Making Tax Digital for Income Tax stopped being a future problem this April. The first wave of UK sole traders and landlords is now keeping digital records and facing quarterly deadlines, and two more waves follow in 2027 and 2028. At the same time, AI assistants have become good enough at reading bank statements and categorising expenses that plenty of business owners are asking the obvious question: can AI do the boring part? The honest answer is yes for preparation and no for submission, and the difference matters. This guide walks through what HMRC actually requires, where AI genuinely saves hours, and where trusting it blindly would cost you money.
What is Making Tax Digital for Income Tax and who does it affect?
Making Tax Digital for Income Tax is HMRC's requirement for sole traders and landlords to keep digital records and send quarterly summaries of income and expenses through compatible software, replacing the single annual Self Assessment scramble. According to HMRC's eligibility guidance, it applies from 6 April 2026 for those with qualifying income over £50,000 in the 2024 to 2025 tax year, from 6 April 2027 for income over £30,000 in 2025 to 2026, and from 6 April 2028 for income over £20,000 in 2026 to 2027.
The phasing catches people out because the qualifying year sits two years behind the start date. HMRC looks at what you earned in 2024 to 2025 to decide whether you were in scope this April. If your income has grown since then, you may join a later wave, but you cannot rely on a quiet year to keep you out once you have crossed a threshold in the relevant qualifying year.
Note also that qualifying income combines self employment and property income. A tradesperson with a single buy to let, or a landlord with a side business, gets assessed on the total, which is why so many people who think of themselves as small are in scope earlier than they expected.
When do the quarterly deadlines actually land?
For the 2026 to 2027 tax year, HMRC's guide to using the service sets the quarterly update deadlines at 7 August 2026, 7 November 2026, 7 February 2027 and 7 May 2027, with the tax return itself still due by 31 January 2028. Each update is a cumulative summary of income and expenses submitted through compatible software.
Four deadlines a year instead of one changes the rhythm of record keeping. The old habit of reconstructing twelve months of transactions each January simply does not survive contact with a 7 August deadline. The businesses coping best are the ones treating bookkeeping as a small weekly or monthly task rather than an annual archaeology project, and that shift in cadence is precisely where automation and AI assistance earn their place.
What records does HMRC expect you to keep digitally?
For every self employment and property transaction, HMRC requires three things to be recorded digitally: the amount, the date the transaction took place, and the income or expense category, with categories matching those used in Self Assessment. That is the core of the digital record requirement set out in HMRC's guidance, and it is deliberately narrower than many people fear.
You are not required to photograph every receipt or abandon spreadsheets. HMRC's guide explicitly allows an all in one software product, a combination of products, or spreadsheets paired with bridging software that handles the submission. What you cannot do is keep records on paper and type totals into a portal once a year.
The hard part is not the format, it is the discipline. Amount and date come straight from your bank feed. Category is the field where errors creep in, because it requires judgement, and it is the field that determines your taxable profit. Get categorisation right consistently and the rest of MTD is largely plumbing.
Can AI really get your records ready for Making Tax Digital?
Yes, for the preparation stage. A general purpose AI assistant can take an exported bank statement and sort hundreds of transactions into Self Assessment categories in minutes, flag duplicates and anomalies, identify transactions it is unsure about, and produce a clean structured spreadsheet ready for MTD software or bridging tools. What it cannot do is submit anything to HMRC, because quarterly updates must travel through compatible software connected to HMRC's systems.
The realistic workflow looks like this. Export a CSV from your business bank account. Give the AI your fixed category list and a handful of rules, for example that a named supplier is always materials and that transfers to your personal account are drawings, not expenses. Ask it to categorise every line, mark anything below a stated confidence level for review, and return the result as a table. Review the flagged lines yourself, then import the finished sheet into your MTD software.
Done monthly, that turns a multi hour chore into perhaps twenty minutes of review. There is a fuller walkthrough of this approach in our guide to AI bookkeeping for UK small businesses, and a worked comparison of the economics in Claude AI versus a bookkeeper on cost.
Which jobs should you hand to AI first?
Start with the high volume, low judgement work: transaction categorisation from bank exports, duplicate detection, matching invoices to payments, and producing quarterly summaries by category. These tasks are repetitive, rule based and easy to verify, which makes them ideal for AI assistance and keeps the risky judgement calls in human hands.
A sensible priority order for a sole trader or landlord:
- Categorising bank transactions against your Self Assessment category list.
- Flagging gaps, such as months with income but no recorded expenses, or invoices with no matching payment.
- Drafting a quarterly summary you can sanity check against your software's figures before the deadline.
- Writing plain English questions for your accountant about anything ambiguous, so professional time is spent on judgement rather than data entry.
There are worked examples of these prompts in our piece on running small business accounts with Claude, and the broader Claude for business hub collects the rest of the series.
Where does AI fall short on tax records?
AI falls short wherever tax judgement replaces pattern matching. Repair versus capital improvement on a rental property, mixed personal and business use, VAT edge cases and the treatment of mortgage interest are all places where a confident sounding but wrong answer carries a real cost. AI also cannot take legal responsibility, and it cannot file, so a human review step and recognised software remain non negotiable.
Two failure modes deserve particular respect. The first is silent misclassification: an AI that categorises 500 transactions will get some wrong, and unlike a human it will not feel uneasy about the hard ones unless you explicitly ask it to flag low confidence lines. Always ask for an uncertainty flag. The second is invented detail. If a transaction description is vague, a model may guess a plausible category rather than admit it does not know. Your prompt should make "unknown, needs review" an acceptable answer.
What happens if you miss a quarterly update?
Missed submissions are handled through a points system rather than immediate fines. According to HMRC's penalties guidance, each missed deadline earns a penalty point, and at the threshold of 4 points you receive a £200 penalty, then a further £200 for each subsequent missed submission. Late payment of the tax itself is charged separately, starting at 3% of the amount outstanding at day 15 for the 2026 to 2027 tax year.
The late payment schedule stiffens quickly. For 2026 to 2027 it is 3% of the tax owed at day 15, another 3% of what is still owed at day 30, and then an annual rate of 10% charged daily from day 31. HMRC's penalties guidance also sets out the rates that apply in later tax years. None of these numbers is ruinous on its own, but they compound with the stress of falling behind, and the cheapest fix is a record keeping routine that makes each quarterly update a formality.
How do you set up an AI-assisted MTD workflow this quarter?
Pick one bank account for business transactions, choose HMRC compatible software or a spreadsheet plus bridging tool, write a one page category guide, then run a monthly loop: export transactions, have AI categorise them against your guide with uncertain lines flagged, review the flags, and import the result. Do a dry run well before your next quarterly deadline rather than discovering problems on the day.
The one page category guide is the piece most people skip and most regret skipping. It is simply a list of your Self Assessment categories with two or three examples of your real suppliers under each, plus your standing rules. Every AI session starts by pasting it in, which is what keeps quarter three's categorisation consistent with quarter one's. When your accountant reviews the year, consistency is worth more than any individual clever judgement.
If you are still deciding between full accounting software and the spreadsheet route, start from how many transactions you handle a month. Under a hundred or so, a disciplined spreadsheet feeding bridging software is entirely workable and HMRC explicitly permits it. Beyond that, a bank feed into proper software with AI handling the review layer usually wins on time.
Is a spreadsheet plus AI enough, or do you need full software?
A spreadsheet is a legitimate digital record under Making Tax Digital provided it connects to bridging software for submission, and HMRC's guide states you can choose one all in one product or combine several. AI makes the spreadsheet route far more viable than it used to be, because the tedious parts, categorisation and consistency checking, are exactly what AI automates well. The deciding factors are transaction volume and how much you value a live bank feed.
The trade off is honest on both sides. Full software gives you automatic bank feeds, built in submission and an audit trail, at a monthly cost and with some lock in. The spreadsheet and bridging route keeps you in control and cheap, but every safeguard is one you built yourself. AI narrows the gap by doing the machine like work in the manual route, but it does not remove the need for your review. Whichever route you choose, the quarterly deadlines are the same, and the habit of little and often is what actually keeps you compliant.
The short version: Making Tax Digital is a cadence change more than a technology change. AI cannot file for you and should not make your judgement calls, but it can absorb most of the repetitive work between you and four clean quarterly updates a year. Start the routine now, while the first year grace periods are still on your side.
Making Tax Digital — FAQ
When does Making Tax Digital for Income Tax actually start?
It has already started for the first group. HMRC's eligibility guidance sets out three phases. If your qualifying income from self employment and property was over £50,000 in the 2024 to 2025 tax year, you needed to use Making Tax Digital for Income Tax from 6 April 2026. If it was over £30,000 in 2025 to 2026, you join from 6 April 2027. If it was over £20,000 in 2026 to 2027, you join from 6 April 2028. So even if you are under the first threshold, the sensible move is to check which phase you fall into now and get your record keeping into shape a full tax year before your start date, because the qualifying year is always the one before HMRC looks at you.
Can I use AI on its own instead of MTD software?
No. AI tools like Claude or ChatGPT cannot submit quarterly updates to HMRC, because submissions must go through compatible software that talks to HMRC's systems. HMRC's own guide says you can use an all in one software product, several products together, or spreadsheets combined with bridging software. Where AI earns its keep is everything before the submission step: sorting a year of bank transactions into Self Assessment categories, spotting duplicates and missing receipts, drafting queries for your accountant, and turning a shoebox of records into a clean spreadsheet that your MTD software or bridging tool can ingest. Think of AI as the preparation layer and recognised software as the filing layer, and keep both in the workflow.
What counts as a digital record under Making Tax Digital?
HMRC's guidance is more modest than most people expect. For each self employment and property transaction you need to record three things digitally: the amount, the date the transaction took place, and the income or expense category, with categories matching the ones used in Self Assessment. You do not need to scan every paper receipt into the system, although keeping copies is still wise for evidence. A well structured spreadsheet can satisfy the digital record requirement provided it links to bridging software for submission. The practical challenge is consistency across four quarterly updates a year, which is exactly the kind of repetitive categorisation work that AI handles well when you give it a fixed category list and clear rules.
What are the penalties if I miss MTD deadlines?
HMRC runs a points based system for late submissions. Each missed deadline earns a penalty point, and once you reach the threshold of 4 points you receive a £200 penalty, then a further £200 for every missed submission after that. HMRC's penalties guidance also notes that points for missed quarterly updates apply for tax years after 2026 to 2027, so the first year is softer on quarterly slips, though the tax return deadline still counts. Late payment is separate: for the 2026 to 2027 tax year HMRC charges 3% of the tax owed at day 15, another 3% at day 30, and then an annual rate of 10% charged daily from day 31. HMRC's guidance also sets out the rates that apply in later tax years.
Do landlords need to follow Making Tax Digital too?
Yes, if their qualifying income crosses the thresholds. Making Tax Digital for Income Tax applies to sole traders and landlords alike, and qualifying income combines both sources. A part time consultant with modest rental income can be pulled in by the combined total even though neither activity alone would cross the line. For landlords the record keeping burden sits in slightly different places: rent received, letting agent fees, repairs versus improvements, mortgage interest treatment, and apportioning costs across properties. AI assistants are genuinely useful for a first pass at those distinctions and for keeping category choices consistent quarter to quarter, but classification questions with real tax consequences, such as repair versus capital improvement, deserve a human professional's sign off.
