How Much Should I Spend on Groceries? Receipts vs the USDA Plan

The USDA grocery benchmark is a table. Every month the department publishes what a week and a month of food costs for a male aged 20 to 50, a female aged 20 to 50, a child aged 6 to 8, and about a dozen other age bands, at four spending levels from Thrifty to Liberal. The family-of-four figure quoted in every article on this subject is one row of that table, added up for one specific family.

Which means your household has its own row, and building it takes about a minute. Two adults in their thirties and one seven-year-old, on the Low-Cost plan, using the May 2026 release: $316.00 plus $274.80 plus $261.30, or $852.10.

Then comes footnote 2, where most of the internet stops reading. The published per-person costs assume the person lives in a four-person household. Smaller households buy in smaller quantities and pay more per unit, so USDA tells you to adjust after summing: add 20% for one person, 10% for two, 5% for three, nothing for four, and subtract 5% for five or six people and 10% for seven or more.

The same build at the other levels gives you the range this household is being measured inside.

A quirk of the two separate tables: Thrifty publishes its adult bands as 20-50 while the other three use 19-50, which is why the Thrifty male figure comes out a dollar higher than the Low-Cost one. Nothing is wrong with the arithmetic.

What the plan is actually counting

Every figure above assumes all meals and snacks are prepared at home. No restaurants, no delivery, no coffee bought on the way to work. The Thrifty basket also comes from 2013-16 consumption data and 2015-16 prices, reflated to current dollars with CPI, so it describes what a careful household bought a decade ago at this year's prices.

That premise collides with what American households do. The BLS Consumer Expenditure Survey for 2024 puts food at home at $6,224 a year and food away from home at $3,945, which works out to $518.67 and $328.75 a month. Food away is 38.8% of the food budget. Hold a grocery-only total up against a USDA plan and you are grading yourself against someone who spends nothing on that 38.8%, which is a flattering comparison rather than a useful one.

Sixty-three transactions and four receipts

The file is synthetic. I generated 63 card transactions covering May 1 to July 31, 2026 for a household of two adults (male 36, female 34) and a seven-year-old, then wrote itemized receipts for the four trips to stores that sell more than food. Running my own account would make the exercise ungradeable, since nobody could check the answer. A planted file is an exam with a key.

One thing in that file is unrealistic and I left it in. Every delivery and restaurant charge falls in May or June, so July shows zero food away from home, and the household's split lands at 89.1% at home against 10.9% away. The national split from the CE survey is 61.2% and 38.8%. I built the file grocery-heavy to load the sorting problem, which means the split figure the model reports describes my construction and not a plausible American household. Read the classification work rather than the ratio.

Five tasks in one prompt

You are a personal finance assistant. Attached: 3 months of card
transactions (grocery-3mo.csv) and itemized receipts for the
mixed-basket stores (grocery-receipts.txt). Household: two adults
(male 36, female 34), one child (age 7).

[USDA Cost of Food, May 2026, monthly, individuals in 4-person
households]
- Thrifty: Male 20-50 $317.10 / Female 20-50 $252.40 / child 6-8 $208.20
- Low-Cost: Male 19-50 $316.00 / Female 19-50 $274.80 / child 6-8 $261.30
- Moderate: Male 19-50 $397.00 / Female 19-50 $335.60 / child 6-8 $304.30
- Household-size adjustment: 3-person household = sum individuals,
  then +5%

Tasks:
1) Separate food-at-home from food-away-from-home and non-food. Do not
   trust the bank_category column blindly. Itemize what you excluded
   and why.
2) Compute the household's true monthly food-at-home average and the
   at-home vs away split.
3) Compute this household's USDA benchmark at all three plan levels
   (show the +5% adjustment) and position the actual average against
   them.
4) The beef line: is the increase price or volume? Show the math.
5) Flag anything ambiguous rather than silently deciding.

Pasting the USDA figures into the prompt is a choice worth explaining. A model asked to recall this table from memory will reach for whichever month it absorbed during training and give you a number that looks right to the dollar and is four months stale. Giving it the current figures moves the task from recall to arithmetic, and arithmetic is the part it does well.

The instruction about the bank category is the other load-bearing line. Bank categories come from a merchant-code lookup, and a model that treats them as ground truth will reproduce the bank's mistakes with more confidence than the bank had.

What the model returned

Checking it against the file

I rebuilt every figure in Python with exact decimal arithmetic against the same CSV and receipts before publishing any of it.

Two rows there are worth more than the seven matches above them.

The beef answer beat the key I wrote. I expected a sentence saying the per-pound rise is price inflation. What came back split the $7.32 increase in the receipt line into $2.26 of price on the original two pounds and $5.06 of volume on the extra half pound, which is the correct decomposition and the opposite emphasis from the one I had planned to report. It did leave an observation on the table: the June 14 receipt shows ground beef at $9.55 a pound, sitting between the May $8.99 and the July $10.12. Three points make the climb look steady, roughly 6% and then 6% again, rather than a single jump that might have come from a promotion ending.

What it never questioned was the comparison itself. It placed $822.45 of food at home against a plan built on the assumption that no food is bought anywhere else. Total food spending for this household, delivery included, is $922.74 a month, which sits $28.03 above the Low-Cost plan rather than $72.26 below it. The verdict turns entirely on whether you accept the plan's premise, and the model applied the benchmark exactly as instructed without ever asking what the benchmark was for.

What the bank called groceries

Add up every transaction the bank tagged Groceries and you get $2,333.94, or $777.98 a month. That number is $38.61 under the Thrifty plan, and a household reading it would conclude it spends less on food than the cheapest benchmark the government publishes. The corrected figure is $822.45, which is $5.86 over Thrifty.

The tag is wrong twice, in opposite directions. It pulls in $300.86 of restaurants and delivery that are not groceries, and it leaves out $434.28 of actual food sitting inside the Target and Costco charges it filed under Shopping. The two errors nearly cancel, leaving the monthly figure $44.47 light, which is small enough to look plausible and large enough to put the household on the wrong side of the line it is being measured against. Neither error announces itself anywhere on the statement, which is why the classification rule has to be written into the prompt rather than assumed.

Prices, or what you put in the cart

Food at home rose 2.7% over the twelve months ending June 2026, per the BLS release of July 14, 2026, against 3.4% for food away from home. USDA's Economic Research Service forecasts the same 2.7% for full-year 2026 within a 1.6% to 3.9% interval. So roughly 2.7% of any increase in your grocery total this year is the price level, and everything above that came from your cart.

That subtraction only works if your basket resembles the CPI basket, and 2026 makes that assumption uncomfortable. ERS forecasts beef and veal up 10.7% for the year, pork up 1.6%, poultry up 1.0%, and eggs down 27.9% year over year as of June. A household that eats a lot of beef and no eggs will run hot against the index for reasons that have nothing to do with how carefully it shops.

The file shows exactly that. Ground beef moved from $8.99 to $10.12 a pound in about ten weeks, a 12.6% climb that outruns even the annual beef forecast. One cut at one store across three receipts is not a price index, and the input I do not have is what the same beef cost at the other four grocery chains in the file, since those trips came with no itemization. The July CPI release lands on August 12, three days after this posts, and will move the food-at-home figure again.

The bulk-buying advice does not survive its own research

Standard grocery advice runs: plan meals, shop the sales, buy in bulk, waste less. Two of those hold up. The bulk one has a study pointing at it.

Pierre Chandon and Brian Wansink, writing in the Journal of Marketing Research in August 2002, worked through scanner data, a field study and two lab experiments and found that stockpiled products get consumed faster. Having more of something in the house raises how visible it is, which raises how often people eat it for convenient products and how much they eat for both convenient and inconvenient ones. Part of the per-unit saving on a multi-buy leaves through the pantry rather than staying in the budget.

The waste half of the warning is worse off. Nienke van Lin and colleagues, in the Journal of Consumer Research in December 2023, found that multi-unit promotions produced less household food waste than regular-price purchases, by 0.156 packages and 47.175 grams, both significant at p below .001. The same purchases produced more eating, up 0.545 packages, and more freezing, up 0.937 packages. Buying too much appears to trigger the behaviour that prevents waste.

Which leaves the less quotable version of the advice: the extra will not end up in the bin, it will end up eaten. That is a consumption leak, and no grocery total on a bank statement measures it.

What I would do with this number

I would leave the grocery total alone. At $822.45 against a $894.71 Low-Cost plan, this household is already shopping below the middle benchmark, and squeezing further means dropping the meal kit or the Whole Foods trips to save perhaps $40 a month. The line I would move is the $100.29 of delivery, and not because it is large. It is the only line in the file where the same food carries two prices, and it is the line the bank filed under Groceries, which means nobody in the household has ever seen it counted on its own. My threshold for acting on a spending category is whether anybody has looked at it yet, ahead of how big it is.

Run the exercise quarterly rather than monthly. USDA moves these tables by a few dollars a month and four weeks of receipts will not tell you anything new. The sorting is what changes the answer, and the sorting only has to be fixed once.

FAQ

How much should a household of my exact size and ages spend on groceries a month?

Build it from the USDA Cost of Food tables rather than reading a family-of-four headline. The tables give a monthly food cost for each person by sex and age band at four spending levels, and you add up the people in your household. For two adults aged 19 to 50 and a child aged 6 to 8, the May 2026 Low-Cost figures are $316.00, $274.80 and $261.30, which sum to $852.10. Then apply footnote 2, which is the step almost every article skips: the published per-person costs assume the person lives in a four-person household, so USDA tells you to add 20% for a one-person household, 10% for two, 5% for three, nothing for four, and subtract 5% for five or six and 10% for seven or more. A three-person household at $852.10 becomes $894.71. The same build gives $816.59 on the Thrifty plan and $1,088.75 on the Moderate plan. USDA republishes these tables monthly, so pull the current month rather than copying a figure out of an article.

Does the USDA food plan include restaurants and takeout?

No, and this is the assumption that decides whether your comparison means anything. Every USDA plan level assumes all meals and snacks are prepared at home, with nothing spent on restaurants, delivery or convenience food. The Thrifty basket is also built from 2013-16 consumption and 2015-16 price data, reflated to current dollars with CPI, so it describes a careful decade-old shopping pattern at this year’s prices. Meanwhile the BLS Consumer Expenditure Survey for 2024 puts the average household at $518.67 a month on food at home and $328.75 on food away from home, meaning 38.8% of food spending happens outside a grocery store. Compare a grocery-only total to a USDA plan and you are grading yourself against a household that spends nothing on that 38.8%. In my test file the food-at-home average came to $822.45, which is $72.26 under the Low-Cost plan. Add the $100.29 a month of delivery and restaurants and total food spending is $922.74, which is $28.03 over it.

How do I separate groceries from household goods and takeout on a mixed receipt?

You cannot do it from the card charge, so give the model the itemized receipt for any store that sells more than food. A $163.27 Target charge in my file contained $60.97 of laundry detergent, school supplies, bath towels and non-food tax, and no arrangement of the transaction row reveals that. A $214.88 warehouse-club charge contained a $54.99 annual membership fee, which is a subscription rather than food. Grocery-store gift cards are a third case: my file had a $31.10 one, and stored value is not food consumed this month. The other half of the job is undoing what your bank did. Bank categories come from a merchant-code lookup, and in my file all six delivery charges and both restaurant charges were filed under Groceries. Tell the model in the prompt not to use the bank category as its classification rule, and ask it to itemize every exclusion with a reason, because the reason is the part you can check line by line against your own receipts.

Is my grocery bill rising because of prices or because of what I am buying?

CPI gives you the price baseline and the rest is behaviour. Food at home rose 2.7% over the twelve months ending June 2026, per the BLS release of July 14, 2026, and USDA’s Economic Research Service forecasts the same 2.7% for the full year within a 1.6% to 3.9% interval. If your grocery total is up more than that, the extra came from your cart rather than from the price level. The rule only holds if your basket resembles the CPI basket, and 2026 is a year where that assumption strains: ERS forecasts beef and veal up 10.7%, pork up 1.6%, poultry up 1.0%, and eggs down 27.9% year over year as of June. A household that eats a lot of beef and no eggs will run hot against the index for reasons that have nothing to do with discipline. In my file ground beef went from $8.99 to $9.55 to $10.12 a pound across three receipts, a 12.6% climb in about ten weeks, which outruns even the beef forecast.

Disclaimer

This article is an educational explainer, not financial or nutritional advice, and it recommends no store, service or spending level. The three-month transaction file and the four receipts behind the test run are synthetic: I generated all 63 rows and wrote the receipt lines, so no figure in them describes a real household. The at-home versus away split in the file was built grocery-heavy on purpose and does not represent a typical American household. The AI output is a real run on August 7, 2026 on gpt-5.6-sol through the Codex CLI, and every number in it was rebuilt with exact decimal arithmetic against the same files before publication; a different model, prompt or day will produce different output. USDA Cost of Food figures are the May 2026 release from a page last updated June 18, 2026 and are revised monthly. CPI figures are from the June 2026 release published July 14, 2026, with the July release due August 12, 2026. ERS forecasts are the July 2026 vintage and carry published intervals. Check the current month's USDA tables and your own receipts before drawing conclusions about your household.

For the same statement read a different way, How to Audit Your Bank Fees With ChatGPT runs the planted-file method on overdraft and ATM charges, and How to Find Forgotten Subscriptions on Your Bank Statement covers the recurring side, including the meal-kit and membership charges that turn up in this file. If you want the price half of the question rather than the sorting half, Calculate Your Personal Inflation Rate With AI weights CPI categories by your own spending, and I Uploaded My Credit Card Statement to ChatGPT starts from a card export instead.

Sources

• USDA Food and Nutrition Service, Cost of Food monthly reports, May 2026 release, page last updated June 18, 2026 (Thrifty reference family $235.00 a week and $1,018.20 a month; per-person costs by age band; footnote 1 on the 2013-16 consumption and 2015-16 price basis; footnote 2 household-size adjustments of +20%, +10%, +5%, 0%, −5%, −10%): https://www.fna.usda.gov/research/cnpp/usda-food-plans/cost-food-monthly-reports
• USDA, Cost of Food at Home at Four Levels, May 2026 (Low-Cost $316.00 / $274.80 / $261.30 and Moderate $397.00 / $335.60 / $304.30 for the bands used here): https://www.fna.usda.gov/sites/default/files/resource-files/cnpp-costfood-3levels-may2026.pdf
• USDA, Thrifty Food Plan monthly report, May 2026 ($317.10 male 20-50, $252.40 female 20-50, $208.20 child 6-8): https://www.fna.usda.gov/sites/default/files/resource-files/cnpp-costfood-tfp-may2026.pdf
• Bureau of Labor Statistics, Consumer Price Index news release for June 2026, published July 14, 2026 (food at home +2.7% year over year, food away from home +3.4%, all items +3.5% with a −0.4% monthly change): https://www.bls.gov/news.release/cpi.nr0.htm
• USDA Economic Research Service, Food Price Outlook, July 2026 forecast (all food +3.1%, food at home +2.7% within 1.6-3.9%, food away from home +3.5%; beef and veal +10.7% within 7.2-14.6%, pork +1.6%, poultry +1.0%, eggs −27.9% year over year as of June 2026): https://www.ers.usda.gov/data-products/food-price-outlook/summary-findings
• Bureau of Labor Statistics, Consumer Expenditures 2024 news release (food $10,169 a year, food at home $6,224 and food away from home $3,945, food 12.9% of total expenditures): https://www.bls.gov/news.release/cesan.nr0.htm
• Pierre Chandon and Brian Wansink, "When Are Stockpiled Products Consumed Faster? A Convenience-Salience Framework of Postpurchase Consumption Incidence and Quantity," Journal of Marketing Research 39(3), August 2002, 292-303 (stockpiling raises consumption incidence for convenient products and consumption quantity for both convenient and inconvenient ones): https://journals.sagepub.com/doi/10.1509/jmkr.39.3.321.19111
• Nienke van Lin, Aylin Aydinli, Marco Bertini, Erica van Herpen and Julia von Schuckmann, "Does Cash Really Mean Trash? An Empirical Investigation into the Effect of Retailer Price Promotions on Household Food Waste," Journal of Consumer Research 50(4), December 2023, 663-682 (multi-unit promotions reduced waste by 0.156 packages and 47.175 grams, both significant at p below .001, alongside +0.545 packages eaten and +0.937 packages frozen): https://academic.oup.com/jcr/article/50/4/663/7071580