An honest comparison
Can ChatGPT Write a Schedule of Works From Construction Drawings?
Partly — and on a small job it may be all you need. Scopa runs on a general AI model too. The difference is not the model; it is the 1,076 lines of estimating rules around it, the way a whole drawing set is handled, and the fact that what arrives is three files rather than a conversation.
We have not benchmarked anybody's assistant. This page compares what each is built to do, and says where it cannot speak for the other product.
In short
A general AI assistant can read construction drawings and list work from them, which is often enough for one or two sheets or for understanding what a schedule of works looks like. Scopa is a purpose-built tool for the same job on a full set: it runs a 1,076-line construction estimating specification on every run, takes the whole drawing package at once, applies a fixed output structure — standard sections or NBS (CAWS) — and returns a spreadsheet, a CSV and a classified list of what the drawings left unclear, by email, for £49. Neither prices the work, and both produce a draft that a professional still has to check.
What a general assistant is genuinely good at
Worth saying before anything else, because it is true and because a comparison that pretends otherwise is not worth reading.
- Reading a drawing and telling you what is on it. Modern models do this well.
- Conversation. You can argue with it, ask why an item is there, and have a section rewritten three ways in a minute. Scopa cannot do any of that — it produces a document and stops.
- One or two sheets. If that is the job, the overhead of a specialist tool buys you nothing.
- Costing you nothing extra, if you already pay for it.
- Everything else in your week. It is a general tool and that is the point of it.
Scopa runs on a general AI model as well
There is no secret model. If the question behind "why not just use ChatGPT" is "am I paying £49 for a prompt", that is a fair question and it deserves a straight answer.
What you are paying for is everything around the model. A 1,076-line estimating specification across 66 sections, 18 of them setting rules for one part of the work at a time, from preparation through to fixtures and fittings. A pipeline that decides how each document in a set should be read. A fixed output contract, so the document does not change shape between jobs. And delivery as files a contract manager can price from.
A prompt you retype, and adjust, and forget half of when you are busy, is not the same thing as a specification that runs identically whether or not you remembered to ask.
What the rules actually say
Four you would have to know to write into a prompt yourself
These are real rules from the specification Scopa runs. They exist because each one is a way a schedule goes wrong, and none of them is obvious unless you have priced work from drawings.
The engineer takes precedence
Where the structural engineer's drawings and the architect's describe the same element differently, the item is written from the engineer's and the conflict recorded. It matters most on foundations, beams, lintels and floor construction, where the architect's drawing is often a generic detail and the engineer's is what gets built.
Do not invent scope
Only work that is drawn, written, clearly necessary, or confirmed by you. No adding ground beams, pad foundations or steelwork on the off-chance — an item that might be there is worse than a confirmation asking about it.
Do not trust a generic note
A note reading "blockwork foundation cavity wall" on a set whose walls are timber frame is a standard detail somebody never tailored. Copying it faithfully is not a defence. The construction type is confirmed from the drawings before any foundation or wall item is written.
Ignore the calculation pack
Structural calculations are read for their conclusions, member schedules and recommendations only. The engineering workings are skipped — they are long, they cost money to read, and they do not produce tender items.
Side by side
Six questions, and what each tool answers
The criteria are set out rather than assumed. Where a fact belongs to somebody else's product — upload limits, data retention, what you pay — it is left for you to check, because it changes and it is not ours to state.
| Criterion | A general AI assistant | Scopa |
|---|---|---|
| Does the output have a structure you can rely on? | It produces what the conversation produces. Ask twice, phrase it differently, and you can get a different shape. | A fixed contract. The same section order every run, or the CAWS arrangement, with unused sections omitted and the order never changing. |
| Does it take a whole drawing set? | Upload limits and file handling vary by plan and change over time — check the current terms for yours. | Built for it. The published example was 17 PDFs and 29 MB, uploaded whole, with drawings read as images and long written documents read as text. |
| Does it apply construction-specific rules? | Only the ones you write into the prompt yourself, every time. | A construction estimating specification with 18 section-by-section rule sets runs on every job, whether or not you think to ask for it. |
| What do you actually receive? | A conversation. Getting it into a spreadsheet is your next job. | Three files by email: the schedule as a spreadsheet, the same as CSV, and the confirmations as a PDF. |
| What happens to your client's drawings? | Depends on your account settings and the current terms of the service — worth reading before uploading a live client set. | Deleted the moment the schedule is emailed, with a nightly sweep for runs that died part-way. |
| What does it cost? | Whatever you already pay for it. No per-job charge. | £49 a job. No subscription, and runs do not expire. |
When not to buy a run
A page like this is supposed to sell you something. It is worth more if it also tells you when not to.
- You have one or two sheets. Use whatever assistant you already have.
- You want to explore, argue and rewrite. That is a conversation, and Scopa does not have one — it produces a document and stops.
- You want the work priced. Neither tool should be doing that, and Scopa will not.
- You want a programme. A schedule of works is not a time schedule, and there are no dates or durations in the output.
- The drawings barely exist yet. Nothing can describe work that has not been drawn.
Evidence
Run the comparison yourself
The most useful thing this page can do is stop arguing and hand you the output. Take a set you know well, put it through whatever assistant you use, then read the published example against it.
- 17
- drawings in
- 29 MB
- uploaded whole
- 105
- line items to price
- 27
- confirmations raised
- 8 min
- £49 a run
The worked example reproduces two sections in full, lists the drawing conflicts it found between the architect's and the engineer's information, and lets you download the actual spreadsheet, CSV and confirmations PDF. It is a domestic extension set that nobody tidied up first.
What this page cannot tell you
We have not run a scored comparison. Doing it properly would mean putting the same sets through both, having a quantity surveyor mark the results, and publishing the method — and until that exists, any claim about which produces the better schedule would be marketing rather than evidence.
General assistants also change constantly. Anything stated here about another product's upload limits, file handling, retention or price would be out of date before long, so those are left for you to check against the current terms rather than asserted.
What is on this page is the structural difference — what each tool is built to do — plus a published example you can hold against whatever you get elsewhere. That seemed more useful than a table of ticks.
Questions people actually ask
- Can ChatGPT write a schedule of works from construction drawings?
- It can produce something. General AI assistants read PDFs and images and will happily list construction work from a drawing. For one or two sheets, or to get a feel for what a schedule looks like, that may be all you need and it costs you nothing extra. The difference shows on a full set, where the job is less about reading one drawing and more about cross-referencing a package, applying a consistent structure and being told what the drawings never settled.
- Is Scopa just ChatGPT with a prompt on top?
- Scopa runs on a general AI model — that part is fair, and worth saying plainly. What sits around it is the difference: 1,076 lines of estimating rules across 66 sections, a pipeline that decides how each document should be read, a fixed output contract, and delivery as real files. A prompt you retype each time is not the same thing as a specification that runs identically on every job.
- Why does a fixed structure matter?
- Because a schedule of works gets priced by somebody else. If the sections move around between jobs, or an item appears under a different heading than last time, whoever is pricing it has to re-learn the document every time. The standard order runs from Preparation to Fixtures & Fittings and does not change; unused sections are dropped rather than reordered.
- Have you benchmarked Scopa against ChatGPT?
- No, and this page does not claim to. Comparing them properly would mean running the same sets through both and having a quantity surveyor mark the results, which we have not done. What is set out here is the structural difference — what each is built to do — rather than a scored test. The published worked example is there so you can judge the output yourself.
- Can I just try both?
- Yes, and it is the sensible thing to do. Put a set you already know well through whatever assistant you use, read the published worked example, and see whether the difference is worth £49 on your jobs. If it is not, do not buy runs.
- Does Scopa price the work? Does ChatGPT?
- Scopa does not — no rates, no quantities, no totals, and the PC sum and price columns in the CAWS output arrive empty on purpose. A general assistant will give you a number if you ask it for one, which is a reason to be careful rather than a feature.
- What about confidentiality?
- Scopa holds a client's drawings only for the few minutes a run takes and deletes them when the schedule is emailed. For any general service, retention depends on your account settings and its current terms — read them before uploading a live client's set, because that is somebody else's commercial information as well as yours.
Read the example before you decide
If it is not better than what you already get for nothing, do not buy a run. If it is, it is £49 and about 8 minutes.
An architect? There is a page written for practices. Issuing to a quantity surveyor? The NBS (CAWS) output is set out in full.