Voice to Text Job Notes: How to Dictate on Site in 2026

Voice to text job notes work well in 2026, but only if you speak in a fixed order rather than rambling. The habit that makes the difference: hold the phone about 20 cm from your mouth, say the job or room first, then what you did, then what is outstanding, then any material used with quantities, and stop. Four sentences, roughly 25 seconds. Transcription engines now handle drills, radios and a windy roof fine; what they cannot fix is a note that has no structure, so the office reads 90 words and still has to ring you. Dictate the shape, not the story.

What has actually changed

Two things. First, transcription quality: site noise used to break dictation, and current models cope with a compressor running in the background. Second, availability. Voice notes used to be a paid add-on, or a separate app you had to remember to open. Now a field app records straight into the field it belongs in, so the words land on the job card instead of in your voice memos, never to be seen again. The question has moved on. It is no longer "does it work". It is "what do I say, and when do I not bother".

The four-part script for voice to text job notes

Say these four things, in this order, every time. It takes a week to become automatic and it is the single highest-value habit in this article.

1. Where. "Flat 3, kitchen." "Plant room, second riser." Location first, always. Everything after it inherits that context.
2. What you did. Past tense, one clause per action. "Replaced the consumer unit. Tested and it passed."
3. What is outstanding. Flag it explicitly: "Outstanding: needs a second visit to fit the cooker point, customer is sourcing the appliance."
4. Material and time. Numbers and units out loud: "Used 18 metres of 2.5 twin and earth, four double sockets, two hours forty on site."

Then stop talking. The strongest signal of a bad dictated note is that it keeps going. If you find yourself saying "and, er, basically", you have finished and your mouth has not noticed.

Say the punctuation you actually need

Most engines add sentence breaks reasonably well now, but they will not guess a list. Say "full stop" and "new line" where the meaning depends on it. "New line" before each outstanding item turns a wall of text into something the office can tick off. Do not bother saying "comma" everywhere; it makes you sound like a robot and buys you nothing.

How to speak so it lands right the first time

• Distance beats volume. Phone 15 to 25 cm from your mouth, slightly off to the side so you do not pop the mic. Shouting from a metre away is worse than talking normally from 20 cm.
• Pause the noise for 20 seconds. Stopping a hammer drill for the length of a note costs you nothing and removes the only thing that reliably breaks transcription: noise louder than your voice.
• Slow the ends of sentences. People trail off. That is where errors cluster.
• Speak at your normal pace. Over-enunciating hurts modern models; they are trained on natural speech. Speak like you are explaining it to an apprentice standing next to you.

Trade jargon, brand names and part numbers

This is where dictation earns its reputation for being useless, and it is fixable. Transcription is a language model as much as an audio model: it predicts words that go together. Real words in real sentences come out right. Random strings of digits and unusual brand names do not.

• Give jargon a sentence to live in. "RCD" alone can come out as "our CD". "The RCD tripped on test" comes out right, because the surrounding words anchor it.
• Read part numbers in pairs, not as one blur. "Twenty-one, forty-four, sixty-two" transcribes far more reliably than six digits fired off at speed. Then check it. Always check numbers.
• Spell unusual brand names with the phonetic alphabet the first time in a note: "Wago, whisky alpha golf oscar." It feels daft. It works.
• Say units out loud. "Millimetres", "metres", "amps". "Sixteen" on its own is useless a week later.

Swedish product names and terms, for readers outside Sweden

Swedish trade vocabulary is a hard case worth understanding even if you never work there, because the same problem hits any non-English term you use daily. Swedish electricians work with a jordfelsbrytare (RCD), a gruppcentral (consumer unit), dvärgbrytare (MCBs) and kopplingsdosor (junction boxes), and they order material by E-nummer, a national article number for electrical goods; plumbers use the RSK number for the same purpose. Two rules follow:

• Set the app language to the language you speak, not the language you write invoices in. A Swedish-language engine handles jordfelsbrytare and gets it right. An English one hears "your fell breaker". If you code-switch mid-sentence, expect mangling of the minority language.
• Do not dictate article numbers into free text at all. Scan them, pick them from your price list, or photograph the box. An E-nummer or RSK number that is one digit wrong is worse than no number, because nobody knows it is wrong until the wrong part turns up.

The general principle: dictate prose, capture identifiers by scanning or picking from a list.

Dictating egenkontroll and inspection answers

In Sweden, a registered electrical firm has to run an egenkontrollprogram — a documented self-inspection routine — under the electrical safety rules overseen by Elsäkerhetsverket. The nearest equivalents elsewhere are an Electrical Installation Certificate in the UK or a Certificate of Compliance in New Zealand. Different paperwork, same reality: tick boxes plus a comment field that everyone leaves blank.

The comment field is exactly where voice belongs, and the tick boxes are exactly where it does not.

• Tap the boxes, dictate the deviations. Yes/no answers are one thumb tap and zero ambiguity. Speaking "yes" into a checkbox is slower and can be misheard.
• For a deviation, say what, where, why it matters, and what you did. "Deviation: earth continuity high on the bathroom circuit. Traced to a loose terminal in the junction box above the ceiling void. Re-terminated and re-tested, now within limits."
• Never dictate a pass you have not measured. A voice note is a record. It carries the same weight as anything you typed, and a fast, confident, wrong sentence is a liability.
• Photo first, then talk. A picture with 15 seconds of narration attached is the most complete evidence you can produce on site.

Dictating quote basis while you are still standing there

The best time to capture a quote is on site, in front of the thing, before the details go soft in your head. You are not dictating the quote. You are dictating the basis — what the office needs to price it.

Cover five things: access, existing condition, scope, unknowns, and gut feel on time.

"Quote basis, Lindberg, terraced house, ground floor. Access is fine, off-street parking. Existing board is a rewireable fuse box, no RCD protection. Scope: full board change, twelve ways, plus two new sockets in the kitchen. Unknown: cannot see the meter tails until the board is off, might need the supplier out. Gut feel: one day, two of us, unless the tails are a problem."

Under a minute. The "unknown" line is the one that saves money, because it is the thing that turns a profitable quote into a loss and the thing you will have forgotten by the time you sit down to price it.

When typing is still faster

Voice is not always the answer, and pretending otherwise is how people lose faith in it. Rough guide: you speak at roughly 130 words a minute and thumb-type maybe 30, so anything over a sentence favours voice. But speed is not the only cost. Correcting a mangled transcript is slower than typing it right, and some content is high-cost when wrong.

The bit everyone skips: read it back

Spend five seconds looking at the text before you leave the field. Not a proofread — a glance at the numbers and the names. That single habit is the difference between voice notes that the office trusts and voice notes that the office quietly re-types.

Also be deliberate about what you say out loud. A dictated note is stored data and falls under GDPR like anything else. Do not narrate a customer's personal circumstances, their alarm code, or where they keep the key. Describe the installation, not the household.

A ten-minute setup

• Set your phone's dictation language to the language you speak on site. If you work bilingually, know how to switch it in one tap.
• Add a home-screen shortcut, or use the assistant button, so recording starts in under two seconds. If it takes more than that, you will not use it.
• Record in the app that owns the job, not in a notes app. A voice note that is not attached to a job card is admin you have moved, not admin you have removed. In OdinTask, transcription is part of the included AI assist and drops straight into the job card, egenkontroll or quote you are standing in — no separate app, no keys to configure.
• Practise the four-part script on three jobs. By job four you will stop thinking about it.

That is the whole method. Structure beats software: an app with a mediocre engine and a disciplined four-line script beats a perfect engine and 90 seconds of rambling, every time. Start a free trial if you want the notes to land where the work already lives.