Can ChatGPT or Claude price an M&E tender? What works and what doesn't

Short answer: Not on its own. A general chatbot can read drawings and give you numbers quickly, but it doesn't know how your firm measures, what the consultant's BQ looks like or what your rates are, and it rarely tells you which sheet a number came from. Set up properly, with your rules, your BQ layout and your estimator checking every flagged line, it can draft the takeoff so your estimator spends their time checking and pricing instead of counting.

What happens when you upload tender drawings to ChatGPT?

Most M&E firms that have tried this tell a similar story. Someone drops a lighting layout PDF into ChatGPT or Claude and asks how many lighting points are on the floor. The answer comes back in seconds, often as a range, sometimes with a cable length and even a total cost in RM.

Then the estimator looks at it and asks three questions. Which sheet did that come from? Which revision? Whose rates? The chatbot can't answer clearly, so the estimator counts the whole floor again by hand. Nothing was saved.

That isn't because the AI is useless. It's because the question was asked with no context. A general chatbot has never seen your BQ, doesn't know that you count a twin socket as one point or two, and doesn't know that the consultant measures trunking by the metre with bends enumerated separately. It fills those gaps with guesses.

What can AI actually do well on an M&E takeoff?

Used with clear instructions, current AI models are good at the repetitive parts of a tender that eat an estimator's week:

Sorting the pack. Listing every sheet in a tender set with its number, title and revision, so you know what's current before anything is counted.
Counting symbols against a legend. Lighting points, socket outlets, data points, FCUs, exhaust fans and DBs, floor by floor, with the sheet recorded on each line.
Reading schedules. Pulling FCU capacities, DB ways and ratings, and equipment tags from schedules into a structured list.
Following a BQ layout. Putting quantities into the consultant's item numbering and units, instead of a loose list you retype later.
Comparing revisions. Listing what changed between Rev A and Rev B of a sheet when an addendum arrives.

Where does a chatbot go wrong on M&E drawings?

The weak spots are predictable, which is useful, because predictable weak spots can be flagged instead of trusted.

Task	Chatbot on its own	Set up properly
Counting points on a lighting layout	Gives a number, often with no sheet reference	Counts per sheet, records sheet and revision on each line
Measuring cable and trunking runs	Estimates lengths, may ignore scale notes	Measures where scale is reliable, flags sheets marked "not to scale"
Clashing drawings	Picks one and carries on	Marks the line for checking, with the reason
Your measuring rules	Has to be re-explained every chat	Built into the setup once
Pricing	Invents rates or uses generic ones	Uses your own rate sheet only, after sign-off

A few specific traps worth knowing about:

Scale. Measured runs depend on reliable scale. Scanned or reduced PDFs, and sheets noted "not to scale", produce lengths that look precise but aren't.
Small symbols and dense sheets. On a busy power layout, symbols overlap text and grid lines. A count can be off by a few points without any warning.
Things that aren't drawn. Containment between floors, cable to equipment shown only on a schematic, or items covered by a note in the specification. A chatbot counts what it sees.
Confident totals. The most dangerous output is a neat RM figure. It reads as finished work when it's a guess built on guesses.

Can AI price the BQ, or only count?

It can apply rates, but only yours, and only after your estimator has checked the quantities. Pricing is where your margin, your supplier terms and your read of the main contractor live. No AI model knows what your copper cable costs this month, what your installation gang achieves per day or how keen you want to be on this job.

What a setup can do is take your own rate sheet and apply it line by line in the consultant's BQ layout, so the priced BQ is ready for your review. The decision on the final number stays with you.

What makes the difference between a chatbot and a working setup?

The model is the same. The difference is everything around it:

Your measuring rules written down. How you count points, how you measure runs, what allowances you add, how you treat bends and accessories.
Your BQ layout and the consultant's. Item numbering, units and descriptions the output must follow.
A rule to show the source. Every quantity carries its drawing sheet and revision, so it can be checked in seconds.
A rule to flag rather than guess. Anything unclear, clashing or unscaled is marked with a reason, not smoothed over.
A sign-off step. Your estimator reviews marked lines and approves before anything is priced.
Proof on past work. Run it on tenders you've already priced and compare, before you rely on it for a live closing date.

This all runs inside your own Claude or ChatGPT business account. Your team opens it in a browser. There's no new software to learn.

Should we just write better prompts ourselves?

You can, and some firms do. The honest catch is time. Writing down how your firm measures takes longer than people expect, because much of it lives in your senior estimator's head. Then you need to test it on several past tenders, find where it differs from your own BQ, adjust, and test again. Most busy subcontractors don't have a quiet fortnight to do that between closing dates.

If you do try it yourself, start with one trade and one drawing type, such as lighting layouts only, and compare against a BQ you've already priced before you trust anything.

FAQ

Is ChatGPT or Claude better for M&E takeoff?

Both can do the work when set up properly. The choice usually comes down to which one your team already uses and which business plan suits you. The setup matters far more than the brand.

Will AI replace our estimator?

No. It takes the counting and retyping off their desk. They still check flagged lines, set rates and decide the price. Most firms would rather use that time to price more tenders.

Can it read scanned or hand-marked drawings?

Often, but quality matters. Clean PDFs from the consultant work best. Poor scans and hand mark-ups are more likely to be flagged for checking, which is the right outcome.

Does it work for ACMV and solar as well as electrical?

Yes. FCU layouts, ducting and refrigerant piping, and solar panel layouts with inverter and DC cabling all follow the same approach: count, measure, show the sheet, flag what's unclear.

Try it on a tender you've already priced

Send one past tender. We return a draft BQ with every quantity traced to its sheet, so you can compare it with your own. Free, under NDA.

Book a free BQ test