AI service reports from technicians’ voice notes: how it works and what to check

We make BoltWright, so weigh our take accordingly. Product details are as of October 16, 2026.

What the tech does on site

In BoltWright’s Field app, each unit on a stop has a Capture button. The tech taps Voice and talks through what they found and what they did, for up to five minutes, or types it instead. Photos, readings and checked-off tasks go on the same unit. Anything that still needs fixing goes under “Recommend anything?” one item at a time, so it reaches the report as a recommendation for the office to quote, never as work that was done.

Everything saves to the phone before the app says it’s saved, so a dead battery on the way down the ladder doesn’t lose it. Until the office has it, the note carries a small cloud mark. When it lands, the stop shows “Office has it.”

If a recording picks up nothing, such as wind or a muted mic, the note says no speech was heard. It doesn’t fill the gap with made-up text.

What happens at the office

The recording is turned into text after it reaches the office, so in a basement the tech sees the audio with no words under it yet. That’s normal. Once the crew’s notes are in, BoltWright usually starts the draft on its own. If it hasn’t, a manager presses Draft customer report.

The draft puts each note under the unit it was captured on, and by building when a site has more than one. It matches the unit itself rather than its label, so RTU-1 on the north building and RTU-1 on the south building keep separate histories. A unit the tech mentions that isn’t in the equipment list is marked “Not in your equipment list” for the office to attach or add.

If the tech’s last notes arrive after the draft was written and nobody has touched it, BoltWright redrafts it about 15 minutes after the notes stop coming in. Once someone in the office has started editing, their work is left alone, and the report says a note came in after the draft.

The capturing a visit and reviewing a report help pages cover each screen in detail.

What to check before trusting AI drafting

Ask these of any tool that writes reports from field notes. Under each is how BoltWright handles it.

• Does a person read it before the customer does? In BoltWright, finishing a visit sends the customer nothing, not the report and not a text. The report goes out only when someone presses Send report.
• Can you see where each sentence came from? Every line BoltWright drafts from the crew’s notes has to cite the note it came from. A line with no source is left out of the customer copy and listed for the office, with Add to summary if it’s right. Beside the draft, “Where the draft came from” shows the crew’s words behind each line and tags it Crew’s words, Reworded, Check wording or Shop standard.
• Does it make things up? The drafting instructions say to keep readings, part names and model details exactly as written, and never to invent a test, reading, part, cause or completed action. “Operating normally” doesn’t become an operation test. Instructions aren’t a guarantee, which is why the source check and the review exist. To test any tool, give it a note with no readings and see whether readings appear.
• Are completed work and recommended repairs kept apart? BoltWright only recommends what the tech recommended. When the tech names a problem but recommends nothing, the office may be offered the usual recommendation, and it stays off the customer copy until someone presses Add.
• What stays out of the customer’s copy? Prices, office instructions and internal chatter are dropped. Something meant for the office, like a new roof hatch code, shows under For the office instead. Supplier receipts a tech photographs never print on the report, and the AI that drafts it doesn’t see them.
• Does capture work without signal? Notes, recordings and photos save on the phone and send when there’s signal. Talking live with Bolt, the assistant, needs a connection. The table below sets out which is which.
• Where does the audio go? To write the text and the draft, BoltWright sends the relevant notes, audio, photos and shop records to OpenAI or Anthropic, using their business services, which don’t train their models on that content. Providers keep some requests for a period under their own policies. The privacy policy lists each provider and what it receives.

What works without signal, and what needs a connection

The offline field app guide has a test you can run on any vendor’s app.

A worked example

Here’s one made-up service call, from capture to the draft the office opens. We wrote the draft by hand to show its shape, so a real one will word things differently.

The hatch code never reaches the property manager, the motor replacement is a recommendation rather than completed work, and the 4.2 amp reading is the number Priya said. The office reads it once, sends it, and drafts the quote for RTU-1 when the supplier’s price is in.

Try it on one visit

In any demo, record a voice note in airplane mode, close and reopen the app, then reconnect. Read the draft beside the tech’s words, find a line it left out and why, and check one reading was copied exactly. Nothing should reach the customer until someone in the office sends it.

For the report itself, the service report guide covers what a commercial report needs, and the worked report example shows a finished two-unit report. The free report builder formats what you type and doesn’t use AI.