 m island, centred | 1.0 m and 1.25 m | 1.0 m | Passes, but tight | Clear via the right-hand aisle |
| B: spine | 3.0 by 0.6 m run down the middle | 1.7 m and 0.25 m | 0.25 m | Fails | Left half has no clear route to the fire exit |
| C: feature table | 1.2 by 1.2 m round table, centred | 1.4 m and 1.15 m | 1.15 m | Passes | Clear |
The gaps are subtraction: the clear width less the feature, halved. For option A that is (4.0 less 2.0) divided by 2 for the sides, and (3.5 less 1.0) divided by 2 for front and back. Option B looks attractive on paper, with a long display run and wide side aisles, and it fails twice: the 0.25 metre gaps at each end are not passable, and because the fire exit is on the right, anyone on the left of the spine can reach it only by squeezing through one of those gaps. That side depends on the front door alone, which breaks the rules-sheet line about a clear route to each fire exit. This is the kind of option an assistant offers happily and the arithmetic removes in thirty seconds.
A and C both pass, though these sums cover only the central feature. Run the same subtraction for the till counter and the two gondolas wherever each option puts them, because on 14 square metres the smaller pieces are often what pushes an aisle below 0.9 metres. A gives more display surface and a tighter aisle; C gives breathing room and less display. The choice is yours, and it comes after the floor test. Our comparison of AI store design tools covers dedicated layout software if you want to draw these properly, and AI floor plan generation for architects shows why generated plans keep failing real constraints.
Prove it on the floor: the tape test and the layout checklist
Mark the footprint of every moved fixture with masking tape, using the coordinates from the option you chose, and put cardboard boxes at the heights of the tall units. Measure the narrowest gaps with a tape instead of trusting your eye. Walk it with a trolley or a pushchair, open every door fully, film a walk-through on your phone, and time yourself from the furthest corner to the nearest exit. Then work through the checklist and tick each row on the floor, not on the screen.
| Check | Test on the floor | Pass looks like |
|---|---|---|
| Fire exits | Walk from every corner to an exit | A clear route from everywhere; nothing in the swing of exit doors; exit signs visible |
| Fire equipment | Stand where someone would look for it | Extinguisher, call point and panels visible and reachable |
| Aisle widths | Measure the narrowest gap in each aisle | At or above your own rule, not rounded up |
| Turning space | Turn a trolley or pushchair at the till | A full turn without moving a fixture |
| Reach | Reach the top and bottom shelf of each unit | Best-selling items between knee and shoulder height |
| Floor and cables | Look along every route | No loose mats, leads or thresholds you could catch a wheel on |
| Sightlines | Stand at the till | You can see the door and the busiest corner |
| Queue | Line up three people at the till | The queue does not block the main route |
| Restocking | Carry a box to each bay | No display has to be moved to refill it |
| Cleaning | Push a mop around the new positions | Every floor area can be reached |
| Stability | Check each tall unit against its instructions | Anchored as instructed; nothing heavy stored high |
A layout passes only when every row passes. If one fails, change the layout and run that row again. The AI's opinion counts for nothing in this table.
Plan on about half a day for the whole exercise: roughly 45 minutes to measure and write the sheet, 30 minutes with the assistant for the read-back, the rules and the options, and two hours for the tape test and the checklist, ideally with a helper to hold the other end of the tape. The only spend is a measure, masking tape and some cardboard boxes. The heavy lifting comes after the layout has passed, which is the point.
Display ideas for one bay: constraints first
Display ideas work best at the scale of a single bay, with its real dimensions. Give the assistant the bay, the products and the limits, and ask for options you can build in an afternoon.
Suggest three ways to display candles in one wall bay 1.5 m wide, with five shelves at 0.3 m spacing, 0.35 m deep.
Stock to show: 24 large jars (0.10 m wide, 0.12 m tall), 30 small jars (0.07 m wide, 0.08 m tall), 12 gift sets in boxes (0.20 by 0.12 by 0.15 m).
Budget for props: $40. We already own: two wooden crates, a cake stand, a small mirror.
For each option: which items go on which shelf, how many fit, the sign wording, and the time to set it up. Do not suggest anything that needs wall fixings.A reasonable result offers a tiered stack with large jars at eye level, an arrangement by scent family with small signs, and a gift-set shelf at eye level beneath a mirror to bounce light. Check the numbers before using any of it. A 1.5 metre shelf holds at most 15 large jars in a single row (1.5 divided by 0.10, with no gaps), so a draft that lines all 24 up along the eye-level shelf has miscounted; they need two rows or two shelves. Check, too, that the heaviest items are on the lower shelves, that the sign wording is yours and true (no "bestseller" without a reason), and that the setup time is realistic for your staff.
Window displays: three concepts a season, with props and a budget
A window brief works the same way: dimensions, stock and constraints first, concepts second. The window in this example is 1.8 metres wide and 1.1 metres deep, with a shelf at 0.4 metres, and a budget of $60.
Give three window concepts for an autumn change in a homewares shop. Window: 1.8 m wide, 1.1 m deep, 2.2 m high, one glass shelf at 0.4 m, strong afternoon sun on the glass.
For each concept: the theme in five words, a prop list with sizes (use only what we own or can buy for under $60 in total), which 7 products from our range to feature, a lighting note, sign wording, and a change plan for week 2.
Do not feature products we do not stock. Do not put anything heavy on the glass shelf.An illustrative first answer proposes a cosy-corner scene, a harvest table and a "gift wrapped" stack. The fixes are the usual ones.
- Too many products. The harvest table lists fourteen items for a window that comfortably holds seven. Cut to the seven you most want to sell.
- Props at the wrong scale. A "two-metre branch" is longer than the 1.8 metre window is wide, and stood upright it fills almost the whole 2.2 metre height, crowding the shelf display.
- Sun damage. Candles and printed fabrics fade or warp in strong afternoon sun, so move them to the interior display and keep the window for items that can take it.
- Weight on glass. The draft puts a cast-iron pot on the glass shelf. Move it to the floor of the window.
- Stock you do not hold. The assistant invented a "hand-thrown ceramic jug". Replace every product with a real one from your range.
Mood boards: words first, then pictures, then your own samples
A mood board helps when you are briefing a painter, a fitter or a friend with a van. Build it in three steps and keep it honest about what it is.
First, ask the assistant for six mood words, a palette described in plain colour names, materials, lighting and a "to avoid" list, based on your products and customers. Second, make six to eight reference images with an image tool you already have. Gemini, for example, creates and edits images, and its overview page lists a free image model alongside a more advanced one on a paid Google AI plan. Claude reads images but does not make them: Anthropic's vision page says it is an image understanding model only. Midjourney and Adobe Firefly both offer style-reference features, though Firefly's work only with Adobe's own image models. Third, assemble the images in Canva, Google Slides or PowerPoint, and add real material samples and paint swatches beside them.
The board sets a tone, not a specification. AI images show lighting you cannot install and fixtures that do not exist, so the dimensioned plan stays the thing you build from. For internal planning the ownership questions are minor; before using an AI image in an advert or on your window, read the tool's terms. OpenAI's and Anthropic's, for example, assign you their rights in an output only "if any". Our tutorial on briefing a designer with an AI mood board goes further on the brief itself, and the guide to a floor plan and display brief for your fitter covers the handover.
Check whether the new layout earned its place
A new layout is a hypothesis. Move one bay at a time, leave the prices and promotions alone for that period, and compare three weeks before with three weeks after, using your till data. Also compare against the same weeks last year, and against a category you did not move. A weekly sheet from till data is easy to build; see our tutorial on a retail KPI dashboard in Excel.
Here is a sum, with illustrative numbers. Candles sold 3.1 units a day in the three weeks before moving them to the front bay and 4.0 a day in the three weeks after. Last year, the same weeks rose 10% on the weeks before, which is the season at work. So the expected figure without any change is 3.1 times 1.10, or 3.41. The lift beyond the season is 0.59 units a day, about 17% over expectation. Over 26 trading days that is 15 extra candles; at $12 each and a 45% margin, around $83 a month in gross profit.
Two cautions keep that honest. Three weeks of 3.1 a day is only about 56 sales over 18 trading days, and ordinary chance can move a count that small by around its square root, roughly 7 or 8 either way, so a 17% lift is promising rather than proven. Repeat the test with a second category, or run three more weeks, before you rearrange the whole shop on the strength of it. And look beyond sales: note whether customers ask where things have gone, whether the queue at the till still clears, and how long restocking takes. A layout that sells 17% more candles and doubles the restocking time may not be a win. If you want visitor counts by area, the footfall camera tutorial covers what they count and the privacy limits.
Questions shop owners ask about AI and store layout
Should I tell regular customers before moving things round?
Yes, briefly. A small sign by the door for the first two or three weeks, saying what has moved and where, saves regulars hunting for their favourites, and staff can walk anyone who asks to the new spot. Keep a tally of how often people ask where something has gone: if the questions have not tailed off after a fortnight, the new position is not working.
Which tool should I use to draw the floor plan?
Paper or a spreadsheet grid is enough, with one cell standing for 50 centimetres. For a drawn plan, Floorplanner's pricing page describes its free plan as free for casual use, with up to five projects and watermarked standard-definition exports, so check its terms for business use. Whatever you draw in, keep the dimensions in text as well.
Can AI render my shop to show a shopfitter?
Image tools can produce atmosphere pictures, not accurate scaled drawings, and they often invent fixtures and lighting that do not exist. Send the fitter your dimensioned plan and the mood board together, and treat any AI picture as tone rather than specification.
How often should a window display change?
There is no universal rule, so choose a rhythm you can keep. A theme change every four weeks with a small refresh at the halfway point is a workable start. Then note whether customers mention the window, ask about featured items or buy more of them, and adjust the rhythm to what you see.
Further reads
- Retail Staff Scheduling: Build a Rota From Hourly Till Sales With AI — Use hourly till sales to staff the busiest hours of the new layout.
- Can AI Design a Restaurant? Layouts, Concept Images and Limits — The same layout-and-mood approach for a dining room.
- How Interior Designers Use AI for Mood Boards and Presentations — How professionals build and present AI mood boards.
- Examples of AI in Retail: Eight Jobs a Small Shop Can Copy — Eight other jobs a small shop can hand to AI.
- How AI Can Help in Sales: A Gift Shop's Bundle and Follow-Up Plan — Bundles and follow-ups once the layout is working.
- AI Mirror for a Clothing Store: Is a Smart Fitting Room Worth It? — A small-store payback sum for an AI smart mirror, the vendor claims to test, camera privacy questions and cheaper ways to get the benefit.
- AI Software for Retail Stores: Five Checks Before You Buy — Five pass-or-fail checks for any AI product a shop is about to buy, with a scoring sheet and sample vendor questions.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Anthropic's vision documentation (image support, spatial-reasoning limits, no image generation), Gemini's image generation overview page and Floorplanner's pricing page, all checked October 2026. Dimensions, sales figures and rule values in the examples are illustrative stand-ins.