AI in the Fishing Industry: Catch Logs, Buyers and Boat Paperwork

A small fishing business can realistically use AI for four paperwork jobs: turning catch notes into a tidy sheet, checking buyer invoices against agreed prices, drafting price and availability messages, and planning engine and gear maintenance. It cannot decide quotas, licences or safety. Those stay with your regulator and your own judgement.

The gain is evenings, not catches. A skipper who lands for four days, then spends two evenings reconciling notebook pages with invoices, is the person this helps; how much time comes back depends on how messy the notebook is, so measure it over two weeks before trusting any estimate. Everything here assumes one boat with a notebook, a phone and a few regular buyers, not sea-wide sensors or satellite analytics.

Follow me on Instagram@sagnikteaches

What one boat's paperwork looks like in a month

Before choosing a tool, count what actually lands on the kitchen table. For a day boat or small netter with a crew of two or three, the pile is surprisingly consistent. There is a catch record for each trip: date, area or ground, gear, species, weights, sometimes grades. There are buyer slips and invoices. There are price messages, usually texts or calls, about what the buyer will pay this week. There is fuel, ice and bait receipts, plus engine hours, service dates and gear repairs. And there are periodic returns and declarations that the regulator asks for.

Connect on LinkedInSagnik Bhattacharya

AI is good at three things in that pile: reading messy text and putting it into columns, comparing two lists and pointing out where they disagree, and drafting wording you would otherwise write from scratch. It is poor at anything that needs the official record, a current rule, or a number it was never given. That split decides every job below.

Subscribe on YouTube@sagnikteaches

A quick audit helps. Over one week, write down each paperwork task, how many minutes it took and how often it goes wrong. The tasks that take 20 minutes or more and happen every trip are the candidates. The ones that happen once a year, such as a licence renewal, are not worth automating, though an AI can still help you draft a covering note.

Catch notes into a sheet you can total

The first job is the one with the quickest payback. Most small boats keep a notebook, a whiteboard in the wheelhouse, or voice notes. The aim is a spreadsheet with one row per species per trip, so that weekly and monthly totals become a click rather than an evening with a calculator.

Set the columns once

Decide the columns and keep them fixed: trip date, ground (your own short codes), gear, species, grade, weight in kilograms, buyer code, price per kilogram agreed, and notes. Fixed columns matter because the assistant will happily invent new ones if you let it, and a sheet that changes shape every week cannot be totalled.

Photograph the page, then paste the text

Take a clear photo of the notebook page or whiteboard, upload it to a chat assistant with image reading, and ask for a table in your columns. Here is a prompt you can reuse as it stands:

You are helping a small fishing boat keep a catch record.
Read the attached photo of my notebook page. Return a table with these
columns only: date, ground code, gear, species, grade, weight_kg,
buyer_code, notes.
Rules:
- One row per species per trip.
- Do not guess. If a figure or word is unclear, write [CHECK] in that cell.
- Do not convert units. If I wrote lb, keep it and put lb in notes.
- After the table, list every cell you marked [CHECK] and why.

The sentence about not guessing is the important one. Without it, a smudged weight turns into a confident number. With it, you get a short list of cells to look at against the page, which takes a minute.

An illustrative result for a Tuesday trip might look like this:

Here the assistant did its job: it refused to pick between 86 and 36. You check the page, see 86, fix the cell, and move on. If the correction rate stays under about one cell in twenty, the method is paying for itself. If it is much worse, the handwriting needs block capitals or you should speak the figures into a voice note and transcribe those instead.

Let the sheet do the arithmetic

Use the spreadsheet's own formulas for totals, not the chatbot. A weekly total by species is a single pivot table or a SUMIFS formula, and it will be right every time. If you do not know how to build one, ask the assistant to write the formula for your actual column letters and then test it on three rows you can add up by hand. The same warning applies throughout: the assistant is a good drafter of formulas and a poor calculator, a point covered in why AI is bad at maths and what to check in quotes and invoices.

Checking a buyer's invoice against what you agreed

Once the catch is in a sheet, the second job writes itself: does the invoice match the landing? Discrepancies in small-boat sales are rarely dishonest. They are rounding, a grade changed at the quay, a deduction for ice or boxes that was not mentioned, or a price that moved after a text conversation nobody saved.

The main worked example, with numbers

Say a two-person potting boat sells to one regular buyer, B1. Over a week the boat lands three trips of brown crab. The agreed prices, taken from the buyer's text on Monday, are $3.80 a kilogram for large and $2.90 for medium. The sheet shows:

Your own sum: large is 120 + 142 + 98 = 360 kg, and 360 x 3.80 = $1,368.00. Medium is 74 + 86 + 61 = 221 kg, and 221 x 2.90 = $640.90. Expected total: $2,008.90.

The buyer's invoice arrives for $1,931.40 with a line saying "crab, 581 kg, mixed". The weights add up (360 + 221 = 581), so the quantity is fine. But $1,931.40 divided by 581 is about $3.32 a kilogram, which is neither of the agreed prices. A blended price of $3.46 would have given the expected total, so the invoice is $77.50 short. You would never spot that by glancing at the page, but you would spot it by asking the assistant to compare.

Paste the invoice text and your sheet rows, with buyer names replaced by codes, and use a prompt like this:

Below are (1) my landing sheet for the week, (2) the prices I agreed
with buyer B1 by text, and (3) B1's invoice. Do three things:
1. Recalculate what the invoice should be from my sheet and the agreed
   prices, showing each line.
2. List every difference between my figure and the invoice, in dollars,
   and say which line it comes from.
3. Draft a short, polite message asking B1 to explain the difference,
   quoting only figures I gave you.
Do not assume anything I did not give you. If a price or grade is
missing, say so instead of filling it in.

A good answer breaks the invoice into the missing grade split, shows the $77.50 gap, and suggests two likely causes: part of the large crab was invoiced at the medium rate (86 kg at 90 cents less would account for $77.40 of it), or a deduction for boxes was taken without being listed. It should not tell you which one it is. Only the buyer knows. Your message might read, in the assistant's draft after you edit it: "Hi [first name], thanks for the invoice. Our landing sheet for the week gives 360 kg large and 221 kg medium, which at the prices we agreed on Monday comes to $2,008.90. The invoice shows $1,931.40. Could you send the line breakdown so we can see where the $77.50 difference comes from?"

Notice what the message does not do: it does not accuse. It states two totals and asks for the breakdown. That is the right tone for a buyer you want to sell to next week as well.

Keep the price agreement in writing

The gap above arose because the agreement lived in a text message. Ask the assistant to turn each week's price messages into a one-line record for the sheet: date, buyer code, species, grade, price, and any conditions such as "if landed before noon". The record does not need to be elaborate. It needs to exist before the invoice arrives.

Price and availability messages to several buyers

Many boats phone round two or three buyers each evening to say what is coming in and ask what they will pay. That works until you are tired, or the weather has shifted the plan, or you want a record of who said what. A drafted message is quicker and leaves a trail.

Give the assistant a standing instruction once, then feed it the day's facts. For example:

Write a short text from a small fishing boat to a regular buyer.
Facts for today: expected landing tomorrow morning, approx 150 kg brown
crab (mostly large), 20 kg lobster, weather may move us back half a day.
Ask for their price per kg for each, by 6pm tonight.
Tone: friendly, brief, no emojis. Sign off with [your name].
Keep it under 60 words and do not promise quantities, say "approx".

An illustrative output: "Hi [first name], expecting around 150 kg brown crab (mostly large) and 20 kg lobster in tomorrow morning, weather permitting, so possibly half a day later. What prices per kg can you do? Could you let me know by 6pm tonight? Thanks, [your name]". You would send that as it is. The point is that you write the facts once and the assistant does the wording for each buyer, including a different tone for the buyer who likes a longer chat.

Two safeguards. First, never let the assistant state a price you did not give it, because it will happily offer a plausible one. Second, do not paste a buyer's private reply back into the chat with their name and number. Replace the name with a code. If you want a wider look at tidying customer messages like these, the approach for answering shop messages faster carries over almost line for line.

Comparing weekly prices across buyers

If you record each buyer's offered price in the sheet, a monthly comparison shows whether the buyer who always seems easy is also the one paying least. Ask the assistant to build a pivot or a formula for average price per kilogram by species and buyer code, then look at the result yourself. A difference of 20 cents a kilogram on 800 kilograms of crab a month is $160, which is worth a phone call.

Engine hours, gear and a maintenance calendar

A boat's maintenance is usually in someone's head and on a few receipts. An assistant can turn the manufacturer's service intervals into a calendar you will actually see. You need to supply the intervals, because the assistant will not know your engine's exact schedule and will invent one if you let it.

Take the engine service page of the manual, or a photo of it, and ask:

From this service schedule, build a table: task, interval in running
hours or months (whichever comes first), last done (blank), next due
(blank), parts needed. Use only the intervals printed in the document.
If a task has no interval, list it separately as "no interval given".

Now fill the "last done" column from receipts, and let a spreadsheet formula work out next due from the current engine hours. A filled-in line might read: oil and filter, every 250 hours, last done at 1,820 hours, next due at 2,070. If the engine has just been serviced and the boat averages 14 hours a week, the next one is about 18 weeks away (250 / 14), and a conditional format can turn the cell amber at 20 hours remaining.

The same approach works for gear: creels, nets, ropes and hauler parts. Create a list of items, the date each was bought or repaired, and a rough replacement rule you set, such as "replace head ropes every two seasons". The rule is yours. The assistant only formats it and nags you through the sheet.

Where it stops

Safety equipment checks, survey dates and anything a surveyor or insurer sets are dates you take from their documents. Put them in the calendar by hand and have the assistant draft the reminder wording, but do not ask it to tell you when they fall due. For deeper ideas on turning use data into service plans, see predictive maintenance for a small workshop, which covers what a single engine can and cannot support.

Drafting the paperwork you write from scratch

Every boat has a few documents that are written occasionally and always at a bad time: a letter to a buyer about a late payment, a note to an insurer after a gear loss, a covering email with a funding application, a job advert for a deckhand, a crew agreement outline. AI is a fast first drafter for all of them.

A late-payment reminder is a good model. Give it the invoice number, the amount, the original date and your terms, and ask for three versions: friendly, firm and final. Keep the friendly one for the first reminder. A drafted example: "Hi [first name], a quick reminder that invoice [number] for $2,008.90, dated [date], is now 14 days past our agreed terms. Could you let me know when we can expect payment? Thanks, [your name]". Again nothing in it was invented.

For anything that has legal effect, treat the draft as a starting point and have the right person look it over: a crew agreement, an insurance claim narrative or a contract with a buyer is worth a read by an adviser who knows your sector before you sign or send it.

What stays with the regulator, and why the assistant cannot help

Four areas should never be delegated, however confident the answer sounds.

• Licences and permits. Whether you hold the right authorisation for a species, a gear or an area is a fact about your account with the regulator, not something a chatbot can see.
• Quota and effort limits. The official record of what you have landed against your allocation lives in the regulator's system. Your own sheet is a working copy for planning. If the two differ, the regulator's wins.
• Logbooks and declarations. If your vessel must file electronically, the official entry is the one that counts. Use the assistant only for the notes you transfer from.
• Closures and size limits. These change. An assistant trained months ago cannot know this week's notice. Read the regulator's current page or call them.

A helpful habit is to write each rule-bound figure in your sheet with its source and date: "quota remaining 1,240 kg per regulator portal, checked Monday". If a number in the sheet has no source, it is an estimate and should look like one.

When the assistant gets a fishing detail wrong

Here is a plausible slip. A skipper asks the assistant to total the month's landings and it returns a figure 40 kilograms higher than the sheet. The cause: two rows for the same trip, one typed from the notebook and one pasted from a voice note, both read in. Nobody noticed because the total looked reasonable. The fix is a duplicate check: sort by date, species and grade, and ask the assistant to list any rows that share all three. Two minutes, run before each invoice check, removes the whole class of error.

Three other things go wrong often enough to list. Species names drift between "lobster", "lobsters" and a local nickname, which splits a pivot table into three rows, so agree your list once and make the assistant stick to it. Units slip when a weight is noted in pounds or stones for one trip. And an assistant asked "is this price fair?" will give a polite, vague answer based on general knowledge, not on this week's market. Prices come from your buyers, not the model.

Which assistant and what it costs

Any mainstream chat assistant that reads photos will do the catch-note and invoice work: ChatGPT, Claude or Gemini. A paid personal plan, such as ChatGPT Plus at $20 a month (the list price when checked in November 2026), is plenty for one boat. The free tiers are enough to test the method for a fortnight, though ChatGPT's free plan limits uploads, which is what the photo work uses. Spreadsheet work can sit in whichever one you already use, such as Excel or Google Sheets. Whichever plan you pick, find the training switch: OpenAI's pricing page says content on its Free, Go, Plus and Pro plans is used to train its models unless you opt out, while its business plans are not trained on by default. Opt out before you paste commercial prices: in ChatGPT the switch is "Improve the model for everyone" under Settings, then Data controls.

Your time is the larger cost. Setting up the sheet and your standing prompts takes a couple of hours once. After that, a trip's notes take about five minutes to photograph, convert and check, an invoice check takes about ten, and the weekly price messages take a couple of minutes each. If you currently spend four to six hours a week on all of this, one to two hours is a fair target to test against, not a promise; your own count after a fortnight is the figure to trust.

A first fortnight that will tell you if it works

1. Day 1: set up the sheet with fixed columns and a list of species and grades.
2. Days 2 to 6: photograph each trip's notes, convert them, and correct every [CHECK] cell against the page. Count the corrections.
3. Day 7: run the duplicate check and compare the sheet's weekly total with your own sum for the week.
4. Week 2: do one invoice check using the prompt above, and keep every buyer message in the sheet's notes column.
5. End of week 2: if corrections are under one in twenty cells and the invoice check found, or cleared, a difference you would have missed, keep going and add maintenance. If not, drop it and keep the notebook.

Boats differ a great deal, and if you run more than one vessel, sell to many buyers, or process the catch yourself, the sums become closer to those of a small food producer. For that, see costing a batch with yield loss and margin. If you would rather map this with someone against your own notebook and invoices, that is what my AI implementation consultation covers; you can reach it through the contact page.

Charter operators have a different set of jobs, mostly bookings and weather call-offs, covered in the tutorial on fishing charter bookings. And if you move your own catch to market, the guide to what a small fleet owner can use now covers the dispatch and invoice side of that run.

Questions skippers ask before trying this

Can AI tell me whether I am within my quota?

No. Quota and licence conditions are set by your regulator and held in their own system. AI can add up the weights you typed into your own sheet, but only the official record counts. Treat any total it gives as a prompt to check the real figure, never as the figure.

Is it safe to type catch weights and buyer names into a chatbot?

Weights and prices are low risk, but buyer names and bank details are not worth sharing. Switch off model training in your plan's settings (business plans are not trained on by default), replace buyer names with codes such as B1 and B2, and never paste licence numbers, vessel registration or bank details.

Do I need a special fishing app for any of this?

No. Everything here works with a notes app, a spreadsheet and a general chat assistant. A dedicated electronic logbook may be required by your regulator for some vessels, and if so it stays the official record. The spreadsheet is your own working copy for checking prices and totals.

Will it read my handwriting in the wet-weather notebook?

Often, but not reliably. Tidy block capitals on a photographed page usually come out well, while smudged pencil and shorthand do not. Always compare the typed result against the page for the weights, because a misread 8 as a 3 is the usual error and it changes the invoice.

Further reads

• AI in the Packaging Industry: Five Jobs a Converter Can Hand Over — Another small-producer list of handover jobs, with costs.
• How to Catch Made-Up Figures in AI-Drafted Proposals — How to catch invented figures in anything AI drafts.
• Can Farmers Use AI for Farm Records and Paperwork? — Record-keeping for another weather-driven primary producer.
• How to Catch Duplicate Invoices and Payment Fraud With AI — Spotting duplicate and wrong invoices more broadly.
• AI in Food Retail: Jobs a Deli or Farm Shop Can Hand Over First — If you sell part of the catch direct to the public.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI's ChatGPT pricing page, business data page and its help article on how your data is used to improve model performance (checked 5 Nov 2026); the reader's own regulator for licence, quota and logbook rules. All boat, catch and price figures are illustrative.