AI in Core Inspection

AI writes the summary. Your inspector does the inspection.

Core reads the record your inspector made and writes it up in plain English: a paragraph for the client, a note for the site manager, and the item's history for whoever examines it next. It never inspects, passes or fails anything.

Read how it works

The inspector's record

Equipment
Chain hoist, WLL 3.2 t
Serial
KR-CH-0005
Examination
6-monthly thorough examination
Defect
Brake slipping under load
Existing or imminent danger
Yes
Remedy
Scrap and replace
Safe to operate
No
Result
Fail

Summary

AI-generated

The 3.2 t chain hoist (serial KR-CH-0005) failed its thorough examination on 2 October 2026 because its brake was slipping under load. The inspector recorded that it is not safe to operate and is to be scrapped and replaced.

Example with made-up data.

Three summaries, each for a different reader

About a third of the inspections recorded in Core carry a typed comment. Somebody then reads those comments and writes them up again for the client. Core now writes that first draft a few minutes after the job is marked complete.

For your client

Job summary

One paragraph on the whole visit: how many items were inspected, how many failed, the failures grouped by type and cause, and what the inspector noted on the rest.

AI-generated

186 items were inspected on job 10482: 179 passed and 7 failed. The failures were 5 chain slings with stretched or nicked links and 2 webbing slings with cut stitching. The inspector also noted surface corrosion on several shackles that passed.

For the site manager

Inspection summary

Up to three sentences on a single item, written for someone who was not there. Only inspections that failed or carry an inspector's comment get one.

AI-generated

The 3.2 t chain hoist (serial KR-CH-0005) failed its thorough examination on 2 October 2026 because its brake was slipping under load. The inspector recorded that it is not safe to operate and is to be scrapped and replaced.

For the next inspector

Equipment history summary

What happened to this item before: past failures, repairs, parts replaced, things an earlier inspector marked to monitor. It is on the technician's device before they arrive.

AI-generated

This hoist failed in March 2025 for a worn load chain, which was replaced in April 2025. In October 2025 the inspector noted slight play in the bottom hook swivel and marked it to monitor.

Examples use made-up data. A job where everything passed and nobody left a note gets one sentence written by Core from the counts, with no AI involved.

What it does, and what it leaves to your inspectors

LOLER, OSHA, ASME B30 and the AS/NZS standards all give the examination, the judgement and the signature to a person. Core keeps AI on the paperwork side of that line.

The AI does

  • Writes a plain-English summary of a completed job for your client
  • Summarises each failed or commented inspection for the site manager
  • Tells the next inspector what happened to an item before
  • Fixes typos and expands obvious abbreviations in inspector comments
  • Writes in the language your inspectors write in
  • Prints on certificates, reports and the client portal, where you choose

The AI does not

  • Inspect, examine or test anything
  • Decide, change or soften a pass or fail
  • Judge equipment from a photo
  • Give safety advice or recommendations of its own
  • Count items or calculate anything (Core does that)
  • Sign a report or replace your competent person

How Core keeps the summary true to the record

An AI model sounds as sure when it is wrong as when it is right. So the parts that must be the same every time are done by Core, and the AI's work is checked before anyone sees it.

Core does the counting

Items inspected, passed, failed and not seen are calculated by Core and handed to the AI as fixed numbers. The AI writes the sentence around them.

Every number is checked

After a summary is written, Core compares each number in it with the record. If one is not in the record, the summary is discarded.

Labelled wherever it appears

AI-written text carries an "AI-generated" label on screen, in the app, on the portal and on printed documents.

An authoriser can edit it

Summaries are shown during batch authorisation. An authoriser can correct the wording, and the label then shows who edited it.

Fixed at approval

Once a job is approved, its summaries cannot be edited or rewritten, so what your client received stays what is on file.

Off until you ask

AI summaries are off by default. Showing them to clients on the portal is a second switch, also off by default.

Why we drew the line here

We asked AI to inspect lifting gear 380 times

We showed two leading AI models 19 photos of real lifting equipment, ten times each, and asked for a pass, a fail or an unsure. Damaged gear failed every time. Sound gear was the problem: one model named a defect on new equipment 58 times out of 60, and the other would not commit. The same models read clean tags and wrote up records accurately.

Read the full test and what the rulebooks say
380

answers from two AI models on 19 photos

1

pass, for a shackle with no visible markings

58 of 60

fails one model gave to new or clean kit

Where the summaries show up

Nothing changes on a page or a document until a summary exists for it, and nothing prints until you add the tag.

Web app

On Job Details, Edit Inspection, Equipment Details and Batch Authorisation.

Mobile app

In the Inspection History box on each item, synced ahead of the visit. No app update needed.

Certificates and reports

Add a tag to a Word or Excel template and the summary prints there, as a paragraph or a table column.

Client portal

On the client's job and equipment pages, once the job is approved. Your choice.

Where your data goes

Summaries are written by Claude, made by Anthropic, through Anthropic's business API. Under Anthropic's commercial terms, content sent through the API is not used to train its models.

The AI has no login to Core. It receives the text of one record and returns one paragraph.

The full list of what is and is not sent

Sent

  • Equipment type, description and serial number
  • Checklist answers and comments
  • The recorded result and date
  • Job number and counts

Not sent

  • Client names, sites and addresses
  • Inspector and user names
  • Photos, signatures and files
  • Internal-only checklist items

Questions buyers ask about AI in Core

Does Core's AI inspect equipment?

No. A competent person inspects the equipment and records the result in Core, as they always have. After the job is completed, the AI reads that record and writes a short summary of it. It has no part in the inspection and no way to change the result.

Can the AI pass or fail an item from a photo?

No, and we tested why. We showed two leading AI models 19 photos of lifting gear ten times each. Out of 380 answers there was one pass, and it was for a shackle with no visible markings. One model failed brand-new equipment 58 times out of 60. A thorough examination needs gauges, function tests and a person who signs, so Core keeps AI on the writing.

What does the AI write?

Three things. A job summary for your client, covering what was inspected, what failed and why. An inspection summary for each inspection that failed or carries an inspector's comment. An equipment history summary for the next inspector, covering past failures, repairs and things marked to monitor.

How do you stop it making things up?

Core calculates every count itself and hands the numbers to the AI. After the summary is written, Core checks every number in it against the record and discards the summary if one does not match. The AI is instructed to use only what is in the record, never to soften a result, and to quote unclear comments exactly. Every AI-written summary is labelled AI-generated, and an authoriser can edit it before the job is approved.

Which AI model does Core use, and where does our data go?

Summaries are written by Claude, made by Anthropic, through Anthropic's business API. Core sends the text of the record: equipment type, serial number, checklist answers and inspector comments. It does not send client names, addresses, inspector names or photos. Under Anthropic's commercial terms, content sent through the API is not used to train its models.

What does it cost?

AI summaries are a paid add-on to your Core subscription and are off by default. Ask us for pricing and we will switch them on for your account.

Do our clients see the summaries?

Only if you want them to. Showing summaries on the customer portal is a separate switch and is off by default. Summaries print on certificates and reports only where you add the tag to the template.

Does it work in the mobile app?

Yes. Equipment history summaries appear in the Inspection History box on each item, and they are written ahead of a scheduled visit so they are on the device before the technician arrives. No app update is needed.

Setting it up? The AI summaries documentation covers where summaries appear, how authorisers edit them and the template tags for certificates and reports.

See a job summary written from a real job

On a 30-minute demo we will complete a job and show you the summaries it produces, on the web, in the app and on the certificate. AI summaries are a paid add-on, switched on per account.

Ask for pricing