It depends on your till. Toast IQ only works for Toast customers and Square AI only for Square sellers, so the real choice is your point-of-sale assistant or a spreadsheet with ChatGPT. Built-in assistants are quicker for routine sales questions; the spreadsheet works with any till and lets you check every figure before you act.
An "AI food insights tool" search will mostly show you food-trend products for big brands. A single restaurant's question is more practical: which items earn their place, which hours are slow, and what to change this week. For that, the tools worth comparing are the ones that already hold your sales.
What each option is, from the vendors' own words
Toast IQ
Toast IQ is the AI assistant inside the Toast platform. Toast announced its conversational version in late October 2025, describing a restaurant operator asking questions in plain language and getting answers and suggested next steps. The company's examples included top-selling items by daypart, adding a menu item from a single prompt, and staffing questions, and it says the assistant can use data from the Toast platform as well as context such as local weather and events. It can also take actions in the platform, for example updating a menu or editing shifts. Toast says it is available through the Toast Now mobile app and Toast Web, and a Toast article notes that availability may depend on your subscription, permissions and account set-up. Toast's own product page, which I fetched in October 2026, says the base assistant is included with Toast POS at no additional cost, and sells two bundles on top, Toast IQ Grow (an AI marketing agent with human support) and Toast IQ Team (a labour platform with an AI agent); it lists no prices for those two. It also says actions need your confirmation and that offerings may vary by region. Check whether it is on for your account, and which team members have the rights to use it. If you are weighing those tiers, Toast IQ Team or Toast IQ Grow goes through them.
Square AI
Square's AI page labels its "Ask anything about your business" feature "Beta" and describes questions about best-selling and slow-selling items, sales over time, and local weather, events and reviews. It also mentions related tools such as AI voice ordering, booking management through an assistant for appointments, an ordering guide that compares ingredient prices, and content generation for item descriptions and marketing. The page I fetched did not state a cost. Square's help article, which I also fetched, calls the Dashboard assistant "Managerbot" and says that account owners or team members with full-access permissions can use it from the compose box on the Dashboard homepage. It states that the assistant uses only data in your Square account and publicly available web information, that it can make mistakes and that you should check its answers. Because the two pages use different names, look at which one you actually see in your Dashboard, and confirm its status and any charge with Square.
A spreadsheet with ChatGPT
This is the do-it-yourself route: export a weekly or monthly sales report from the till as a CSV, open it in Excel or Google Sheets, and ask an AI assistant to analyse it, or ask it to write the formulas and pivot tables. It works with any till. It needs a business-grade account if you want a contractual commitment on training and data, and for a small team a paid seat on ChatGPT Business Standard lists at $25 a user a month or $20 billed annually. The sibling tutorials on using ChatGPT with Excel and AI prompts for Google Sheets cover the mechanics.
The four questions to put to each tool
A fair comparison asks the same questions of each, on the same period, and checks the answers against a number you trust. These four cover the decisions a restaurant owner makes most often, and they get progressively harder.
- "Which five items sold most by revenue at dinner last month, and how did they change from the month before?" A plain ranking with a comparison. It tests whether the tool can filter by daypart and compare periods correctly.
- "Which items have fallen the most in units sold over the last eight weeks?" A trend question. It tests whether the tool reads a trend or just compares two endpoints.
- "On which days of the week was labour cost the highest share of sales?" A cross-source question. It needs sales and labour data together, which a till assistant may hold and a plain sales export may not.
- "What would you change on the menu this month, and why?" A judgement question. It tests whether the tool explains its reasoning from the data or produces generic advice.
Run them in your own account. I have not stated what any of the assistants answered, because that depends on your data, plan and the current version, and answers I could not verify would be invention.
A scoring sheet for the comparison
Score each answer 0 to 2 on four things: Was the figure right when checked? Did it say where the number came from, including the date range? Was it fast enough to use during a busy week? Could it do something useful with the answer, such as export, chart or act on it? Here is a filled-in illustration of how a result might look. The scores are invented to show how to use the sheet, not a verdict on any product.
| Question | Built-in assistant (example) | Spreadsheet and ChatGPT (example) |
|---|---|---|
| 1. Top five by daypart | 7 out of 8 | 8 out of 8 |
| 2. Falling items | 5 out of 8 | 6 out of 8 |
| 3. Labour share by day | 6 out of 8 | 3 out of 8 |
| 4. Menu change advice | 4 out of 8 | 5 out of 8 |
| Total (max 32) | 22 | 22 |
In this made-up result, the tie hides a useful pattern: the built-in assistant wins on the cross-source labour question because it holds both sets of data, while the spreadsheet wins on the plain ranking because you can see the formula. That is a realistic shape for the result, and it suggests a hybrid: use the assistant for quick questions, and a spreadsheet to check anything that changes a menu price or a rota.
The spreadsheet route, step by step
- Export item-level sales for the last 8 to 12 weeks, with date, time, item, category, quantity and net sales. Leave out guest names and payment details.
- Clean the file: one header row, no merged cells, consistent item names ("Margherita" and "Margherita Pizza" are different rows to a computer). The older tutorial on cleaning messy data covers this.
- Add a daypart column with a formula, for example lunch before 16:00 and dinner after.
- Ask the assistant to analyse the file, giving it the exact question and the date range, and ask it to show the formula or steps it used.
- Check one number by hand with a pivot table or a SUMIFS formula.
A prompt and an illustrative reply for question 1:
Prompt:
The attached file is item-level sales for a 45-cover restaurant, weeks
1 to 8. Columns: date, time, item, category, qty, net_sales.
Dinner = orders from 16:00. For the latest 4 weeks, rank the top 5 items
by net_sales at dinner, then compare each with the previous 4 weeks.
Show the formula or steps you used and flag any item whose name appears
in two spellings.
Illustrative reply:
Item Last 4 wks Prior 4 wks Change
Ribeye 300g $4,860 $4,410 +10.2%
Seabass $3,990 $4,120 -3.2%
Mushroom risotto $2,880 $2,300 +25.2%
House burger $2,760 $2,790 -1.1%
Lamb shank $2,420 $2,980 -18.8%
Note: "House Burger" and "House burger" appear as separate items;
I combined them.What I would fix: confirm the merge of the two burger spellings was right (it changes the ranking), and check the lamb shank drop, since a seasonal menu change would make the comparison unfair. Then verify one row by hand: filter dinner sales for ribeye in the last four weeks and add up the net sales. If it does not match $4,860, find out why before trusting the rest.
Question 3 in practice: labour as a share of sales
The labour question deserves its own walk-through, because it is where a till assistant and a spreadsheet differ most. Suppose last week's figures, invented for illustration, were as follows.
| Day | Net sales | Labour cost | Labour share |
|---|---|---|---|
| Monday | $1,100 | $520 | 47% |
| Tuesday | $1,250 | $520 | 42% |
| Wednesday | $1,600 | $560 | 35% |
| Thursday | $1,900 | $600 | 32% |
| Friday | $3,100 | $780 | 25% |
| Saturday | $3,600 | $820 | 23% |
| Sunday | $1,450 | $500 | 34% |
Total sales are $14,000 and labour is $4,300, or about 31 percent, which matches the café example later. The assistant's answer should name Monday and Tuesday as the days with the highest labour share. A good answer goes one step further and says that a skeleton crew of two instead of three on Monday would save a fixed amount, though only you know whether the kitchen can run that way. The spreadsheet route needs a second file: labour hours or cost by day from your scheduling or payroll tool. If you do not have that file, the spreadsheet cannot answer question 3 at all, which is a fair reflection of why a platform that holds both sources has an edge here.
Whichever route you use, ask the tool to show the days it considered. If it reports a Monday figure of 47 percent and your own sums say 47 percent, you can trust the method on this question, at least for this week. Then test another week, ideally one with a holiday in it, which tends to break date logic.
A twenty-minute weekly routine that uses either route
- Minutes 0 to 5: Ask for last week's top five and bottom five items by net sales, with the dates in the question.
- Minutes 5 to 10: Ask for the labour share by day, and compare against your target (many owners work to a figure of their own; use yours).
- Minutes 10 to 15: Ask for any item that moved by more than 15 percent against its recent average, and decide if the cause is known (a special, a weather swing, a stock-out).
- Minutes 15 to 20: Write down one decision, one thing to check and one question for next week. Check one number by hand.
The habit matters more than the tool. An owner who spends twenty minutes a week asking and checking will spot a drifting item or a bloated Tuesday rota within a month, whichever tool they use. An owner with a clever assistant and no routine will not.
Where each option wins
| Criterion | Toast IQ | Square AI | Spreadsheet and ChatGPT |
|---|---|---|---|
| Works with any till | No, Toast only | No, Square only | Yes |
| Holds your live sales data | Yes, in the platform | Yes, in the account | Only what you export |
| Can act, not just answer | Vendor says it can update menus and edit shifts | Help page says it can take some actions and can be undone | No; it advises, you act |
| Shows how it got the number | Check in the product | Check in the product | Yes, you see the formula |
| Combines sales with invoices or reviews | Within the platform's data | Account data plus public web information | Yes, if you bring the files |
| Stated cost | Base assistant included with Toast POS; Grow and Team bundles unpriced on the page | Not stated on the page I fetched; the ask-anything feature is labelled beta | Seat price of the assistant, such as $20 a month annual |
| Effort per question | Lowest | Lowest | Highest, until you build a template |
Where "check in the product" appears, that is deliberate: I would rather tell you to look than state what I could not see. During your trial, open the answer and see whether the assistant shows its date range, filters and data source.
A worked choice for a small café-restaurant
Imagine a 45-cover owner-run site on Square with one manager. Weekly sales about $14,000, labour about $4,300 (31 percent), food cost about $4,200 (30 percent). The owner spends two hours each Monday pulling reports to decide what to promote and how to staff the week, about 100 hours a year. If the assistant halves that, 50 hours saved a year at $20 an hour is $1,000. Set against a $20-a-month assistant seat, which is $240 a year, that is a comfortable margin, and an assistant that carries no extra charge would be a clearer win still.
But the owner's trickiest decision is whether to drop three weak items. The assistant says they are the bottom sellers. The spreadsheet shows two of them are bought mostly alongside a best-seller, so removing them would reduce basket size. That second view came from checking in a pivot table. So the sensible set-up is: the assistant for the Monday questions, a monthly spreadsheet review for menu changes. Neither replaces the other, and the extra monthly hour is cheap insurance against a wrong cut.
Permissions and privacy
- Who can ask. Square's help article says full-access team members and account owners can use its assistant, with only owners and admins able to manage the apps it creates. Toast says availability depends on subscription and permissions. Decide who in your team should be able to ask, and especially who can let it change menus or shifts.
- Actions need a safety net. An assistant that can edit a menu or a shift can make a wrong edit faster than a person. Test actions on an unused item first, and check that you can undo them. Square's help article says most actions can be undone by asking for it, but restoring an earlier version of one of its apps does not reverse changes already made.
- Exports. A spreadsheet that has left the till lives in your email and downloads folder. Keep it in a shared drive with limited access, and delete old exports.
- Staff data. Labour questions involve pay rates. Keep named individuals out of any file that goes to a general chat tool. If you have staff or guests in the EU, data-protection law applies, so ask a qualified adviser about the final arrangements.
A mistake that looks like a right answer
Suppose an owner asks, "What was my best-selling item last week?" and gets a confident answer: the house burger. The week shown was in fact the previous one, because the assistant interpreted "last week" as the last full Monday-to-Sunday period while the owner meant the last seven days. The burger was top in both weeks, so nothing seemed wrong, but the ranking below it differed and the owner moved a special on the wrong evidence. The fix is a habit: always state the dates in the question ("1 to 7 March"), and look for the date range in the answer before using it.
What these tools cannot see
- Why customers did something. A slow item may be badly described, poorly placed on the menu or simply unlucky in a wet week. The data gives a signal, not a reason.
- Sales that never reached the till. Cash tabs, comps and voids appear if your staff use the system properly, and not at all if they do not.
- Your food cost. Unless your recipes and invoices are loaded, "profitable items" are based on sales only. A big seller with a thin margin looks great on a revenue ranking.
- Reviews and word of mouth. For these you need a different source; see using AI to analyse your reviews.
Following up each insight-led decision after a month
Pick one decision made from each route and follow it up. Did the promoted item rise? Did the staffing change reduce the labour share without hurting service? Keep a log of the questions asked, the answer given, whether you checked it and what happened. After a month, count how many answers were right when checked, how many changed a decision, and how many hours you saved. If the assistant was right on nine out of ten checked answers, relax the checking to one a week; if it was right on six out of ten, keep verifying everything and consider the spreadsheet your main tool. For the next step, forecasting covers and cutting food waste with AI uses the same sales history.
Questions restaurant owners ask about AI sales insights
Can I use Toast IQ if I run Square, or the other way round?
No. Each assistant works on the data in its own platform, so it only helps if your till is that vendor's. If you are choosing a point-of-sale system, do not decide on the assistant alone; check hardware, fees, payments and support first, since the assistants are one feature among many.
Do these assistants cost extra?
Toast's product page says the base Toast IQ assistant is included with Toast POS at no additional cost, while the Grow and Team bundles are separate and unpriced on that page. Square's AI page lists no price, so ask Square directly. Get any answer in writing before you rely on it.
Is the spreadsheet route safe for guest data?
It can be, if you export only sales data. Item sales by hour, labour hours and invoice totals need no guest names, phone numbers or card details. Strip anything personal before upload, and use a business-grade account where the provider says it does not train on your content by default.
Why not trust the assistant's answer without checking?
Both vendors say their assistants can make mistakes, and a plausible answer on the wrong date range looks just like a right one. Check any figure you will act on against a report or a quick spreadsheet total. Checking the first five answers teaches you how far you can trust it.
Further reads
- How to Build a Restaurant Staff Rota With AI Demand Forecasts — Use the same sales data to draft a rota.
- Restaurant AI Tools: 12 Questions to Ask Before You Sign Up — Questions to put to any restaurant software vendor.
- Retail KPI Dashboard in Excel: A Weekly Sheet From Till Data With AI — A weekly till-data sheet built in Excel.
- How Much Does AI Cost a Small Restaurant Each Month? — The monthly cost of the AI tools a restaurant might add.
- How to Use ChatGPT for Restaurants: Five Jobs to Hand It Each Week — Other weekly jobs the spreadsheet-and-ChatGPT route can take.
- Restaurant Menu Price Increase: Test Each Change by Dish With AI — Test a price change by dish once the sales data is clean.
- Best AI for the Food Industry: Picks for a Cafe, Caterer or Takeaway — Six AI picks for a cafe, caterer or takeaway: costing, menu copy, allergen records, supplier orders, phone answering and POS-built AI.
- Food Truck POS: Offline Mode, Fees and Which AI Features Pay Off — Square, Toast Go and Clover compared for a truck: offline limits, card fees worked through on real sales, and the AI features worth paying for.
- Free AI Tools for Restaurants: 7 to Try and Where Free Stops — Seven free AI tools for restaurants by job: menu copy, graphics, posts, sales questions, training, enquiries and a second opinion, each with its free limit.
- Restaurant Gift Card Programme: Terms, Offers and Tracking With AI — Set up a gift card programme for a small restaurant: terms to write, offers that work, POS options and an AI-assisted tracking routine.
- 7 Restaurant Upselling Techniques: Train Servers With AI-Built Scripts — Seven upselling techniques, each with a table-side script you can build with AI, a pre-shift huddle sheet and a way to check average spend.
- Recipe Costing Spreadsheet for Restaurants: Build It With ChatGPT — Five tabs, one salmon dish costed line by line, and a ten-minute weekly step that updates every plate cost when an invoice changes.
- Restaurant Inventory Management With AI: Par Levels and Orders — A weekly count sheet, a par for every item and an order draft for your supplier, built with AI from your own counts and POS sales, plus when paid software pays.
- Restaurant Labour Cost Percentage: Test Rota Changes With AI First — A what-if rota test from last month's takings and payroll: baseline the percentage, cost the change and find the break-even sales loss, with AI doing the sums.
- Square AI Voice Ordering for Restaurants: Set Up and Test (Beta) — The Dashboard steps, the plan you need and a seven-day test-call schedule to run before Square's AI answers your restaurant phone.
- AI Automation for Restaurants: Back-of-House Jobs to Start With — Prep lists first, supplier order drafts second, kitchen rotas last: a ten-week rollout with the checks that keep each AI draft honest.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Square's AI page and Square Dashboard help article (fetched October 2026), Toast's Toast IQ product page (October 2026) and its October 2025 announcement as reported in trade press, OpenAI business plan pricing. Check every plan, availability and price on the vendor's own page.