Skip to content
Solo Runbooks
Menu
← All guides

How to Automate Invoice Follow-Ups with AI (Without Sounding Like a Robot)

17 November 2026

Chasing late invoices is one of the least pleasant parts of freelancing. It is awkward, it is easy to put off, and it directly affects your cash flow. It is also repetitive, which makes it a good candidate for AI help.

This guide shows how to use an AI assistant such as ChatGPT, Claude or Grok to handle the thinking and writing part of invoice follow-ups, while you keep control of the sending part. You do not need to write code or buy new software to start.

What “automating invoice follow-ups” should actually mean

Before the how, a note on the goal. There are two things people mean by automation here:

  1. Fully automatic reminders. Software sends emails on a schedule with no human involved.
  2. AI-assisted follow-ups. An AI reviews your invoices, decides who needs a nudge, and drafts the message. You review and send.

Most invoicing tools, such as Stripe Invoicing, QuickBooks, Xero, FreshBooks and Wave, already offer the first kind in some form. It is worth turning on. But scheduled reminders are blunt. They send the same message to your best long-term client and to the stranger who ignored three emails.

The second approach is what this article covers. It gives you tone that fits the situation and a human check before anything goes out. For a freelancer with a handful of clients, that is usually the better trade.

Step 1: Get your invoice data out of your tools

The AI can only work with what you give it. Start by exporting your invoices as a CSV or spreadsheet. Most invoicing tools have an export option in the invoices or reports section; the exact location changes between tools and versions, so look for “Export” or “Download”.

You need these columns:

  • Invoice number
  • Client name and the name of the person who pays
  • Issue date and due date
  • Amount and currency
  • Status (paid, unpaid, void)
  • Payment link, if you have one
  • Date of the last reminder you sent (add this column yourself if your tool does not have it)

Delete anything you do not need, such as bank account numbers or tax IDs. A good rule: if the AI does not need it to write a reminder, do not share it.

Step 2: Give the AI clear rules

The difference between a useful reminder assistant and a risky one is the instructions. Put these rules in a Project, a custom instruction, or the start of the chat:

  • Draft only. The assistant never sends anything.
  • Facts only from your data. It must not invent amounts, dates or payment details.
  • Skip closed invoices. Paid, void and cancelled invoices are ignored.
  • Calculate days overdue from today’s date, which you state in every run. Do not assume the AI knows the date.
  • Escalate gradually. Friendly nudge first, firmer later.
  • Hand over serious cases. Anything very overdue goes on a “needs your decision” list rather than getting a draft.

Here is a compact version you can start from:

You are my accounts-receivable assistant. You only DRAFT reminder emails. You never send anything.
Use only facts from the invoice data I provide. If something is missing, write "MISSING: <field>".
Ignore invoices with status paid, void or cancelled.
Days overdue = today's date minus due date. Today's date: [YYYY-MM-DD].
Stages: 1-7 days overdue = friendly nudge; 8-21 = clear reminder; 22-45 = firm notice;
over 45 = do not draft, list under "Needs your decision".
Tone: polite, direct, human. No threats, no legal language, no late fees unless I ask.
Combine several overdue invoices for one client into one email.
Each draft: subject, invoice number(s), amount and currency, original due date, payment link,
one clear request. Under 120 words.
Finish with a summary table and total overdue per currency (never add different currencies together).

The full version of this prompt, with more rules and a testing checklist, is in the Solo Operator Agent Kit.

Step 3: Run it weekly and read the output

Pick one day a week. Paste or upload the export, state today’s date, and ask the assistant to follow its instructions. A good response contains four things:

  1. A summary table of overdue invoices.
  2. Totals per currency.
  3. A draft for each client.
  4. A list of anything it could not handle, such as missing data.

Now read each draft as if you were the client receiving it. Does it sound like you? Is the amount right? Is the payment link right? Is it the right person? Then copy it into your email and send it yourself.

What a good reminder looks like

Whatever tool writes it, a reminder works best when it is short, specific and easy to act on. Here is an example of a stage 2 draft (invented details):

Subject: Invoice 1042: payment reminder

Hi Jordan,

A quick reminder that invoice 1042 for EUR 1,800, issued on 3 October, was due on 17 October and I do not see it as paid yet. You can pay by card using this link: [link].

If it has already been paid, please ignore this and send me the date so I can match it on my side. If there is a problem with the invoice, let me know and I will sort it out.

Thanks, Sam

Notice what it does: it states the facts, makes paying easy, offers a graceful way out for an honest mistake, and does not accuse anyone of anything. Most late payments are oversights, not bad faith, so start from that assumption.

Where the AI helps, and where it should not

Good uses:

  • Spotting which invoices are overdue and by how many days.
  • Grouping several invoices into one email.
  • Picking a suitable tone for each stage.
  • Rewriting a draft to be warmer or firmer.
  • Drafting a reply when a client proposes a payment plan.

Keep these human:

  • Sending. You should see every message that goes out under your name.
  • Final notices, late fees and legal steps. Rules about late-payment interest and fees differ by country and by your contract. Ask a professional, not a chatbot.
  • Relationships. If a long-term client is having a hard time, a phone call may be worth more than any email.
  • Checking payments. Your export may be out of date. Check your bank or invoicing tool for payments received before you send anything.

Testing before you trust it

Do not point a new setup at your real clients on day one. Make a fake export of five or six invoices and check that the assistant:

  • Ignores a paid invoice.
  • Does not treat an invoice due today as overdue.
  • Chooses different stages for invoices 3, 15 and 30 days overdue.
  • Refuses to draft for one that is 60 days overdue and flags it for you.
  • Combines two invoices for one client into one email.
  • Keeps currencies separate.
  • Never says it sent anything.

If any of those fail, tighten the instructions and test again. This takes about twenty minutes and saves you from sending a wrong amount to a real client.

Common mistakes

Not stating the date. AI assistants do not reliably know today’s date. Without it, days-overdue calculations can be wrong. Always include it.

Sharing too much data. Remove bank details and anything unrelated to the reminder. Also check your AI tool’s privacy and data settings for your plan.

Skipping the review. Language models can write fluent text that contains a wrong number. A thirty-second read is your safety net.

Sending the same tone to everyone. The point of AI assistance is fit. Tell it which clients are long-term or sensitive.

Letting reminders pile up. The fix for late payment is a steady rhythm. A weekly slot beats a monthly panic.

Prevent the problem upstream

Follow-ups work better when late payment is less likely in the first place. A few habits that cost nothing:

  • Put payment terms in your proposal and on every invoice.
  • Send invoices promptly, and to the person who actually approves payments.
  • Make paying easy with a payment link.
  • Ask for a deposit on larger projects.
  • Send a friendly heads-up a few days before the due date for clients who often pay late.

You can ask your AI assistant to draft that pre-due-date message too.

A realistic setup

Here is what a sensible, low-risk weekly routine looks like:

  1. Monday, 10 minutes: export invoices, paste them into your assistant, state the date.
  2. Review the summary and the drafts.
  3. Send the ones you approve from your own email.
  4. Note the date you sent them in your sheet.
  5. Look at the “needs your decision” list and decide what to do with each one.

Later, if you want to go further, you can connect your tools so the assistant reads invoices directly and saves drafts to your mailbox, or schedule the weekly run. Connector and scheduling features vary by AI product and plan, and they change often, so check what your tool currently offers. Whatever you automate, keep the sending step in your hands until you have watched the setup behave correctly for a good while.

Get the ready-made version

If you would rather not write and test all of this yourself, the Solo Operator Agent Kit includes an Invoice Follow-Ups runbook with the complete master instructions, weekly prompt, follow-up prompts, a testing checklist and the pitfalls to avoid. It is one of ten runbooks for the admin work freelancers dislike most, including inbox triage, bookkeeping prep and subscription review.

See the Solo Operator Agent Kit

Not ready to buy? Try a free runbook first: get the free New Client Onboarding runbook.