chatgpt construction takeoff, ai takeoff, supplier quotes, trade accounts, distributors
Your customers are pasting your quotes into ChatGPT
November 10, 2026
Buyers may now check a quote and plan a takeoff with ChatGPT. What an assistant can and cannot see on your quote, a 25-run test, and what to print.
Executive summary
ChatGPT can do a simple construction takeoff from dimensions you type in, but it struggles to measure drawings and cannot see your stock, your trade prices or the job. Your customers may already be using it on your quotes, so a quote showing only a price gives an assistant the least to check.
A building supply dealer told me that customers now drop a quote into ChatGPT and ask whether they are being ripped off. The same dealer sees many competing quotes on a single job, so the customer has plenty to paste.
That is one dealer's account, and I am treating it as an early warning. Other trades I talk to have not raised AI with me at all. But the check takes seconds and costs nothing, and my view is simple: on any job with scope, a bare price is now the weakest quote you can send.
A price tells the assistant what you charge. Everything else on the page tells it what the price buys.
Can ChatGPT do a takeoff and check your quote?
Yes for simple jobs described in words, and much less so from drawings. Checking a quote is easier for it than measuring a plan, because a quote is text and arithmetic.
OpenAI's own vision documentation says the model "struggles with tasks requiring precise spatial localization" and "may give approximate counts for objects in images." Reading a scaled plan calls on both skills.
Users report the same split. On OpenAI's forum, one poster wrote in December 2024 that ChatGPT "doesn't handle construction drawings particularly well", yet handled simple takeoffs quickly once given dimensions and material costs. Another, in January 2025, reported "very good results for Plumbing", with accuracy falling on PDFs where lines overlap. Those are forum posts, not measurements.
For you, the weakness matters less than it sounds. Your customer does not need a correct takeoff to walk in with one. A detailed material list on a phone is enough to change the conversation, right or wrong.
How many buyers are really doing this?
Nobody has measured it for trade accounts, as far as I can find. I looked for a primary source on customers using AI to check a supplier's quote, and I did not find one.
What exists is broader, and none of it isolates quote checking:
Forrester's November 2024 report summary says 89% of B2B buyers have adopted generative AI. The detail sits behind a client login, and "adopted" does not mean "checked a quote".
A September 2025 NBER working paper found ChatGPT had reached about 10% of the world's adults by July 2025, with more than 70% of messages unrelated to work. Practical guidance, seeking information and writing made up nearly 80% of conversations.
The NBER paper is the one I would hold on to. The closest category it has to a price check is shopping: questions like "How much are Nikes?" count as Purchasable Products, inside Seeking Information. A piece written for Scrunch, a vendor in this space, calls the larger version a vendor bake-off: one prompt, one structured comparison, something to show the boss.
So treat this as a may. Your customers may be doing it. Preparing costs little, and nobody is obliged to tell you.
I found one piece of research on how sellers respond, and it should be read carefully. Cui, Li and Zhang ran a field experiment with a trading company, published in 2022, before ChatGPT existed. Suppliers quoted chatbot buyers higher wholesale prices than human buyers. Signaling a smart recommendation system reduced the quote to chatbot buyers. That was a different market and a different tool, so I take only one thing from it: who a seller thinks is checking can change the quote.
What can an assistant actually check on your quote?
It can check what is printed. It can compare against what is public. It cannot see your warehouse, your price sheet or the customer's job. In the test below, all 20 quote runs declined to confirm a market price.
Start with price, where two facts weaken an assistant's answer. The first is timing. BLS producer price data for softwood lumber rose 6.4% from June to July 2026, then fell 4.0% by September. Copper wire and cable rose 30.8% from September 2025 to September 2026 (recent months are preliminary, and the percentages are my calculation). Against that, the knowledge cutoff OpenAI lists for GPT-6.1 Sol is April 30, 2026, and Anthropic lists June 2026 for its current Claude models. An answer from memory is an answer from months ago.
Web search helps with that, but it brings the second problem: search finds public listings. An electrical distributor's account price on 12/2 cable, or a safety supplier's contract price on gloves, is not on any page the assistant can read. The test had no web search, so it could not measure this. But 6 of the 10 electrical runs told the customer to compare with a big-box store or local retailer as well as another electrical supplier, and none of the 6 explained that a trade price and a shelf price are different things.
Code has the same shape. Minnesota Rule 1303.1600 sets minimum footing depth at 5 feet in Zone I, which includes St. Louis County, and 3 1/2 feet in Zone II, unless an engineer supports less. One state, 18 inches apart. None of the five takeoff runs below named either depth; all five said to confirm it locally.
This table comes from how these tools work and from the test in the next section, which ran without web search.
The test: two made-up quotes, two versions each, five runs
The test sent the same total to a model twice, once as a bare price and once itemized, and scored what came back. Everything in it was made up, including the prices and the jobs.
The prompt for the quotes was the question the dealer described: "A supplier sent me this quote. Am I being ripped off?"
Bare deck quote: 12 x 16 deck package, $3,121.70 plus tax. Delivery Tuesday.
Itemized deck quote, same total: 13 pcs 2x10x12 treated, ground contact, at $31.40, $408.20. 3 pcs 2x10x16 treated at $44.00, $132.00. 3 pcs 6x6x10 treated posts at $62.00, $186.00. 27 pcs 5/4x6x16 treated decking at $21.50, $589.50. Hangers, screws and ledger fasteners, $412.00. Railing kit, 3 sides, $1,299.00. Curbside delivery Tuesday, $95.00. Total $3,121.70. All in stock on Nov 2. Price held 14 days. Not included: concrete, footing forms, flashing, permit.
Bare electrical quote: 10 spools 12/2 NM-B, 250 ft, $1,285.00 plus tax. Ships Friday.
Itemized electrical quote, same total: 10 spools 12/2 NM-B copper, 250 ft, at your account price of $128.50 a spool, $1,285.00. 10 in stock on Nov 2. Copper-based item, price held 7 days. Delivered free on our Friday route.
The itemized deck quote carried a planted error: 27 boards at $21.50 is $580.50, so the printed line is $9 too high. Its exclusions were planted too. The takeoff prompt added one fact that could be checked: "I'm building a 12 x 16 deck attached to my house, about 3 feet off the ground, in Duluth, Minnesota. Do a material takeoff I can order from. Include waste." Duluth is in St. Louis County, so the answer to check was the 5-foot footing.
Each prompt ran five times, 25 in all, each a new conversation with no shared history, on November 2, 2026. This ran through OpenAI's API on GPT-6.1 Sol at default settings, without web browsing or the ChatGPT app, so a customer's experience in the app may differ. The traps were set before the run, and the same party that set them scored the answers. The counts for lump sums, "account price", the stock line and the big-box comparison were added after reading the responses, so treat those as observations and not as planned measures. Five runs of one made-up quote show what this model did with that quote, not a rate for yours.
Not one of the 20 quote runs called the supplier a rip-off. The model did well on the traps: it caught the $9 error and read the exclusions every time, and it held back a Duluth footing depth instead of getting one wrong, though three runs priced concrete on piers 60 to 66 inches tall, each labeled as an example only. Its overreach was quieter: it called an invented electrical price plausible in 5 of 5 runs with no source, while saying in those runs that it had not confirmed local prices. The electrical results are mixed. The bare quote was called plausible in 5 of 5 but asked about freight in 5 of 5. The itemized one was called plausible in 0 of 5 and drew doubt about "account price" in 5 of 5, a phrase only that version carried, yet its price hold was called normal or plausible in 5 of 5 and its free delivery noted as worth counting in 5 of 5. On the deck, the bare quote was judged to have too little detail in 5 of 5, while the itemized one had its unit prices called unalarming in 4 of 5 and its two lump sums questioned in 5 of 5. I read that as support for itemizing a job with scope, and as no case against a one-line price on a simple commodity order.
You can run this on your own business in one sitting. Take your last three quotes, strip the customer's name and address, paste each one with that prompt, and read the answer as your customer would.
What should a quote carry now?
A quote should answer the questions a stranger would ask, because the first reader may be a stranger.
- Lines with quantity, unit price and extension, so the arithmetic can be checked.
- The spec that sets the price: grade, treatment, listing, brand, or "or equal" if you allow it.
- A date, how long the price holds and why, particularly on copper, lumber or anything else that moves.
- Stock status on that date, by line if it differs.
- Delivery method, day and charge, or a plain "pickup".
- What is excluded and what you assumed, such as the footing depth or the job's location.
- A name and a direct number, so the next question comes to you.
Lines 6 and 7 do the most work. An exclusion turns "you left out the concrete" from a catch into something you already said. The name gives the customer somewhere to go with what the assistant told them.
No website, app or AI is needed for any of it. If you are wondering whether this is a reason to go online, it is not: a new store will not fix a quote that says only a number. A better printed form or a better email will. If your ERP's quote layout cannot show these fields, that is the first thing to change.
Where does a fuller quote cost you?
It costs you in three places, and for some customers the bare price is still right.
It is easier to shop. Every unit price you print is one a competitor can beat line by line, and with many quotes on a job, some customers will do exactly that.
It takes longer. A rep turning quotes all day will not add seven fields by hand, so this only works if the fields come from your system by default.
And some customers do not want it. A trade account reordering the same item has already decided, and a one-line price with a date is the respectful answer. An industrial supplier told me its buyers know what they need and do no takeoffs. For that business, the takeoff half of this article does not apply, and "not yet" is a fair answer to the rest.
Should you answer with takeoffs of your own?
Yes if you serve construction trades and customers are asking, because the assistant may be setting a new floor for what a quote includes. No if your buyers order by part number.
A dealer told me that younger customers now ask for more service, such as a takeoff, where their parents asked only for a price and a date. A customer who can get a rough material list from a phone for free will expect you to do at least that well, and to know what the phone does not: what you stock, which local rules apply, and what you would substitute.
A takeoff you do for free can be pasted into a chat window and shopped, along with your quote. I would still do it for customers worth keeping, with your assumptions on it, because a list with assumptions is harder to compare.
Staff using an assistant on their own store is a separate subject, covered in the undo button post, and why trust decides these purchases is in my piece on commoditization.
My rule: if a customer might compare the quote with anything, send the lines, the date, the stock, the delivery and the exclusions. Keep the one-line price for someone reordering what they already know.
Where ToolSwift fits
I run ToolSwift, so I have a stake in this. Its Takeoff Agent, trained on your rules and product data, turns a takeoff document into a draft quote matched to your catalog, for staff review. In ToolSwift quotes, your team checks the lines and pricing and sends the quote, the customer accepts from their account, and the accepted quote converts into an order. None of this is needed for the seven lines above, and a spreadsheet and someone who knows the trade will also do a takeoff. If you want takeoffs and quotes handled in one place, book a demo.
Sources
Related in ToolSwift
Quotes built into your store
Create, assign and track quotes, and turn approved quotes into orders.