Cut Dispatch Notes and Customer Questions for Your Trucking Company with Claude Code

6 a.m., the dispatcher’s desk. Ten text-message scribbles that drivers fired off the night before. Three slips of paper with hurried notes from shippers calling to ask “where’s my load right now?” And a dispatch board covered in shorthand that only one person on the planet can read.

The first time I helped out at a trucking company, this is what stopped me cold. The dispatching itself wasn’t the bottleneck. It was the hours spent turning scribbles into something a human can read, and the hours spent explaining the same answer to the same question over and over. Together, those two things ate half the day.

And that shorthand? The moment the dispatcher takes a sick day, nobody else can decode it. “4t reefer, Newark 3 runs, deadhead back, drop Edison” — would the person at the next desk understand that? Probably not.

Today I want to talk about making two of those jobs lighter: cleaning up dispatch notes and handling routine questions. We’ll use Claude Code, a tool that lets you run generative AI from the command line. This is not about letting AI decide your routes. It’s about cutting the busywork that sits in front of the real decisions.

Key takeaways

• For a trucking or logistics dispatcher, the two biggest time sinks — rewriting notes into clean form, and answering the same question repeatedly — can be cut by more than half with generative AI.
• Do not hand the dispatch decision itself to AI. Hand it the work of tidying up, drafting, and flagging gaps — nothing more.
• Below are three copy-paste prompt templates plus a check script that turns dispatch notes into a clean table and warns you about missing fields.
• Shipper names, phone numbers, license plates — anything personal needs a company rule and an anonymization step before it touches AI.
• If 90 minutes of daily busywork drops to 30, that’s roughly 20 hours a month. I’ll show why that’s a realistic number per dispatcher, not a sales pitch.

Who this is for, and what the day looks like now

I’m writing for the person running dispatch at a small or mid-size trucking company. Ten to fifty trucks. Mostly repeat customers as shippers. You talk to drivers by phone and text. You probably have a dedicated dispatch system, but in the end you build the schedule with notes and what’s in your head.

A typical day runs like this:

1. From late evening into early morning, requests and changes for the next day arrive piecemeal from drivers and shippers.
2. The dispatcher reads them and assigns trucks and people.
3. The assignments get written up cleanly onto the board, and each driver gets contacted.
4. Through the day, questions trickle in: “where’s that load,” “can we push the pickup an hour?”
5. After hours, you log what actually happened and get ready for tomorrow.

Of all that, only step 2 needs real judgment. Steps 1, 3, 4, and 5 are “organize the information” and “say the same thing again” work — and that’s exactly where AI shines.

The usual rework and headaches

Here are the snags I’ve watched happen again and again on the floor. At least one of these is true at your company.

• Waiting on shorthand to be decoded: only the original dispatcher can read it, so when they’re out, work stops.
• Double-messaging gone wrong: a change told to the driver never reaches the shipper, and times collide on site.
• “You never told me” disputes: a promise made over the phone leaves no record, and it comes back to bite you.
• Copy-paste accidents on the board: yesterday’s board gets reused and the date or run number doesn’t get updated.
• New hires never grow: the veteran’s knowledge lives in their head, unwritten, so it can’t be handed over.

None of these need a “smart” AI. They all disappear once you have a helper that puts things into a fixed format and points out what’s missing.

Use case 1: Turn shorthand dispatch notes into a table anyone can read

This is where the payoff was biggest. Paste in the shorthand a driver sent you, and have it rebuilt as a table.

For example, you hand it an input like this:

4t reefer Newark 3 runs deadhead back drop Edison / 2t Port pickup 8am produce market return mixed freight / 10t Bayonne container 2pm chassis pool

You ask for a table with columns for truck class, run, pickup, cargo type, and return leg. Nobody has to expand each abbreviation by hand. The prompt template is below.

Before going further, let’s nail down what AI gets and what a human must always decide.

Step	What AI handles	What a human must decide
Cleaning up notes	Expanding shorthand, building the table, flagging missing fields	When shorthand is company-specific, the final check that it expanded correctly
Assigning trucks	Listing candidates and available vehicles	Which truck gets which run — the final call
Writing messages	Drafting the driver-facing and shipper-facing text	The fact-check before hitting send
Logging results	Standardizing the format, transcribing	Whether the numbers and times reconcile

A human holds the assignment decision. Do not give that one up. AI can tell you “the Edison drop has an empty return leg,” but only someone who knows the shipper relationships and the rates can decide what to load on that empty truck to actually make money.

Use case 2: Draft the replies to customer questions

The second job is handling questions. “Where’s that load right now?” “Can we delay the pickup by an hour?” Composing each reply from scratch quietly burns time.

Here you teach the AI a handful of your common questions and answers up front, and when a new question comes in, it produces a draft. A human decides whether to send it.

Run it off a checklist and it won’t drift.

• Did it summarize the question in one line?
• Does the draft reply contain the facts (time, place, run number)?
• Do those facts match the dispatch board and the system? (A human verifies this.)
• Did anything that must not leave the building — shipper names, phone numbers — slip into the text?
• Did the dispatcher read it once under their own name before sending?

Items 3 and 4 are non-negotiably human work. AI will confidently write “arriving around 2 p.m.,” but whether 2 p.m. is actually true is something AI has no way of knowing.

Use case 3: A second pair of eyes on the dispatch board

The third use is having it scan a finished board for gaps. People are bad at catching mistakes in tables they built themselves. A date you forgot to change, the same driver booked at two places at once, a blank return leg. Mechanical checks like these get caught when you run the board through AI once.

Again, no decisions here. You have it flag “driver #7 has two jobs at 9:00 and 9:30 — geographically impossible?” and a human does the fixing. This alone cuts down those copy-paste accidents a lot.

Copy-paste prompt templates

Here are three you can use as-is. Swap in your own company names and shorthand glossary.

Template 1: Turn shorthand notes into a table.

You are a dispatch assistant for a trucking company. Turn the dispatch notes below into a table.
The columns are: Truck class / Run / Pickup / Cargo type / Return leg / Unclear.
Expand shorthand using common trucking terms; put anything you cannot determine into "Unclear."
Do not assign trucks on your own and do not add runs that were not written down.

Notes:
(paste the shorthand notes here)

Template 2: Draft a reply to a customer question.

You are a draft assistant for a trucking company's customer support.
For the question below, write a polite but concise draft reply.
Where a fact such as a time or place is needed, do not assert it — leave a
placeholder like [VERIFY: arrival time]. Do not invent facts.

Question:
(paste the customer question here)

Template 3: Double-check the dispatch board.

Inspect the dispatch board below and list only the inconsistencies, as bullet points.
Look for: the same driver double-booked in time, geographically impossible pickup time and place,
blank return legs, and mismatched dates or run numbers.
You do not need to rebuild the schedule. Only point out the problems.

Dispatch board:
(paste the dispatch board here)

A check script: convert shorthand notes into table data

Verifying the prompt’s output by hand every time is a chore, so here’s a small script that converts the notes into structured data and checks for gaps mechanically. It runs anywhere you have Node.js. It reads the AI’s output (tab-separated) and warns about any row still marked unclear — plain, but it earns its keep.

import { readFileSync } from "node:fs";

// Have the AI return tab-separated rows: "class\trun\tpickup\tcargo\treturn\tunclear"
const raw = readFileSync(process.argv[2] ?? "dispatch.tsv", "utf8");
const cols = ["class", "run", "pickup", "cargo", "return", "unclear"];

const rows = raw
  .trim()
  .split("\n")
  .map((line) => line.split("\t"))
  .map((cells) => Object.fromEntries(cols.map((c, i) => [c, (cells[i] ?? "").trim()])));

let warnings = 0;
for (const [idx, r] of rows.entries()) {
  const n = idx + 1;
  if (!r["pickup"]) { console.log(`Row ${n}: pickup is empty`); warnings++; }
  if (!r["class"]) { console.log(`Row ${n}: truck class is empty`); warnings++; }
  if (r["unclear"]) { console.log(`Row ${n}: needs review -> ${r["unclear"]}`); warnings++; }
}

console.log(`---\n${warnings} of ${rows.length} rows need review`);
if (warnings > 0) process.exitCode = 1;

Running it is one line.

node check-dispatch.mjs dispatch.tsv

This stops the accident where “an unclear note flows straight onto the board.” If even one row needs review, the exit code is 1, so wiring it into an automated flow forces a human to look before anything moves.

Before and after

The effect is clearest in numbers. These are rough figures per dispatcher.

Task	Before	After
Cleaning up shorthand notes	40 min in the morning	About 10 min (paste plus check)
Composing question replies	5 min each	1–2 min each (draft plus edit)
Hunting board mistakes	Found as rework	Flagged on the spot

Say you save 30 minutes on cleanup and 3 minutes each on 20 questions — that’s roughly 90 minutes a day. Across 20 working days, about 30 hours freed up a month. If a dispatcher’s time is worth, say, $30 an hour, that’s around $900 a month in value. Even after subtracting setup effort, a few hundred dollars of breathing room a month is a realistic line.

What you do with the freed time is up to the company, but on the floors I’ve seen, the big win was turning it into “the veteran sitting beside a new hire, teaching.”

Security and personal data

This part is non-negotiable precisely because you’re a trucking or logistics company. Dispatch notes carry shipper company names, delivery addresses, driver names, license plate numbers, and sometimes the contents of the freight. That’s bona fide personal and commercial information.

At a minimum, follow these:

• As a company, decide first which generative AI service, on which plan, may be used for work. Don’t paste business data into someone’s personal free account.
• Anonymize shipper names, personal names, and phone numbers before they reach the AI — turn them into “Company A,” “Contact B.” Putting a replace step ahead of the script above makes this safer.
• Confirm that, by setting or contract, the data you submit isn’t used for training.
• Don’t send the output straight outside. Along with the fact-check, have a human confirm no personal data crept in.

The thinking behind anonymization and the rules around it are best worked out together with your operational design. How to teach Claude Code your rules is covered in claude-md-best-practices, and a safe on-ramp for non-engineers is in claude-code-for-non-engineers. For the very first step, claude-code-getting-started-guide is the easiest read. As a reference point for handling data, the U.S. FTC’s guidance on protecting personal information is worth a look.

FAQ

Q. Can I hand the whole dispatching job to AI? No. You hand it organizing, drafting, and flagging — that’s it. Which truck gets which run is a judgment that involves rates, shipper relationships, even a driver’s condition that day. AI doesn’t know far too many of those things. Keep the decision with a human.

Q. Will it understand our company’s shorthand and house rules? Not out of the box. You either pass the shorthand glossary in the prompt or keep your house rules in a file the AI reads each time. For how to teach it, claude-code-prompt-engineering-advanced is a good reference.

Q. Can a dispatcher who isn’t great with computers use this? If it’s just pasting a prompt and checking the result, it’s not hard. Start with copy-paste, and once you’re comfortable, move on to automating with the script. Day-to-day tips are in claude-code-productivity-tips.

Q. How much does it cost to get started? Generative AI usage runs from a few dollars a month. The bigger cost isn’t money — it’s the effort of writing internal rules, and skipping that leaves a real personal-data risk.

What happened when I actually tried it

On my own machine, I made up 15 dummy shorthand notes and 10 questions and ran the whole flow.

After turning the notes into a table with Template 1 and running them through the check script, three rows came back with “pickup empty” and “return leg blank.” Those were spots the original notes genuinely left out — places I’d missed when I eyeballed them myself. The mechanical gatekeeper caught them.

The reply drafts from Template 2 kept the [VERIFY: arrival time] placeholder properly in place. Even when I tried to trick it into asserting a fact with a leading question, it didn’t invent a time, which was reassuring.

I also hit a limit. Company-specific shorthand — a code for our own warehouse, say — gets dropped straight into “Unclear” unless you hand over the glossary. That’s actually the correct behavior; safer than being guessed at.

The bottom line: keep the dispatch decision with a human as always, and hand the work in front of it — tidying, drafting, flagging gaps — to AI, and the busywork gets noticeably lighter. If you want to set up the rules company-wide first, we can work out an approach that fits your floor through training and consulting.