Guide

Where to start with AI in a UK property business

Guillaume Black Chief Executive, Property Filter 29 July 2026

The most common thing we hear from investors about AI is not scepticism, it is paralysis. They know it matters, they have read that it changes everything, and they have no idea what to actually do on a Tuesday morning. So this guide is deliberately narrow: seven days, five tasks, three habits, and nothing to buy.

Day one: pick the task you hate most

Not the most valuable task. The most avoided one. In the first week motivation beats strategy, and a small unpleasant job you have been putting off for three weeks gives you a much sharper read on whether this is useful than an ambitious project you abandon on Thursday.

For most people it is one of: the legal pack nobody has read, the reply to the difficult tenant or agent, the receipts that need sorting before the accountant asks, or the deal that has been sitting in a tab unanalysed.

Day one, second half: write your standing brief

This is the highest-leverage twenty minutes in the whole week, and almost nobody does it.

Write one page about your business. Your strategy and why. The areas you actually buy in. Your finance terms: rate, fees, loan to value, whether you buy personally or through a company. Your portfolio as it stands. Your refurbishment basis. Your target returns and your walk-away rules. And, importantly, what you have tried that did not work.

Then paste it at the top of every serious conversation, before the actual question.

The reason this matters so much is worth being blunt about. When people say the output was generic, what has happened is that they asked a stranger a question about a business the stranger knows nothing about, and got an answer built from the average of everyone who has ever written about property. That is not a limitation of the technology, it is a consequence of starting from zero. The standing brief is the cheap fix. It is tedious, it works, and it will tell you within a week whether your real bottleneck is capability or memory.

Days two and three: one document job

Give it something real and long. A lease, a survey, a legal pack, twelve months of statements. Then ask for output you can act on rather than a summary:

• What would a buyer worry about here, in priority order.
• What is missing that you would expect to see in a pack like this.
• Everything that implies a cost, grouped by urgency, with the document's own wording quoted next to each item.
• The five questions I should put to my solicitor about this.

That last one is the sleeper. Arriving at a professional's call with a list rather than a vague worry halves the call and doubles what you get out of it, and you stop paying professional rates for the discovery phase.

Insist on the quoting. It keeps you honest about what the surveyor actually wrote rather than what you would like them to have written, and it is the habit that later protects you from accepting invented detail.

Day four: one drafting job

The letter, the reply, the contractor brief. Give it your best previous example as the model, then the specifics.

The value here is not literary. It is that most difficult correspondence sits for days because starting is unpleasant, and a first draft in ten seconds turns writing into editing. Different task, different level of resistance.

Days five and six: one real deal

Now something that matters. Take a deal you are actually looking at, or better, one you already completed or walked away from so you know the answer.

Give it the listing in full including floor area and EPC, your real finance terms, your refurbishment basis, whatever comparable evidence you have with addresses and dates, and the exit you intend. Then ask for the test that matches the strategy rather than a generic yield.

Finish with the instruction that is worth more than any prompt template: list everything that has to be true for this to work, then tell me which of those is least likely. Doing this on a deal you already know the answer to is the fastest way to calibrate how much to trust it. The full method is in the deal analysis guide.

The three habits that decide whether it sticks

These are not optional extras. They are the difference between someone who gets years of value out of this and someone who tells you at a networking event that they tried AI and it made things up.

Supply ground truth rather than trusting recall

Ask it to reason over material you give it, not to remember a rule. The current HMRC page, the actual lease, the council's own licensing register. This single distinction accounts for most of the difference between useful and dangerous.

Ask for the source of every number

You want one of two answers: something you can open and check, or a plain admission that it does not know. A confident figure with no provenance is the one that eventually ends up inside an offer.

Never let it be the last word on anything with a penalty attached

Possession, licensing, deposit handling, tax filing. UK property tribunals spent 2025 and 2026 dealing with parties who submitted AI-generated case law that did not exist, and the sanctions escalated to costs awards and a debarment. That story is worth reading before you rely on anything.

Day seven: decide what to keep

Be strict. Keep only what saved you time twice. Anything that impressed you once and has not been used since was entertainment.

Then run the one honest test: name a task that used to take an hour and now takes fifteen minutes, and that you have done twice. If you cannot name one, nothing has changed, however good the conversations felt. If you can name three, you have a different problem now, and it is worth naming precisely.

That problem is repetition. You will be spending real time every week re-pasting your standing brief, re-explaining a deal, re-uploading the same documents, re-establishing what you told it last Tuesday. That is the point at which the bottleneck stops being what AI can do and becomes what it remembers, and no amount of better prompting fixes it. It is also, not coincidentally, the point at which a workspace that holds your deals and your documents starts to be worth paying for. The full sequence, including that stage, is in how to invest in UK property with AI.

What to ignore for now

• Prompt packs. A hundred prompts written for someone else's business, mostly American. Your standing brief is worth more than all of them.
• Choosing a tool in week one. You cannot judge a tool until you can name the job you are hiring it for, and a week of real tasks is how you find that out. Then read the guide on judging tools.
• Automating anything. Automation multiplies a process. Get the process right by hand first, or you will scale a mistake.
• Which model is best. The gap between models is far smaller than the gap between an assistant that knows your business and one that does not.

Frequently asked questions