Is ChatGPT Good for Electrical Estimating? Use It to Check Bids

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Is ChatGPT Good for Electrical Estimating? Use It to Check Bids.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for Is ChatGPT Good for Electrical Estimating? Use It to Check Bids.

ChatGPT is not good at pricing electrical work, but it is useful for checking a bid you have already built. It can spot likely omissions, question labour hours against ranges and recompute arithmetic. It does not know your site, your rates or your crew, so never submit its numbers unchecked.

The forum arguments on this question usually split between "it changed my life" and "it made up a price". Both are right, because they describe different jobs. Asking it to produce a bid hands it work it cannot do. Asking it to cross-examine a bid you wrote hands it work that rewards its strengths: wide recall of what usually gets forgotten, patient arithmetic and tireless questioning.

Follow me on Instagram@sagnikteaches

Where it fails, and why it fails there

Electrical estimating blends three kinds of information: quantities from drawings, unit costs and hours from your history, and judgement about the building. A chat assistant has none of the three unless you supply it.

Connect on LinkedInSagnik Bhattacharya
  • It does not see the building. Ceiling void depth, existing containment, a long cable run to a remote board and the difficulty of working in an occupied space decide hours. A description in a prompt is a poor substitute for a site visit.
  • It does not know your rates. Material prices change, supplier discounts are yours alone, and labour rates include your overheads. Any figure it offers is a plausible number, not your number.
  • It produces confident ranges. A range of "0.3 to 0.8 hours per socket" looks like data. It is a spread wide enough to hide a loss or lose the job.
  • It can slip on arithmetic and on long lists. It may drop a line or mis-total silently, so reconcile its sums against your own.

Those limits shape the way to use it: as a reviewer, with your numbers in the prompt, and with every observation checked against something you trust.

Subscribe on YouTube@sagnikteaches

A sample bid to check

An illustration with invented figures. A small office fit-out of about 1,800 sq ft. The estimator has built this labour schedule from takeoff counts and their own unit hours:

ItemQtyHours eachHours
Double socket outlets, surface or flush400.624.0
LED panel luminaires240.512.0
Data points (cabling only)60.754.5
Distribution board, install and terminate18.08.0
Emergency light fittings80.54.0
Total52.5

At a labour rate you set, say $60 an hour, that is $3,150 of labour, plus materials and margin. Two things should already make an experienced electrician uneasy: there is no line for testing, and 52.5 hours for 1,800 sq ft of fit-out looks light. Those instincts are what the prompts below are meant to confirm or challenge.

The Complete Excel Course with AI Integration thumbnail

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

Prompt one: ask for omissions, not for a price

Give the assistant the job description and the item list, and tell it to list what is missing. State that it must not add prices.

I am an electrical contractor checking my own bid. Do not price anything.
Job: office fit-out, about 1,800 sq ft, existing building, occupied floors above, suspended ceiling.
My labour schedule: [paste table].
List items typically needed on a job like this that are NOT in my schedule.
Group them as: testing and paperwork, builder's work and making good, access and temporary works,
supervision and travel, and risks to confirm on site. One line each, with the question I should ask.

An illustrative reply, trimmed: under testing and paperwork it would list inspection, testing and certification of the new circuits, as-built drawings and handover documents. Under builder's work: core drilling or chasing, fire-stopping where cables pass through fire walls, and making good. Under access: a tower or steps for ceiling work, a temporary supply, out-of-hours working if the building is occupied. Under supervision: travel time, site induction time, waste removal. Under risks to confirm: whether the existing board has spare ways, the condition of existing containment, and whether any fire alarm interface is required.

What you would do with it: tick each item against your bid, add the real ones with your own hours, and cross out any that do not apply. A reply of ten suggestions will usually contain three that matter and seven that do not. That is a good return for ninety seconds, as long as you do the sorting. Where an item raises a regulation question, such as testing requirements, check the wiring rules and the contract; the assistant will not know which edition or requirements apply to your work.

The related tutorial on setting fitting allowances and provisional sums covers what to do with items you cannot yet price, such as the fire alarm interface in this example.

The sample bid after the omissions check

Here is how the same illustrative schedule might look once the real omissions are added, using the estimator's own hours for each. The estimator decided which suggestions applied, so the additions are deliberately only a subset.

Added itemBasisHours
Inspection, testing and certificationOwn past jobs: about 6 per cent of install hours, rounded up to the next whole hour4.0
Fire-stopping and making good at penetrationsSix penetrations at 0.5 hours3.0
Tower access and set-up for ceiling workOne day of access handling3.0
Site induction, travel and waste removalFour visits at 0.75 hours3.0
As-built drawings and handover packOwn record from similar jobs2.0

Those five lines add 15 hours, moving the schedule from 52.5 to 67.5 hours, an increase of 28.6 per cent. At $60 an hour the labour moves from $3,150 to $4,050, a $900 difference that would have come out of margin. The figures are invented, but the pattern is real: the work you forget is the paperwork, access and clean-up around the visible fittings. Notice that each added line has a basis the estimator owns. The assistant never supplied an hour.

A second sample shows how the list changes by job type. For a domestic rewire of a three-bedroom house, the omissions an electrician might get back are different: lifting and relaying floorboards, a consumer unit upgrade, bonding to gas and water, smoke and heat alarm interlinks, redecoration allowance (usually excluded), and a temporary supply for the occupants during the works. The assistant reuses the same categories but fills them differently, which is why the prompt includes the job type and building condition. If you leave those out you get a generic list, and generic lists are the ones to ignore.

Prompt two: sanity-check labour hours against your own records

The second use is a challenge to the hours. Do not ask "how long does a socket take"; ask it to compare your numbers with what usually moves them, and to name the conditions under which they would be wrong.

Here are my labour hours per item for an office fit-out. They come from my past jobs.
[paste table]
For each line, say what site conditions would make my hours too low and what would make them too high.
Do not give me new hours. Flag any line where the ratio to the others looks odd,
for example a very high or very low hours-per-unit compared with the rest of the schedule.

A realistic reply: for sockets, hours rise with chasing into masonry, surface trunking runs, working above ceilings that must stay in place and sockets on long radial runs; fall when first-fix cabling is done in bulk on open walls. For luminaires, hours rise with ceiling grid obstructions and fall when fittings are pre-assembled. Ratio flags: emergency fittings at 0.5 hours each equal ordinary luminaires, though they need testing and are often wired to a separate circuit.

What you would fix: the assistant may suggest "typical" hours despite being told not to. Ignore them. The useful content is the list of conditions, which you turn into questions on the site visit: how are the walls built, is the ceiling grid clear, what is the run length to the board? If the answer to "is the ceiling grid clear" is "no", you now know where your 12 hours for luminaires is soft. Compare with the tutorial on counting fittings from drawings, since a wrong count undermines any hours check.

Prompt three: recompute the arithmetic with working shown

Ask it to total the schedule, line by line, and then to total the cost at your labour rate and a stated material allowance, showing every step. Then compare with your spreadsheet. In the sample bid above, a correct reply gives 24.0 + 12.0 + 4.5 + 8.0 + 4.0 = 52.5 hours and 52.5 x $60 = $3,150. If the reply gives a different total, the discrepancy is either your error or its error. Find out which with a calculator before touching the bid.

A before and after example from this step. Suppose your spreadsheet total shows 54.5 hours because the distribution board line was cut from 10 hours to 8 during review, but the total cell had been typed in by hand rather than summed. The assistant's recompute of the listed lines returns 52.5, and the two-hour difference is traced to the hand-typed total. This is the kind of silent error that a second pass catches, whoever does it. Spreadsheet checks help here too; the tutorial on bid checks after the takeoff lists more of them.

Averaging a new-build hours range into a refit

Picture an electrician who asked an assistant for "typical hours for a commercial fit-out" and averaged its range into the bid. The reply had a tidy table: sockets at 0.4 hours each, luminaires at 0.35. Those numbers were for new-build, open ceilings, with first-fix and second-fix done in bulk. The actual job was a refit above a live ceiling grid with ducts in the way, and the job ran 30 per cent over. The error was not the tool; it was treating an out-of-context range as a unit rate. The cure is the rule used above: hours come from your jobs, and the assistant only asks which conditions might change them.

What a bid check is worth over a year

Say a firm submits 40 bids a year, averaging $4,000 of labour each, and a review catches an average of $300 of missed work on one bid in four. That is ten bids a year with $300 recovered, $3,000 in total. If the check costs eight minutes a bid on a $20-a-month plan, that is about five hours a year plus the subscription. The numbers are illustrative; the exercise is useful because it forces you to count catches rather than assume the benefit. If your own count shows a $300 catch is rare, adjust your expectations.

What to do when its answer disagrees with your instinct

Disagreements are the point of a second opinion, but they need a way to be settled. When the assistant says an hours figure looks low and your gut says it is fine, go to your records for the nearest comparable job: hours booked against the completed installation, not hours estimated. If there is no record, that is itself a finding, and the answer is to start recording. When the assistant says an item is missing and you believe it is covered, find the line in the contract or your exclusions that says so. If you cannot point to it in writing, it is not covered, and the customer will say the same.

Building a reference set so it stops guessing

The more of your own information it has, the fewer plausible guesses it makes. Create a one-page sheet of facts: your labour rate, your standard exclusions, the allowances you always carry (for example for testing and certification), the units you use, and your past hours per item with the job type. OpenAI stopped the creation of new custom GPTs in late September 2026 and existing GPTs stop running on 11 December 2026, so keep this in a ChatGPT Project (or paste it at the start of a chat) rather than building a GPT. Check the current Project features on OpenAI's help pages, as they change.

Before you paste anything, remove client names, addresses and site details. On business plans, OpenAI says content is not used for training by default, while consumer plans have a setting you must check; read the data controls page for your plan before uploading anything commercial. The tutorial on whether an electrician needs an AI app or ChatGPT is enough looks at when a dedicated tool is worth paying for.

Prompt four: reading a specification extract against your bid

Tender specifications bury electrical requirements in general clauses. Paste the electrical section only, with project names removed, and ask for a requirement list you can tick against your schedule.

Below is the electrical section of a specification. List every requirement that adds work or cost
(testing, labelling, warranties, spares, samples, working hours, access restrictions, submittals).
For each, quote the clause wording exactly and give its clause number.
Do not interpret. If a clause is ambiguous, say so.

An illustrative reply lists, for instance, "Clause 4.2: all circuits labelled at both ends", "Clause 7.1: operating manuals in three copies", and "Clause 9.4: works outside normal hours where noise is likely". Each is something a quick bid can miss. Because the prompt asks for exact wording and clause numbers, you can verify each in seconds, and any item with no matching clause in the original is an invention to discard. Never rely on a summary of a specification that you have not checked against the text.

Four more places it can help, and one trap

  1. Exclusions and clarifications. Paste your standard list and the job description, and ask what is missing for this job type. Illustrative additions: "no allowance for fire alarm interface", "ceiling tiles to be lifted and replaced by others", "work in normal hours only".
  2. Comparing supplier quotes. Paste two wholesalers' quotes for the same materials list and ask for a side-by-side table that flags substituted products, missing lines and different quantities. Illustrative catch: one quote lists 24 LED panels, the other 20, because a revised drawing went to only one supplier. Check every flag against the quotes themselves before you order or price.
  3. Cover email. Have it draft a short note to go with the bid, listing the key exclusions in plain words. You read and send.
  4. Post-bid review. After a loss, paste your bid summary and ask what a client might have read as risk or unclear scope. Treat it as a prompt for reflection.

The trap is the temptation to ask for "a bid for a three-bedroom rewire". The answer will look professional, carry a total and be entirely ungrounded in your costs. If it is ever pasted into a quote, you have submitted a guess. A simple rule helps: any figure that goes to a customer must be traceable to your own record or a supplier quote, never to the assistant.

Six steps to run on every electrical bid

  1. Build the bid yourself from takeoff counts and your own hours.
  2. Run the omissions prompt and sort the answers into add, ignore and ask on site.
  3. Run the labour-conditions prompt and turn the answers into site-visit questions.
  4. Have it recompute the totals, and reconcile any difference by hand.
  5. Write a one-line note of what changed because of the review. If nothing ever changes, the review is not worth the time.
  6. Submit your numbers, not the assistant's.

Keep the whole routine inside one chat or Project per bid, so the assistant sees the same schedule each time and your notes stay together. Save the final version of each schedule beside the prompts that checked it, with a line saying which suggestions you accepted and which you rejected. After a few months that folder is a record of your own typical omissions, and you can fold the recurring ones straight into your standard bid template, so the assistant has less to catch next time. Track the benefit over ten bids: how many omissions were real, how many hours you changed and how many bids you would have submitted with an error. If the answer is close to zero, stop. If it is a few real catches, you have a cheap safety net. For more ideas on where else AI fits in an electrician's week, the tutorial with ten ChatGPT prompts for quotes, emails and terms pairs well with this one. If you would like help turning this into a repeatable routine for your estimators, that is what my AI implementation consultation covers, via the contact page.

Questions electricians ask about ChatGPT and bids

Can ChatGPT give me labour hours per fitting?

It can give ranges, and those ranges are only a prompt to compare against your own records. Hours depend on access, ceiling and wall types, cable runs and your crew's speed, none of which it knows. Use your completed-job data as the source of truth.

Should I upload my price list to ChatGPT?

Only your unit-rate sheet without customer names, and only on a plan whose privacy settings you have checked. Business plans do not train on your content by default, while free consumer plans may. A Project can hold the sheet so you do not paste it each time.

Will it find mistakes in my arithmetic?

Often yes, if you give it the line items and ask it to recompute and show working. Always recheck any discrepancy it reports with a calculator or spreadsheet before changing your bid, since it can make arithmetic slips of its own.

Is it safe to send a bid that ChatGPT wrote?

No. Numbers, quantities and scope you did not verify are your liability once submitted. Use it to question your bid, then change the bid yourself.

Further reads

Sources: OpenAI ChatGPT plan and privacy help pages, and general electrical estimating practice. The sample bid and all hours are illustrative and not a benchmark.

Want a bid-checking routine your estimator trusts?

On a 1:1 call we'll look at how your bids are built, choose the checks worth handing to an assistant, and set up prompts and a template your team can reuse without relying on its numbers.

Book a 1:1 call with me