A good field sales visit report holds nine things: who you saw, when, what was discussed, what was agreed, any order or quote, objections, competitor mentions, the next step with a date, and a rating of the account's temperature. AI can build that from a two-minute voice note, then a person checks it before it reaches the CRM.
The reason reports are not written is simple: reps drive to the next call, and typing in a car park is the first thing to go. Speaking takes a quarter of the time. The risk is that the transcript is long and rambling, so the value is in a fixed template and a prompt that turns speech into fields, rather than in the transcription itself.
The nine fields a visit report needs, and why each is there
Resist the temptation to add fields. Every extra one is a thing the rep forgets to say and the AI has to guess. These nine cover what a manager, a colleague taking over the account and the CRM's reports need.
| Field | What it holds | CRM home |
|---|---|---|
| 1. Account and contact | Business name, person seen, their role | Company and contact records |
| 2. Visit date and type | Date, first visit or repeat, in person or call | Activity (meeting) |
| 3. Discussion summary | Three sentences at most, in plain facts | Note |
| 4. Agreed | What both sides said yes to | Note, deal stage |
| 5. Order or quote | Products, quantities, values, terms | Deal amount and line items |
| 6. Objections and concerns | Price, stock, terms, shelf space, anything blocking | Note or custom property |
| 7. Competitors seen | Which rival products are on the shelf, at what price | Note or custom property |
| 8. Next step | What, who, by when | Task with a due date |
| 9. Account temperature | Hot, warm or cold, with one reason | Deal stage or a rating property |
The last field is the one reps resist and managers use most. A single word and a reason, such as "warm: likes the range, waiting on a new buyer", turns a stack of reports into something you can scan on a Monday. It is also where people differ, so write down what hot, warm and cold mean for your business: hot might mean an order within 30 days, warm a decision within 90, cold no date at all.
How a rep should record the note
Any phone's voice recorder will do, and a note recorded in the car straight after the visit is more accurate than one recorded by Friday evening. The way to get clean output from the AI is to give the rep a speaking order that matches the template. It fits on a sticker on the dashboard:
- Say the date, the business and who you saw.
- Say what you talked about, in a few sentences.
- Say what was agreed.
- Say any order or quote, with products and numbers.
- Say what held them back, and what competitors you saw.
- Say what happens next, who does it, and the day.
- Finish with hot, warm or cold, and why.
Ask reps to say dates as dates ("Thursday the 14th") and not as "next Thursday", because the AI cannot know when the note was recorded unless you tell it. Ask for numbers to be said slowly, and for names of people and products to be spelt out when they are unusual. These two habits remove most of the errors described later.
Turning audio into text is a separate step from structuring it. Many phones and chat assistants transcribe a recording; some CRMs and call tools now include transcription. Whichever you use, check that you can use it with customer information under your own rules. Business plans of the main chat assistants do not train on business content by default; check that setting on the plan your reps use. Moving a transcript into the template is where the AI does the real work.

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 nowThe prompt that fills the template
Paste the transcript below the instructions. Give the AI the visit date so it can resolve "tomorrow" and "next week".
You turn a field sales rep's transcribed voice note into a visit report.
The visit date is [date]. The rep is [rep name]. Our company sells [product range].
Return exactly these fields, in this order:
1. Account and contact (business, person, role)
2. Visit date and type
3. Discussion summary (max 3 sentences, facts only)
4. Agreed
5. Order or quote (product, quantity, price if stated, terms)
6. Objections and concerns
7. Competitors seen
8. Next step (action, owner, due date as a calendar date)
9. Account temperature (hot, warm or cold, with a one-line reason)
Rules:
- Use only what is in the transcript. If a field was not mentioned, write "Not stated". Do not infer.
- Write numbers and names exactly as spoken. If a spoken number is unclear, write it and add "(check)".
- Convert relative dates ("Thursday", "next week") to calendar dates using the visit date, and say that you did.
- Leave out personal comments about people that are not about the business.
- Do not add recommendations.
The "Not stated" rule is the most valuable line. Without it, the assistant fills empty fields with plausible sentences, and a report full of plausible filler is worse than one with gaps, because nobody can tell which parts are real.
One retailer visit, from recording to filled template
An illustrative case: a small stationery maker sells notebooks, planners and cards to independent gift shops. A rep visits a gift shop on Tuesday 10 November and records this in the car (lightly edited from what a transcription tool would produce, with the verbal clutter left in):
Okay, so it's Tuesday the tenth, just seen the owner at the gift shop on the high street, her name's [first name], I think she's the owner, she runs the buying. We talked mostly about the autumn range, she liked the A5 planners, said the cover colours were good for gifting. She took twelve of the A5 planners and six of the card packs, so that's twelve at eleven fifty trade and six at, I think, four ninety, call it. Payment thirty days, which is our standard. Her worry was the minimum order, she wants to try the small notebooks but not at a box of fifty. She's got a competitor range on the front table, a, uh, brand she called [rival brand], notebooks at about nine ninety retail. I'm going to send a sample pack of the small notebooks, I'll post it Thursday, and call her the Wednesday after to see what she thinks. She's warm, I'd say, she bought something and is open to more.
Run through the prompt, the structured report might look like this (illustrative output):
1. Account and contact: Gift shop (name not stated). [first name], owner and buyer.
2. Visit date and type: 10 November (Tuesday), in-person visit; first or repeat not stated.
3. Discussion summary: Discussed the autumn range. The owner liked the A5 planners, particularly the cover colours for gifting. She is interested in the small notebooks but not in a box of 50.
4. Agreed: Order placed; payment terms 30 days (standard). Sample pack of small notebooks to be sent.
5. Order or quote: 12 x A5 planner at 11.50 trade (check); 6 x card packs at about 4.90 (check). Terms: 30 days.
6. Objections and concerns: Minimum order on small notebooks is too high (box of 50).
7. Competitors seen: "[rival brand]" notebooks on front table, about 9.90 retail (check spelling of brand).
8. Next step: Rep posts sample pack of small notebooks Thursday 12 November (converted from "Thursday"); rep calls [first name] Wednesday 18 November (converted from "the Wednesday after").
9. Account temperature: Warm. Bought, open to more.
Read it as a manager would. Four things need a person. The business name was never said, so "name not stated" is correct, and the rep fills it in. The prices were spoken loosely ("I think", "call it"), and the (check) flags are the AI doing its job; the rep confirms them against the order. The brand name might have been spelt differently. And the order value, 12 x 11.50 plus 6 x 4.90, is 138.00 plus 29.40, or 167.40, which the AI did not total because the prompt did not ask it to. Add "calculate the order total" only if you are prepared to check the sum, because language models make arithmetic errors; why AI gets sums wrong in quotes explains the habit to build.
The AI saved the rep ten minutes of typing and gave the manager a record the CRM can use. The rep spent about two minutes correcting three flags before pushing it on.
Where each field goes in HubSpot or Pipedrive
The report is only useful once it is attached to the right account. The two CRMs name things differently, but the structure is the same: a company, a contact, a deal for money in progress, a note for history and a task for the next step.
| Report field | HubSpot | Pipedrive |
|---|---|---|
| Account and contact | Company and contact records, associated to the deal | Organisation and person, linked to the deal |
| Visit | A logged meeting or a note on the contact timeline | A completed activity of type meeting, or a note |
| Summary, agreed, objections, competitors | One note, with headings, or custom properties for objections and competitors | One note on the deal, or custom fields |
| Order or quote | Deal amount, close date and line items | Deal value and products |
| Next step | A task with a due date, assigned to the rep | A scheduled activity with a due date |
| Temperature | Deal stage, or a dropdown property | Deal stage or a custom field |
Most of the field work is a single note. Custom fields for objections and competitors are worth adding only when you want to report on them: a field you can filter by ("show me every account where the competitor is [rival brand]") is a different thing from text in a note. Keep the custom fields to the ones your managers will use, and leave the rest in the note. HubSpot's free CRM allows two users, so a two-person team can run this at no CRM cost; larger teams need a paid plan on whichever CRM they pick, and prices change often, so check the vendor's page. HubSpot's Starter Customer Platform lists at $20 per seat per month as of the last check, often with promotional first-year pricing, which is why the live page matters more than any article.
Three ways to get the report into the CRM
Choose by volume and by how much you trust the output. Start with the first.
1. Copy and paste, with a person in the loop
The rep pastes the AI's report into the CRM note and adds the task by hand. It takes about a minute, costs nothing, and the rep sees every line. Use it for the first two weeks whatever your plans, since errors surface while they are cheap.
2. An automation tool between a form and the CRM
Ask the AI to return the report as a list of labelled fields, and use a form or shared document as the hand-over. An automation tool such as Zapier then creates the note and the task in the CRM. A free Zapier plan allows 100 tasks a month and two-step workflows only, which is enough to test one rep, but a flow that includes an AI step, a note and a task needs a paid plan. Count the steps before you choose. Zapier, Make and n8n compared on cost shows how the pricing grows.
3. The CRM's own AI features
HubSpot, Pipedrive and most other CRMs advertise built-in AI or call-summary features, but which ones exist, and on which plan, changes often. Pipedrive's pricing page, for example, lists Pipedrive Nova from its Lite plan upwards: a meeting-intelligence tool that gives a brief before a meeting, a transcript of each call and ready-to-approve CRM updates afterwards. Check whether a tool like that can take a voice memo recorded on a phone after an in-person visit, or only calls and online meetings it joins. If your CRM does it, trial it against your template and see whether it produces the nine fields or only a generic summary. If it gives a summary only, keep the prompt above in front of it.
Whichever you choose, keep the person who reviews the report. The rule that holds for every route is that automation moves data and a person approves anything that changes a deal. A broader view of keeping the CRM up to date after calls is in updating the CRM automatically after sales calls.
What voice notes get wrong, and how to catch it
Recordings in cars, shops and car parks produce specific errors. These are the ones to train for.
- Numbers. "Eleven fifty" might be 11.50, 1,150 or 11 and 50 of something. The prompt's "(check)" rule and a rep reading the order lines against the order form fix this.
- Names. Customer and competitor names are mis-transcribed, especially unusual spellings. Spell them in the note or confirm them from the CRM.
- Dates. "Next Friday" is ambiguous even for people. Passing the visit date and asking the AI to say when it converted a date lets a manager see the assumption.
- Sentiment. The AI can rate a visit warmer than the rep felt. Take the rep's own word for the temperature, and treat the AI's reason as a summary of what they said.
- Merged visits. Reps who record three visits in one note get one confused report. One note per account.
A manager's quick check each Friday catches most of it: open five reports at random and read each against what you know about the account. If two or more have a significant error, tighten the prompt or retrain the speaking order.
A before-and-after for the rep and the manager
Before: a rep with six calls a day typed three sentences into the CRM at the end of the week, from memory, for the calls they remembered. The manager saw "Good visit, will follow up" against a dozen accounts and could not tell which were worth a second visit.
After: each report carries the same nine fields. The manager filters Monday's pipeline for "warm" accounts with a next step due this week, reads two lines on each, and sees that three accounts raised the same objection about minimum order size. That one pattern, visible only because objections are in a consistent field, is worth more than the time saved: the maker might offer a smaller starter box.
A quick sum, illustrative: if a rep makes six visits a day and each report takes eight minutes to type, that is 48 minutes. With a two-minute voice note and a two-minute review, it is 24 minutes, and the reports are more complete. Over a five-day week, a rep gains two hours. The figure is only worth believing once you time your own reps for a week, so do that before and after.
A weekly roll-up the manager can run on a folder of reports
Once reports are consistent, the AI can summarise a week of them. Paste the week's reports, with names removed if you prefer, and ask:
Here are this week's visit reports (one per account). Please:
1. List accounts by temperature, with the next step and its due date.
2. Count how often each objection appears and list the accounts for each.
3. List each competitor named and the accounts where it was seen.
4. Flag any account with no next step, or a next step overdue.
Use only what is in the reports. Do not add advice.
Check the counts against your own tally for the first few weeks, since counting is a task language models do imperfectly. The roll-up is also a good input to account tailoring; personalising outreach to each trade account shows how the notes feed the next message.
A visit with no sale: the report that matters most
Reports for visits that produced an order write themselves. The ones that teach you something are the visits that did not. Take a second, shorter illustration: a rep calls on a garden centre that stocked the range two seasons ago and has not reordered. The recording says, roughly, "Saw the buyer, a new person since last year. Said the old buyer set the range up and she is cutting suppliers to cut admin. Didn't order. She said the planners sold slowly because they were on the back wall. I saw our cards by the till doing fine. Nothing agreed. I'll send her the lighter-weight display stand details and ring in January. Cold, but not dead."
A good report from this keeps the facts a manager can act on. Agreed: nothing. Objection: supplier consolidation, and poor placement of the planners. Competitors seen: not stated. Next step: send the display stand details, call in January (the AI should write a calendar date only if the rep gave one, so it asks for one). Temperature: cold, with the reason "buyer is cutting suppliers, cards selling at the till". This is not a failed record. It tells the manager that one account's problem is shelf position, which the company can fix with a stand, and that the account is not lost to a rival.
It also shows why "Not stated" is a feature. The AI reports "Competitors seen: Not stated" and does not invent a rival, and the manager does not read a made-up competitor into the pipeline.
A blank copy of the template to paste into your CRM or notes app
If you start with plain notes before any automation, give reps the template in this shape, which can be pasted as a note and filled by hand or by the AI.
VISIT REPORT
Account / contact:
Date and type:
Summary (max 3 sentences):
Agreed:
Order or quote (items, quantities, price, terms):
Objections:
Competitors seen:
Next step (action, owner, date):
Temperature (hot / warm / cold) and why:
Hold the template in one place, such as a pinned note in the CRM, so that the prompt, the speaking order and the CRM note always agree. When you change a field name, change it in all three the same day.
Keeping the record fit to be read by others
A CRM note is a business record. It may be read by a colleague who takes over the account, by a manager, and, depending on the law where you operate, by the person it describes if they ask to see what you hold about them. That is a good reason to keep the notes about the business. Write facts about orders, products, timing and decisions. Leave out remarks about someone's health, family or temperament, and tell reps so. Add one line to the prompt to make the AI do the same, as in the version above. If you have customers in the EU, data-protection rules apply to contact details in the CRM, so keep contact records to what the account needs, and ask a qualified adviser about how long to keep them.
Rolling it out to a team of reps without a revolt
Reps accept new reporting when it saves them time and when it is not used against them. Do three things:
- Pilot with the two reps who dislike paperwork most. If it works for them, it works for everyone.
- Tell the team plainly what the reports are used for, and what they are not used for. If notes will be read in performance reviews, say so.
- Keep the template fixed for a month. Changing fields weekly makes the AI's output and the reps' habits unstable.
Give each rep a printed card with the speaking order, a copy of the prompt saved in their assistant, and a rule that every note is reviewed before it goes into the CRM. Add the five-report Friday check for the first month, and the report template becomes part of how the team works.
Questions about voice-note visit reports
How long should a rep's voice note be?
Two to three minutes covers most visits. Shorter notes miss the next step, longer ones drift into stories the CRM does not need. A fixed spoken order, with who, what was agreed and what happens next, keeps it short and makes the AI's job easier and more accurate.
Can the AI write straight into HubSpot or Pipedrive?
Yes, through an automation tool or the CRM's own AI features, but start with copy and paste for two weeks. You learn which fields the AI gets wrong before an error reaches the pipeline. Automate only once the report format has stopped changing and a person reviews the first ones.
What if a rep mentions something sensitive on the recording?
Teach reps to keep personal comments about people out of the note, since it ends up in a record the customer could ask to see. Review a sample of reports in the first month, and tell the AI to leave out anything not related to the account's business.
Do I need a paid CRM for this?
No. HubSpot's free CRM allows two users, most CRMs offer a free trial, and a shared spreadsheet works for a very small team. The report template matters more than the tool. Choose a CRM when you need reminders, pipeline views and a place to keep every visit against the account.
Further reads
- How to Turn Sales Call Recordings Into Coaching Notes With AI — Reuse recordings as coaching material for new reps.
- AI Tools for Car Sales: Test-Drive Follow-Ups, Quotes and Call Notes — See how a different sales team handles call notes and follow-ups.
- AI Business Development Software: Do You Need It or a CRM? — Decide whether you need a CRM or only a better routine.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: HubSpot's Starter pricing page and knowledge base (logging activities, products and line items), Pipedrive's pricing page and knowledge base (activities, products), Zapier's pricing page, and general field-sales reporting practice. Check each CRM's current plan before you rely on a feature named here.