AI Training for Electricians: What the Crew Should Learn First

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for AI Training for Electricians: What the Crew Should Learn First.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for AI Training for Electricians: What the Crew Should Learn First.

Teach three jobs first: customer emails, quote wording and job photo notes, because they are everyday writing tasks where a person can check every line. Put data rules in place before anyone starts, train in two short sessions on real jobs, and keep code answers, pricing and certificates for later.

Most advice on AI training for electricians is written for one person taking a course. A firm has a different problem: six people, one set of customers and nothing to stop a keen apprentice pasting an alarm code into a free chatbot. What fixes that is one shared tool, one prompt sheet and a page of rules, taught on your own jobs.

Follow me on Instagram@sagnikteaches

Why a crew plan beats a stack of individual courses

A course can teach a person. It cannot write your data rules or hold the crew to them. A firm needs three things that sit above any one person's skill:

Connect on LinkedInSagnik Bhattacharya
  • The same rules for everyone: what may be pasted into a tool and what may not.
  • The same wording for common jobs: a shared prompt sheet, so a reply from the office and a reply from a van sound like one firm.
  • One person responsible: an AI lead who owns the accounts, the settings and the prompt sheet.

Without those, each person finds a free tool, customer details end up in places nobody chose, and quality varies by who wrote the message. That pattern has a name, shadow AI, and the usual cure is a supported alternative, not a ban; stopping staff pasting client data into free tools covers the mechanics. The AI lead is often the person who runs the office, or a lead electrician who likes this sort of thing. The owner sets the rules, the lead keeps the sheet current, electricians use the sheet, and apprentices use it with someone checking.

Subscribe on YouTube@sagnikteaches

The three first jobs, and why they come first

JobWhy it comes firstWhat a person must checkWho starts
Customer emailsDaily, low risk, and a bad draft is easy to spotFacts, promises, dates and pricesOffice admin and owner
Quote wordingTime-consuming, and the wording matters as much as the figureScope, inclusions, exclusions and every numberWhoever prices jobs
Photo and job notesCaptures the work while it is fresh and improves your recordsThat each caption matches what is visible, and that no safety claim slipped inElectricians, with a checker

All three share one property: the output is words that a trained person can check against what they already know. None asks the AI to make a technical judgement. The order matters as well. Emails teach the basic habit of giving context, facts and limits. Quote wording adds a structured input and a rule about numbers. Photo notes add field data, and with it the privacy rules about what appears in a picture.

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

Job one: customer emails, from prompt to checked draft

Start with an enquiry the office already has. The prompt gives the AI four things: who it is writing as, what to ask the customer, what to offer, and what it must not say. The last of those does most of the work. This is a prompt for a charger enquiry, with the enquiry text pasted at the bottom:

You write replies for a small electrical firm. Reply to the enquiry below in 90 to 120 words, friendly and plain, no jargon.
Ask for: a photo of the main panel, a photo of where the charger would go, and whether the garage has its own supply.
Offer a site visit on Tuesday or Thursday afternoon.
Do not give a price, a start date, a promise about speed, or any statement about regulations or what is allowed. Do not say we can definitely do the job.
Sign off as [your name], [firm name].

ENQUIRY
Hi, we'd like a quote for a 7 kW EV charger on the side of our garage. The panel is in the cellar. Thanks, [first name]

An illustrative draft: "Hi [first name], thanks for getting in touch about the charger. To quote accurately, could you send three things: a photo of the main panel in the cellar, a photo of the spot on the garage wall where the charger would go, and a note on whether the garage has its own supply? I can visit on Tuesday or Thursday afternoon to look things over, and we'll have it sorted quickly once we know what is involved. Best wishes, [your name], [firm name]."

The draft is nearly right, and it contains the sort of slip the crew has to learn to see. "Sorted quickly" is a promise about speed that nobody in the firm has made. Teach one reading rule: scan every draft for verbs of commitment, such as will, can, guarantee and sorted, and ask of each whether the firm has actually decided that. Then fix the prompt so the slip does not recur, by naming the thing in the do-not list (as the prompt above now does with "a promise about speed"). Finally, check the offered days against the diary before sending, because the AI will happily offer a Tuesday you are on holiday.

Job two: quote wording, with the numbers left to you

Quote wording is where electricians lose evenings. The AI can turn rough notes into clean text under four headings, and the estimator then adds every price. The rule that makes this safe is that the AI sees the scope and never the rates. These are filled-in notes for the same charger job:

Turn these job notes into customer-ready quote wording with four headings: What we will do, What is included, What is not included, What we have assumed. Use only what is in the notes. Do not add items, materials, certificates, warranties or timings that are not in the notes. If something a customer would expect is missing from the notes, list it at the end under "Ask the estimator". Do not write any prices.

NOTES
Supply and fit 7 kW EV charger on garage wall. Customer supplies the charger. New circuit from main panel, 18 m cable run, panel has 2 spare ways. Cable surface-mounted and clipped along the side wall and through the garage wall. Excludes: groundworks, decorating, any supply upgrade. Assumes: access to the panel on the day, brick wall.

An illustrative result: under What we will do, fitting the customer-supplied charger to the garage wall and running a new circuit from the main panel. Under What is included, one of the two spare ways, 18 m of surface-mounted cable clipped to the side wall, and the fixing of the charger. Under What is not included, groundworks, decorating and any supply upgrade. Under What we have assumed, access to the panel on the day and a brick wall. Then, under Ask the estimator: is testing and paperwork included, which cable type is specified, and is making good around the wall entry included?

That last list is the useful part. Before the "do not add" instruction, assistants often fill gaps with plausible inclusions such as "all materials supplied" or "certification included", and a customer will hold you to a sentence like that. Here the estimator answers each question, adds the figures and sends it. Everyone on the crew should know the check: every inclusion must trace back to the notes, and if it does not, it is deleted or confirmed. For a ready-made structure, the electrician quote template with scope and exclusions goes further.

Job three: photo and job notes that match what is visible

The routine is simple. At the end of a job the electrician takes the usual photos and records a 30-second voice note on a phone. Dictation turns the note into text, and the text goes to the AI with an instruction to produce a short customer summary and an internal to-do list. Start with text only. Photos can follow once the crew has the habit, and each photo then needs the extra care covered in the rules below.

Here is a voice-note transcript from a job. Turn it into (1) a five-line job summary for the customer and (2) an internal note listing anything unfinished. Use only what is in the transcript. Do not describe test results, safety conclusions or compliance. If something is unclear, write "unclear" and quote it.

TRANSCRIPT
Kitchen job finished, old panel out, new panel in the utility room, circuits labelled, two cable runs in the kitchen still need tidying, tested all circuits, results go on the paperwork tomorrow, customer wants the old panel taken away, outside light trips now and then so ask customer

An illustrative result: a customer summary of five bullets (old panel replaced with a new one in the utility room; circuits labelled; circuits tested, with the paperwork to follow; two kitchen cable runs to be tidied on the next visit; old panel to be taken away as requested) and an internal note (tidy the two cable runs, remove the old panel, ask the customer when the outside light trips and investigate).

The checking rules for this job are about two words. The first is "safe". If a summary says the installation is safe, or all circuits passed, that is a conclusion for the signed paperwork, not for a note-taking tool. The prompt forbids it, but the reader has to confirm it stayed out. The second is "tomorrow", because the draft passes a promise through from the transcript. Replace it with a date the person has checked. With photos, add one more instruction: describe only what is visible, and write "unclear" for anything else, since an AI will happily call a connection tight or a cable correctly rated from a photo that cannot show either. Keep the original photos untouched, whatever else happens to them. Turning on-site voice notes into job sheets and quotes goes deeper on this workflow.

Data rules for the crew: a page you can paste

Choose the account before the rules, because the account decides what is possible. OpenAI says that, by default, it does not use data from ChatGPT Business, Enterprise or its API for training, and its business pricing page repeats it. Anthropic says the same of its commercial products such as Claude for Work. Personal plans are different. On ChatGPT Free, Plus or Pro, OpenAI's help page says sharing for model improvement is on by default, and you turn off "Improve the model for everyone" under Settings, then Data controls. Its Temporary Chat is not used to improve models but may be kept for up to 30 days for safety. As of November 2026, a ChatGPT Business standard seat is $25 a month ($20 billed annually) and a Claude Team standard seat is $20 a month billed annually ($25 monthly). If the firm has a Microsoft 365 business plan, Copilot Chat is included at no extra cost; ask whoever runs your account what is switched on.

Once customer details are in a business account, the rules are mostly about what not to paste. This is a one-page version for the crew to read aloud in the first session:

AI RULES FOR THE CREW

1. Use only the firm's AI account for anything about a customer or a job. Never use a personal account for work.
2. Never paste alarm or door codes, key-safe numbers, passwords, bank or card details, ID documents, or anything about a customer's health or vulnerability.
3. Use first names and the job type. Leave out full addresses unless the task needs one.
4. Photos: crop out faces, documents, vehicle plates and alarm keypads. Switch off location tagging for photos you upload.
5. AI writes drafts. A named person reads every line before it leaves the firm. Nothing goes out with a price, a date or a safety statement that a person has not checked.
6. Do not use AI for code or regulation answers, cable sizing, test results or certificates.
7. Keep the original photos. Never edit a photo to make it easier for the AI.
8. If unsure, ask [AI lead]. If something goes wrong, tell [AI lead] the same day. Honest mistakes reported quickly get no blame.

Rule 2 gets broken like this. An apprentice pastes a customer's whole email thread into a free personal chatbot to get a quicker reply, and the thread includes the key-safe number the customer wrote for access. Nothing visibly goes wrong, which is the trouble: nobody knows whether the number was kept. The fix is not a lecture. It is the business account, rule 2 written down, and a habit of asking, plus a call to the customer to change the number. Check any contract or tender terms that limit sharing documents with outside services, and ask a qualified adviser about the data-protection rules that apply to you and what your insurer expects.

Training the crew in two 45-minute sessions

Run the sessions yourself, on real jobs, with no slides. Whoever acts as the AI lead leads, and the owner sits in so the rules carry weight. Hold them in the office at the start or end of a day when the vans are in. Apprentices come to both and are paired with a qualified electrician who checks their drafts for the first month.

MinutesSession oneSession two, two weeks later
0 to 10Why the firm is doing this; the rules page read aloudEach person brings one draft they used and one thing that went wrong
10 to 20Live demo of job one on a real enquiry with the details removed; the crew spots the promisesUpdate the prompt sheet: keep what worked, delete what did not
20 to 30Job two: the estimator pastes notes; the crew lists what the draft got wrongTest the data rules against what actually happened
30 to 40Job three: a phone voice note becomes a summary; the crew looks for the word safe and for invented detailDecide the next job to add, or decide to stop
40 to 45Everyone does one task on their own phone; prompt sheet handed outAgree who checks what in the coming month

The prompt sheet is the thing people take away. It holds the three prompts from this tutorial with your firm's name, days and wording filled in, plus the rules page. On a business plan, ChatGPT and Claude both let you share a Project, so the sheet and rules can sit in the tool where the crew already works, and building a shared prompt library covers keeping it tidy. A paper copy in each van costs almost nothing and gets read.

A four-week rollout for a six-person firm, with costs

Take a six-person electrical firm: the owner, who also prices jobs, a part-time office administrator, three electricians and an apprentice. The administrator becomes the AI lead. Every figure below is illustrative, so swap in your own rates.

  • Week 0, setup (4 hours of the AI lead's time): choose one tool, create three seats for the owner, the administrator and the lead electrician, switch off anything that shares data, write the prompt sheet and prepare the demo. Three standard seats cost $75 a month billed monthly, or $60 if billed annually.
  • Week 1, session one: 45 minutes with all six, which is 4.5 hours of staff time.
  • Weeks 2 and 3, real use: the three seat holders use the sheet on real work. The AI lead keeps a simple log: task, minutes saved or lost, and any problem.
  • Week 4, session two and the decision: 45 minutes, again with all six, another 4.5 hours, plus about 3 hours for the lead to review the log.

Staff time comes to 16 hours (4 + 4.5 + 4.5 + 3). At an internal cost of $40 an hour that is $640, plus $75 for the first month of seats, about $715 to get going. The firm then pays $75 a month. To cover the setup within three months and the seats along the way, it needs to save about 21.6 hours over 13 weeks: $640 plus three months of seats at $75 comes to $865, and $865 divided by $40 is 21.6. That is about 1.7 hours a week across the firm.

Is that plausible? A guess to test, not a promise: eight enquiry replies a week at 6 minutes saved each is 48 minutes, four quotes a week at 15 minutes saved is 60 minutes, and nine jobs a week where the lead electrician or the administrator turns a voice note into a summary at 5 minutes saved each is 45 minutes. That adds up to about 2.5 hours, above the 1.7 needed. The log from weeks 2 and 3 replaces these guesses with real numbers. If the measured saving is under 1.7 hours a week, narrow the rollout to the job that works, or stop. How much time and money staff training takes gives a wider range if your firm is larger.

What to skip in the first month, and when to come back

After a good first week the temptation is to add things. Hold them back. Each of these needs either more checking skill or more setup than a new crew has.

  • Code and regulation answers. A chat tool can sound sure about a rule that is out of date, belongs somewhere else or does not exist, and a wrong answer here is a safety matter. Use the standards documents and your technical helpline; whether ChatGPT can answer code questions safely explains why the answer is no. Treat it, at most, as a way to find which clause to go and read.
  • Pricing and cable or load sizing. The tool does not know your rates, your suppliers or your site. Put your own numbers in. Once the basics hold, a price guide built from your own past jobs is a better next step than asking a tool for prices.
  • Certificates and test records. The signed record is yours, and a model cannot see a meter reading you did not give it. Drafting the text around a certificate can wait until you have a rule for who signs.
  • Risk assessments and method statements. The same logic applies: an AI can draft, and a competent person signs.
  • Agents, automations and website chatbots. They need rules, testing and a handover plan. Start with prompts the crew can see and check.
  • Several tools at once. The crew needs one tool and three jobs. Do not build the plan around ChatGPT's Custom GPTs either: OpenAI ended their creation in late September 2026 and existing ones stop running on 11 December 2026, so use a shared Project.

Come back to the list after the week 4 review, one item at a time, and only when the three first jobs are running without problems.

How to tell the training worked

Do not measure attendance. Measure whether people can do the three jobs safely, with a short practical check. Give each person an anonymised enquiry, a set of job notes and a voice-note transcript, and have a second person mark the results against a pass or fail card:

  1. No invented facts, items or prices appear in the output.
  2. No promises about dates, speed or results survive into the final version.
  3. No code, regulation or safety statements appear, and the word safe is absent from customer text.
  4. No customer details beyond what the rules allow were pasted in.
  5. The person can say what they changed and why.

Ask the AI lead to note for a month how many customer-facing drafts went out unchanged. If nearly all do, people may have stopped reading; if none do, the prompts need work. A healthy pattern is most drafts lightly edited and a few rewritten. Other warning signs are a personal account still in use for work, prompts drifting away from the shared sheet, and an apprentice working without a checker.

Repeat the practical check after three months, and again whenever the tool changes in a way that matters. Some regions' rules ask businesses to support staff AI literacy, so ask a qualified adviser whether any apply to you, and keep your session plan and attendance list as evidence either way. A written plan like this also makes a convincing answer if a tender asks how the firm uses AI.

Training an electrical crew on AI: common questions

Should apprentices use AI on customer work?

Yes for drafts, with rules: they read every line, a qualified person checks anything customer-facing, and they never use it for code, sizing or test results. Teach apprentices to check the draft against what they saw on site, because the habit of checking matters more than speed at this stage.

Do we need paid accounts for the whole crew?

No. Start with the two or three people who write most: the owner, whoever runs the office and a lead electrician. Anyone else can pass notes to the AI lead until the review shows who needs a seat. Do not let free personal accounts handle customer details.

How long should the first training take?

Two 45-minute sessions, about two weeks apart, are enough to start. The first demonstrates the three jobs and the data rules on a real, anonymised example; the second reviews what went wrong and updates the prompt sheet. Add roughly half a day of setup for whoever acts as the AI lead.

What if the crew already uses free chatbots?

Do not ban it overnight. Ask what people use it for and move the useful tasks to the business tool. Switch off the model-improvement setting on any personal accounts that stay in use, and write down the rule about customer details. Banned tools get hidden; supported ones get shared.

Further reads

Sources: OpenAI business data and data controls pages, Anthropic commercial and consumer privacy pages, and ChatGPT Business and Claude Team pricing pages, all read in November 2026.

Want a first-jobs plan built for your crew?

On a 1:1 call we'll pick the first jobs worth handing to AI at your firm, set the data rules for your crew, and agree what the team can run on its own.

Book a 1:1 call with me