Can ChatGPT do a takeoff? Not in a chat. With Estimator+, yes

So, can ChatGPT actually do a takeoff?

In a chat window on its own, no, not one you can price. In Codex in the ChatGPT desktop app, with Estimator+ connected to your tender folder, yes. It's the same model either way. What changes is where the number comes from.

You've probably tried it already. You drop the Level 2 lighting plan into ChatGPT, ask how many downlights are on it, and get a number back before your coffee's gone cold. Tempting, right?

And ChatGPT isn't useless on drawings. It'll read a sheet, find the finishes legend and tell you what a code means. That's handy. What a chat can't hand you is a quantity with its evidence attached: the sheet, the revision, the scale, and a shape someone else can look at. A reply carries none of that. Just a number.

You're the one who answers for every quantity in a tender, too. Subbies push back on the ones that look wrong, which is fair enough, it's their job. If you can't show the working for a number, it isn't safe to carry.

What do OpenAI and Anthropic say about it?

Both companies publish the limits of their models' vision in their own docs. OpenAI's vision guide says: "The model may give approximate counts for objects in images." It adds: "The model struggles with tasks requiring precise spatial localization."

Anthropic's vision documentation says much the same about Claude, that it "can give approximate counts of objects in an image but might not always be precisely accurate, especially with large numbers of small objects."

Now picture a reflected ceiling plan. That's a large number of small objects pretty much by definition, and so's a hydraulic layout. Both companies are being upfront: a model looking at a picture gives you its best read of the count, and you'd still have to check it.

Here's where a typed count bites you

Let's run an example. The figures are made up, and money's ex GST. Say you give a chat the Level 2 lighting plan and ask how many DL1 downlights it shows.

1. The reply says "approximately 140 DL1 downlights".
2. The plan shows 152.
3. At A$185 each, supplied and installed, the 12 missing fittings are 12 × A$185 = A$2,220.
4. The electrician's quote allows for 152. Your figure says 140, and nothing shows you which one's right.

Honestly, the A$2,220 is the smaller problem. The bigger one is finding the 12, which means counting all 152 symbols yourself. By then you've done the takeoff anyway.

Here's the annoying part. A right answer wouldn't leave you any better off. If the reply had said 152, you still couldn't trust it without doing that same count.

What changes when ChatGPT works through Estimator+?

Set it up in Codex in the ChatGPT desktop app, with the Estimator+ connector added and your tender folder open. Now ChatGPT isn't squinting at one picture you pasted in. It reads the whole tender first, the spec, schedules, notes, contract and every addendum, then runs the takeoff with AI Takeoff V5.

The big change is that it never types a quantity. It marks the drawing instead: a dot on each fitting, a line along each wall, an outline around each floor finish, all on a published sheet. Estimator+ works out the number from those shapes and the sheet's printed scale, and the scale is proven on the sheet itself.

So a count is the number of dots, and a length comes from the line and the scale. There's no total for anyone to make up.

Do the model's limits vanish? No, and we'd be kidding you if we said so. They become visible. A dot on the wrong symbol or a wall line off its wall is right there on the drawing to fix. In Review mode every shape carries its line's name on the plan, and a symbol with no dot stands out.

Back on that lighting plan. The line says 152, and there are 152 dots on the sheet, each sitting on a symbol. Then the electrician's quote comes in allowing for 146. You don't recount the floor. You send them the marked-up plan. Much shorter conversation, that one.

Does Claude work the same way?

Yep. Claude runs the same takeoff in the Claude desktop app, in its Code tab, with the same connector and the same tender folder. Same read of the whole tender, same shapes, same rule that the AI never types a number. Use whichever one you already pay for. The AI runs on your own Claude or ChatGPT plan, so there's no AI bill from us.

Setup's four steps and you never open a terminal. Connect Claude or ChatGPT to estimating software walks through it, and AI takeoff software covers everything the takeoff reads before it measures.

What about ChatGPT in the browser?

Still useful. ChatGPT in a web browser can work on the estimate, the quotes and the pricing with Estimator+. What it can't do is the takeoff, because a browser has no tender folder to work in. For that you need the desktop app with the folder open.

And plain ChatGPT, with no Estimator+ at all, is fine for reading the text on a sheet: notes, legends, title blocks, what a detail's showing. Just don't carry its counts into a tender.

The honest catch

You still check the takeoff. Where the documents leave something open, the AI assumes, says so, and asks you the open questions together at the end of the run. The price, the quote picks and the strategy stay with you.

It works from PDF drawings, not CAD or BIM models. And Estimator+ is a paid seat, A$450 per seat per month ex GST, billed monthly with no contract.

Want a fair test? Grab a lighting plan you've already counted, bring it along when you book a demo, and check the AI's dots against your own count.