AI bookkeeping tools are genuinely useful, and the pitch is easy to like: upload your receipts and statements, let the tool sort and code them, save an evening a week. The catch is buried in the terms, and it is usually the same catch. Free tiers frequently reserve the right to train on what you upload. Paid tiers usually do not.

That matters because bookkeeping data is not yours alone. It contains your clients' names, their turnover, sometimes their bank details. Feeding that into a system that may use it to improve a shared model is, at minimum, a data protection question, and possibly a breach of the confidence your clients assume. The tool being clever does not change who the data belongs to.

The clause to find is usually near a heading like ‘how we use your data' or ‘model improvement'. Look for whether you can opt out, whether opting out costs money, and whether it applies retroactively to what you have already uploaded. If you cannot find a clear answer in five minutes, that is itself an answer about how much the vendor wants you to know.

None of this means avoid the tools. It means choose the tier that keeps your data out of training, do a quick data protection impact assessment where sensitive figures are involved, and keep a note of what you checked. If a client ever asks whether their numbers went into an AI, ‘yes, and here is the setting that stopped it being used for training' is a good answer to have ready.