It depends which message. A pre-due confirmation, a first polite nudge and a statement of what's outstanding are routine enough for software to draft and send from your domain. Where you're replying to a dispute, talking to a client in trouble or sending a formal notice, what's needed is an opinion about a person more than a well-phrased sentence, so write those yourself.
The short version
- No UK rule requires you to say a message was AI-drafted, and the EU AI Act's disclosure duty binds providers of AI systems in the EU, so it doesn't reach a UK business chasing UK customers.
- The better standard than the legal minimum comes from your professional bodies: treat AI output as if a less experienced junior colleague had drafted it.
- Independent research puts the value of the reminder itself at roughly 25 percent against sending nothing, so its existence matters more than its author.
- Nobody has published research on how customers react to AI-written reminders, so anyone telling you what your clients prefer is guessing.
- What has to be human is reachability: a reply that lands with a person, and a name that can answer it.
Which messages, specifically
Sorting by message gets you to a usable answer faster than arguing the principle.
| Message | Who it should come from |
|---|---|
| Confirmation before the due date | Software, from your domain, on a rule |
| First reminder after the due date | Software, drafted and sent, reviewed if you prefer |
| Second and third reminders | Software, with a person watching the account |
| Statement of everything outstanding | Software, on a trigger you set |
| Reply to a query or a dispute | You, every time |
| Instalment or hardship conversation | You, and preferably by phone |
| Formal notice or letter before action | You, with the figures checked |
The line falls where a message stops being about the invoice and starts being about the relationship. Everything above the dispute row states what's owed and how to pay it, and your client gains nothing from you typing it personally. Below that row sits a judgement about somebody's circumstances, or a legal consequence you're prepared to stand behind.
What the law asks of you here
Very little, on anything we could find. No UK rule requires you to disclose that a business email was drafted by AI, and the ICO's transparency guidance for AI is about how you process personal data, not about authorship.
The EU AI Act is the source of the disclosure duty people have heard about. Article 50(1) obliges providers to design systems that interact directly with people so those people know "that they are interacting with an AI system", which puts the obligation on whoever builds the system, in the EU. A UK business chasing UK customers is outside its scope. Is it safe to let AI email your clients? covers the data protection side properly, including the UK rules that did change in February 2026.
The standard worth using instead
Absence of a rule leaves you deciding for yourself, and your professional bodies have written down something more useful than the legal minimum. The PCRT bodies' guidance on the ethical use of AI tools of 19 January 2026 says at 3.4 that output from an AI tool "should also be regarded as if it were prepared by a less experienced junior colleague and reviewed with appropriate scepticism", and at 3.5 that members "remain ultimately accountable for any work produced, regardless of whether AI has been involved".
Applied to chasing, that framing answers the question almost by itself. You'd let a junior send the standard reminder unread, you'd check their draft of anything mentioning a fee or a legal step, and you'd handle the awkward call yourself. Nobody would suggest telling the client which colleague typed which message, or that the partner stops being responsible for it.
On disclosure the same guidance stays practical. At 1.3 it suggests a statement in the engagement letter about the potential use of AI tools, and that consideration be given to telling a client about actual use when deliverables are provided. So if you're a practice, this belongs in your engagement letter and not at the bottom of a reminder.
The reminder matters more than its author
The strongest independent evidence in this area isn't about authorship at all. Gillitzer and Sinning ran a field experiment on Australian business tax debts and found a reminder letter raised the probability of payment by roughly 25 percent compared with sending nothing. Varying its timing inside a three-week window changed how quickly people paid without changing whether they paid. Tax letters from a revenue authority aren't invoice emails from a supplier, so read the direction and not the number.
That reframes your question. The measurable difference is between chasing and not chasing, and no published study compares AI-written reminders with human-written ones in any jurisdiction, so nobody can tell you which performs better. What's evidenced is that the message going out at all matters a great deal, and the weeks when nobody gets round to it are the expensive ones.
The human element that isn't the typing
Our view, and it's judgement rather than research: what your customer wants from a person isn't authorship, it's access. A reminder signed by a named colleague who didn't write it and can't answer a reply is the version most likely to cost you something, and it costs nothing to avoid. Make sure a reply reaches somebody, name a real contact, and let the routine sequence carry your firm's name instead of borrowing an individual's.
That design also lines up with what Parliament thought made an automated decision fair. Article 22C(2) of the UK GDPR requires, where it applies, that a person can obtain human intervention and contest a decision, and a reply-to address reaching you is the cheapest version of that. We'd build it in whether or not the article touches your reminders.
Nobody has published research on how customers react to AI-written chasing, so treat confident claims about what clients prefer, including ours, as opinion.
Where the routine messages can go
The case for handing over the routine part is about capacity and not writing quality. You know which client answers a text. You know the one who needs the office manager copied in, and the one who pays the moment a statement arrives. Holding that for a dozen accounts is easy and for four hundred it isn't, and businesses affected by late payment already spend an average of 86 hours a year chasing, on Small Business Commissioner research.
Adfin's Customer Agents work from your rules of engagement: which channels are allowed, how persistent they may be and what goes out unreviewed, with adaptive templates sent from your own email domain. Whether that collects more than your current sequence is not something anybody can show you yet, and our argument is that it keeps routine messages moving in the weeks you're busy. Can AI chase invoices for you? sets out what the software does and doesn't do.
Common questions
Should invoice reminders come from a person or an AI? It depends on the message. Confirmations, first reminders and statements are routine enough for software to draft and send. Anything answering a dispute, discussing hardship or asserting a legal consequence is better from you, since each needs a judgement.
Do you have to tell customers a reminder was written by AI? No UK rule requires it. The EU AI Act's disclosure duty binds providers of AI systems in the EU, so it doesn't reach a UK business chasing UK customers. For a practice, the PCRT guidance points at a statement in your engagement letter about potential AI use.
Does it matter whose name is on the reminder? Our view is that what matters is whether a reply reaches a person. A message signed by a colleague who can't answer it is the version most likely to annoy somebody, so use your firm's name and a reply-to address that lands with you.
Which messages should you always write yourself? Anything to a client who has raised a dispute, anything discussing an instalment arrangement or hardship, anything quoting a specific interest or fee figure, and any formal notice. Those turn on judgement or on numbers you'd need to defend.
Do AI-written reminders work as well as ones you write? No published study compares them, in any jurisdiction, so nobody can answer that with evidence. What independent research does show is that sending a reminder matters a great deal against sending nothing, which puts the emphasis on consistency.
Should your engagement letter mention AI? The PCRT bodies' January 2026 guidance suggests including an appropriate statement about the potential use of AI tools, and considering disclosure of actual use when deliverables are provided. That's a professional recommendation and not a legal requirement.
Sources
- Small Business Commissioner — late payments research (accurate as of August 2026)
- ATT — PCRT bodies' guidance on the ethical use of artificial intelligence tools, January 2026 (accurate as of August 2026)
- Tax and Transfer Policy Institute, ANU — nudging businesses to pay their taxes, does timing matter (accurate as of August 2026)
Reviewed by the Adfin team. This summarises the UK legal position and professional guidance on AI-drafted client communications as at August 2026 for information, and it isn't legal advice. UK data protection law on automated decisions changed in February 2026 and the regulator's updated guidance was still in draft when this was written. If you belong to a professional body, its guidance on AI tools covers your duties of competence and confidentiality, and no software setting moves those off you.
