Home · Business · Finance & Accounting · Invoicing & credit control

YES

As of 13 August 2026, AI can categorise your purchase invoices.

This still needs a person who signs their name to it.

Can you do it?

5 minutesto a draft.

30 minutesto something you’d act on.

Cost, all in£0

Skill neededchat-fluent

Who has to check ityou

What the alternative costsNanonets is a purpose-built AI document-processing alternative for invoices, receipts and forms.

If this goes wrong, purchases can land in the wrong accounts or receive the wrong VAT treatment and require correction before your records or returns are finalised.

What to actually do

  1. Hand it to a person

    The route this page recommends

    A person who owns the outcome does this end to end, worth it when the failure is dear.

  2. Use a tool built for this

    Second choice
  3. Do it yourself

    The distant third

    A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

    How to actually do it

    1. Open your current chart of accounts and bookkeeping or VAT policy, then copy the category definitions and relevant rules into a working document.
    2. Gather the purchase invoices for the period as readable PDFs or images, keeping one clear filename per invoice and removing unrelated personal documents.
    3. Record the business name, business type, VAT registration status and accounting basis in the prompt, then paste the chart of accounts and business-specific rules into the marked sections.
    4. Upload the invoices to an AI tool that can read documents, paste the prompt, and ask it to produce the requested table with separate approval and human-decision lists.
    5. Compare each proposed category, VAT treatment, amount and supplier against the original invoice and your bookkeeping rules, and resolve every flagged or low-confidence entry.
    6. Send ambiguous transactions and any entries that could materially affect VAT or the accounts to your bookkeeper or accountant before approving or importing the results.

    Prompt

    Categorise the purchase invoices I provide for a UK business. Use only information visible in the invoices and the bookkeeping rules below. Do not invent missing details, suppliers, amounts, VAT rates or business purposes.
    
    Business name: [BUSINESS NAME]
    Business type: [SOLE TRADER, LIMITED COMPANY OR OTHER]
    VAT-registered: [YES OR NO]
    Accounting basis: [CASH, ACCRUALS OR UNKNOWN]
    Chart of accounts and category definitions:
    [PASTE CHART OF ACCOUNTS]
    Business-specific rules:
    [PASTE RULES]
    
    For each invoice, return a table with: invoice filename, supplier, invoice date, invoice number, net amount, VAT amount, gross amount, currency, proposed account category, proposed VAT treatment, confidence level, and a short reason. Flag any missing, ambiguous or unusual information instead of guessing. Separate invoices that are ready for approval from invoices that need a human decision. Check that net amount plus VAT equals gross amount where the figures are present, and identify duplicates or possible credit notes. This is a bookkeeping draft, not professional advice. Do not submit anything to HMRC or post entries to accounting software.

    Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.

What it gets wrong

  • AI cannot know the business purpose of an invoice that is not stated in the document.
  • AI cannot reliably decide ambiguous categories where your chart of accounts or accounting policy gives no clear rule.
  • AI cannot replace your responsibility for selecting and approving the VAT treatment recorded in your books.
  • AI cannot spot every connected-party transaction, private-use element or unusual commercial arrangement without relevant context.
  • AI cannot sign off your accounts or submit a VAT return to HMRC on your behalf.

Even on a YES, the friction has a name: verification cost, judgement under ambiguity and stakes of error.

How we scored this

Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.

AxisScore (0–2)
Output2
Inputs2
Verification1
Liability1
Effort delta2
Total8 / 10

FAQ

Can ChatGPT categorise purchase invoices?
Yes, it can extract invoice details and suggest categories if you provide your chart of accounts and business rules. Check every category and VAT treatment before approving the entries, because ChatGPT does not know unstated business context.
Can AI categorise invoices for VAT?
AI can suggest VAT treatment from the invoice and the rules you provide, and can flag missing or inconsistent VAT details. You remain responsible for the records used in a VAT return, so send uncertain or unusual invoices to your accountant or bookkeeper.
Is AI invoice categorisation accurate enough for bookkeeping?
It can handle clear, repetitive invoices well, but accuracy falls when an invoice has mixed items, private use, unusual VAT treatment or an unclear business purpose. Treat the output as a draft and approve it against the invoice and your bookkeeping policy.
What information does AI need to categorise invoices?
Give it readable invoices, your chart of accounts, VAT registration status, accounting basis and any rules for common suppliers or expenses. Include instructions to flag missing information and never guess amounts, categories or VAT treatment.

Nearby answers

Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.

The newsletter

AI news, new answers and product picks, straight to your inbox.