PAT Testing Software: Where AI Helps With Reports and Reminders

Yes, for the wording: AI can draft the plain-English customer summary, the reminder emails and the chasers, and explain failed items from the tester's own notes. It should not touch results, pass or fail decisions or retest intervals. Your tester and your PAT software own those, and due dates are simple date arithmetic that software already does.

That split also tells you what to test when you shop for software. On the vendor pages read for this tutorial at the start of November 2026 (Seaward's PATGuard Cloud, Kewtech's KewPAT app listing, iTemit's PAT page and Kewtech's own site), none described an AI feature. What they do describe is scheduling, alerts, emailed summaries and report export. Those claims can be tested in an afternoon, and the script is below.

Follow me on Instagram@sagnikteaches

The worked example is a one-person PAT testing business with 38 customers and about 2,400 appliances on its books. The customers are the usual mix: a tutoring centre, a yoga studio, a removals firm's depot and a landscaping company's workshop. Every figure here is illustrative.

Connect on LinkedInSagnik Bhattacharya

Two rules hold throughout. Every number in a customer email comes from the software or a spreadsheet formula, never from the AI, and every status word (pass, fail, not tested) is copied from the record, never reworded.

Subscribe on YouTube@sagnikteaches

Who owns each job: records, dates and words

Rows five to seven are where AI can earn its place. Rows one, two, three and eight are where it should have no say, and row four belongs to the software. A tool that blurs the line, for instance one that lets a model suggest a retest interval or edit a status, deserves a harder look than one that only drafts text.

What the vendor pages promise, and what each claim leaves open

These are the claims as worded on those pages, with the question each one leaves unanswered. The quotes come from Seaward's PATGuard Cloud page, the KewPAT App Store listing and iTemit's PAT testing page.

Features also differ by plan, so ask which plan carries the one you are testing. On Seaward's page, automatic test scheduling is listed from the Basic plan, while scheduled reports and reminders and guest report access appear on the Business plan, the third of four. The KewPAT listing describes a free iPad app whose remote-control features are for Kewtech's own SMARTPAT tester, so check what your own instrument can do with it. Plans and inclusions change, so read the page you are buying from.

A 45-minute trial script that settles the claims

Most vendors offer a trial. iTemit's page mentions 14 days with no credit card, and Seaward's page has a free-trial button without stating terms. Use the trial on dummy data before any real customer goes in, and run the same script on every product you are comparing.

1. Create a dummy customer called "Test Site" with one location and 12 dummy appliances across four types: a kettle, an extension lead, a monitor and a drill.
2. Enter results: ten passes, one fail with a written comment, and one pass with a comment reading "retest early". If the software imports from your tester, download a real batch of twelve instead and note which fields arrive: values, visual result and comment.
3. Set the dates so that three items fall due in 5 days, three in 35 days and six in 340 days. Back-date the test date or change the interval to get there.
4. Trigger or wait for the alerts and reminders. Write down what arrived, at which address, how many days ahead, and what the subject line said.
5. Open the customer summary or report. Does it name the failed item and its comment, or only give a count? Can you edit the wording before it goes out?
6. Export the data as a CSV file and open it in a spreadsheet. Note the column names. This is what you would hand to an AI later, so check that it carries asset ID, location, test date, result, comment and due date.
7. Delete the dummy site and confirm it has gone, including from exports.

Score each claim in the table above as 0 (did not work), 1 (worked with a workaround) or 2 (worked as described). A 0 or 1 on steps 4 and 5, the customer-facing ones, is a reason to choose another product or plan to cover the gap with your own templates and a mail merge. An hour on dummy data costs less than a month on the wrong plan.

From export to customer summary: a prompt, a result and the fixes

Take one visit from the worked example: a tutoring centre with 88 items on the register for the site, 86 tested and two not available on the day. The software's export has the statuses in column F, and four formulas produce the counts that go into the email.

Those counts reconcile: 83 plus 3 is the 86 tested, and 86 plus 2 is the 88 on the register. If they ever fail to add up, fix the record before writing a word. The counts, the failed items and the not-tested items then go into the prompt, along with rules that close off the usual ways a drafting assistant goes wrong.

You are drafting a customer summary of a portable appliance testing visit. The tester will check it and send it. British English, plain words, 150 to 200 words.

Rules:
1. Copy the numbers from the COUNTS line exactly. Do not add, subtract or recalculate anything.
2. For each failed item, give the item, location, recorded reason and recorded action in the tester's own words. Do not explain why a failure matters.
3. Never use the words "safe", "unsafe", "compliant", "dangerous" or "certified".
4. List each item that was not tested, with the reason recorded.
5. Do not mention retest dates. I will add them.
6. If a field is empty for a failed item, write [tester to complete] instead of guessing.

VISIT: site code TC-07, [date], tester [your name]
COUNTS: on register 88; tested 86; passed 83; failed 3; not tested 2
FAILED (item | location | recorded reason | recorded action):
Kettle | staff room | earth continuity resistance above limit | withdrawn from use, labelled
Extension lead | office 2 | damaged outer sheath | removed by tester with site contact's agreement
Fan heater | reception | insulation resistance below limit | withdrawn from use, labelled
NOT TESTED (item | reason):
Projector | in use during a session
Fridge | not accessible on the day

Here is the sort of draft that comes back. It is illustrative, and every run words it differently.

Hello [first name],

Thank you for having us in on [date]. We tested 86 appliances across the centre: 83 passed and 3 failed.

The three that failed were a kettle in the staff room (earth continuity resistance above the limit; withdrawn from use and labelled), an extension lead in office 2 (damaged outer sheath; removed with your agreement) and a fan heater in reception (insulation resistance below the limit; withdrawn from use and labelled). Please keep these items out of use until they have been repaired or replaced and retested.

Two items could not be tested: the projector, which was in use during a session, and the fridge, which was not accessible on the day. We will arrange a time to test both.

All other appliances are fine and need no further action. The full report is attached.

Kind regards, [your name]

It reads well, which is why it needs a check. Five things need attention, and they are typical.

• "All other appliances are fine and need no further action" is a conclusion nobody gave the model, and "fine" is the kind of word rule 3 was meant to stop. Delete the sentence. The counts already say 83 passed.
• "Please keep these items out of use" is sensible advice, but the prompt did not supply it. Keep it only if it matches your own standard wording.
• "We will arrange a time to test both" is a promise. Replace it with a time you will really offer.
• "Thank you for having us" and "we" are plural. If you work alone, change them.
• Check the numbers against the table above, then check each status word against the export. Every word for a pass, fail or not tested must appear exactly as recorded.

Treat the email as a cover note. The software's own report goes with it as an attachment, and the summary never replaces it.

Failed items: explain the record, don't decide it

A failure is where customers phone you, and where a model's helpfulness does the most damage. It wants to explain why a failure matters, reassure the reader or recommend an action. What the customer needs is the item, the recorded result and what happened to the item.

The last row is why rule 6 is in the prompt. If your notes are thin, an AI will make them richer, and a richer invented comment on a failed appliance is worse than a short true one. Fix the habit at the source: a two-word comment such as "damaged sheath" beats none, and it takes five seconds on site.

Reminders: the software owns the date, AI owns the wording

A reminder has a date, a recipient and some words. The software supplies the first two. The words are yours, and most reminders fail because they are either too stiff or too vague to act on. One that works says how many items, by when, and how to book.

That schedule is a starting point, not a standard, so adjust it to your customers. Group by customer and visit rather than by item: one email per site saying "14 items due within 60 days, 2 already overdue", with the counts coming from a filter or pivot table on your export and never from the AI.

Write five reminder emails for a PAT testing business, to go out 60 days, 30 days and 7 days before a retest date, on the date, and 14 days after. British English. Under 90 words each. Friendly and direct, no pressure language.

Rules:
- Use only these placeholders: [first name], [site code], [number of items], [due date], [slot 1], [slot 2], [booking link], [your name]. Do not invent numbers or dates.
- Say nothing about safety, compliance, insurance or legal duties.
- Each email ends with one clear action.
- Write a subject line for each.

An illustrative result for the 30-day email, after a light edit:

Subject: Your retest at [site code]: two slots to choose from

Hi [first name], [number of items] items at [site code] are due for retesting on [due date]. I have [slot 1] or [slot 2] free. Reply with the one that suits and I will confirm it, or book through [booking link]. [your name]

The overdue email is where models drift. Expect phrases such as "to keep your premises compliant" or "insurers may ask for proof". Both are claims about legal duties and insurance that nobody supplied, so cut them and keep the action. If your software already sends reminders to customers, you may only need the AI to improve the template text once, rather than write each email. The same pattern runs through service reminders that bring garage customers back, where the trigger is a service date instead of a retest date.

No software yet? A due-date sheet that does the arithmetic

A spreadsheet will carry a small business for a while. Four columns do the work: item, last test date, interval in months and next due date, with a fifth for status. Ask an AI to write the formulas, then test them on three rows whose answers you can check by hand. The rows below assume today is 1 March 2027.

Next due (D2):   =EDATE(B2, C2)
Status (E2):     =IF(D2<TODAY(), "Overdue", IF(D2-TODAY()<=60, "Due within 60 days", "OK"))
Count (any cell): =COUNTIF(E:E, "Due within 60 days")

The intervals in the third column are yours, taken from the risk assessment you work to, and the sheet only adds months. The 60-day window is a setting you can change. The pattern is the one described in the tutorial on an expiry date tracking spreadsheet, and the calibration log template applies the same arithmetic to gauges and scales. A sheet is a stopgap. Past a few hundred items, an asset register with labels and tester import saves more time than any AI step.

If a vendor adds an AI claim: five checks before you switch it on

The pages read for this tutorial describe no AI features, but that will not stay true for every product. When a vendor does add an "AI summary" or "smart reminders" button, run these five checks on your dummy site first.

Two failures matter most. A tool that rewrites a status, however helpfully, has crossed from drafting into deciding. A tool whose counts are wrong in the dummy test will be wrong in the real one, and the error will go out under your name.

What goes into an AI chat: codes, not customers

A PAT record carries few personal details. Appliance types, rooms and results are not personal data. The personal part is the customer: names, addresses and contact details. So use a site code such as TC-07, paste only the columns the task needs, and add names back inside your email app.

The plan you draft on is the other half. A personal plan suits a tester working alone, provided the model-training switch in its privacy settings is off. A business plan adds protection by default, because ChatGPT Business and Claude Team say they do not train on workspace content, but each starts at two seats. For one tester that is $40 to $50 a month at list prices, which is more than this work justifies by itself. Whichever you choose, read your customer contracts for any clause on third-party tools, and put the same questions to your PAT vendor if its software ships with built-in AI. For the wider records habit, see the compliance calendar for licences, insurance and certificates.

A year of drafting, timed: hours saved against the plan's price

Back to the one-person business with 38 customers, each visited once a year. Treat the times as illustrative and replace them with your own stopwatch figures.

The sums: 12 minutes saved on each of 38 summaries is 456 minutes, and 3 minutes saved on each of 152 emails is also 456, so 912 minutes or 15.2 hours. A personal plan at about $20 a month costs $240 a year, which works out at roughly $16 per hour saved. That is a modest gain, and an honest one. If drafting emails is the only job you give the AI, a free tier may be enough. The real value in this business is a due date that never slips, and that is the software's job. The AI earns more when it also writes quotes and answers enquiries, as set out in the tutorial on the best AI tools for electricians and their costs.

Questions PAT testers ask about AI and their software

Can AI decide whether an appliance passes or fails?

No. A pass or fail comes from your tester's measurements and your visual inspection, recorded by the software. AI should only copy the recorded status into wording. If a tool lets a model change a status or a retest date, switch that off, because the result is a safety record you sign for.

Do I need AI if my PAT software already sends reminders?

Often not for the dates. Reminders are date arithmetic that software does reliably, and the vendor pages describe them without any AI. AI helps with the words around them: a clearer customer summary, a friendlier chaser, a note explaining a failed item. Add it only where your current templates are the weak point.

Is it safe to paste test results into ChatGPT or Claude?

Remove customer names, addresses and contact details first and use site codes instead. A list of appliance types, locations and results does not need personal data. Use a business plan, or a personal plan with training switched off, and check what your customer contracts say about third-party tools.

How should a failed item be worded in a customer email?

State the item, where it was, the recorded result and the recorded action, in the tester's own words: for example, withdrawn from use and labelled. Do not add words like dangerous or compliant. Let AI tidy the sentence, then compare it with the record before you send anything.

Further reads

• Fire Protection Software: Inspection Reports and Reminders With AI — The same reports-and-reminders split for a fire protection business.
• Electrical Installation Certificate: What AI Can Draft and You Must Sign — What AI can draft on electrical paperwork and what you must sign.
• Emergency Lighting Certificate With AI: Test Log to Signed Paperwork — From a test log to signed paperwork without losing the record.
• Gas Engineer Software: What AI Can Draft and What You Must Sign — Another safety-record trade where the engineer signs, not the AI.
• Chimney Sweep Business Software With AI: Reminders and Online Booking — Annual reminders tied to online booking.
• Plumbing Maintenance Plans: AI-Written Offers and Renewals — Turn a recurring visit into a plan customers renew.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Seaward PATGuard Cloud product page, Kewtech's KewPAT App Store listing and website, and iTemit's PAT testing page, all read at the start of November 2026; OpenAI, Anthropic and Google plan and data-use pages.