How to use AI to find property deals, and why it won't do the deal for you
There are plenty of AI tools that find property. None of them view, offer or negotiate. How I use AI to save time on sold prices, rents and EPC data, where it goes wrong, and the trap of analysing instead of doing.

If you want to know how to use AI to find property deals, there are plenty of tools that will do it. Portals have AI search now, there are apps that rank listings by return, and you can paste any listing into ChatGPT and ask if it's worth chasing. I've used them and I still do, but not a lot.
Here's why. None of them do the work. A house that's for sale on Rightmove today is still sitting there for everyone else to see, whatever an AI tool says about it. You still have to view it. You still have to negotiate, work out whether it's a straight purchase or a creative deal, make the numbers work, get it agreed and put it through the legals. AI does none of that for you.
Where AI does earn its place is the slog in between: sold prices, rents, EPC data and the numbers. It's a big time saver there. Used right, it gives you more time for the jobs that make you money. Used wrong, it becomes a very convincing way of feeling busy while nothing happens.
Below I'll go through what AI can and can't find, what it won't do for you, the trap I see people fall into, how I use it to save time, the prompt I use for sold prices, where it gets things wrong, the free places to get the data yourself, and the tool I pay for to shortcut it.
What I say about lending rules, data protection and the tools themselves is as it stood when I wrote this, in November 2026. The tools change every few months. It's education, not advice, and the small print is at the end.
Can AI find property deals?
It can find properties. It can't find a deal on its own, because a deal is a house plus a vendor with a reason to sell at a price that works for you. The house is on the internet. The reason usually isn't.
People use AI to find property deals in 4 ways:
- The portals' own AI search. Rightmove AI search, called Ask Rightmove, launched on its homepage in February 2026 and lets you describe what you want in plain English instead of ticking filters. It has since been added to listings and search results, where it compares properties side by side, and Rightmove has an app inside ChatGPT. Rightmove's help pages say these are experimental and only shown to some users.
- An AI property deal finder. These are paid apps and sites that pull listings, run yields and returns on them and rank them. Most of them sell this as AI property sourcing.
- ChatGPT, Claude or Gemini with a listing pasted in. You paste the details and ask whether it's worth chasing.
- An automation. A property alert lands in your inbox, and a workflow reads it, pulls the numbers and drops it into a spreadsheet. I built one of these myself.
All 4 work from the same thing: houses that are already listed, and numbers someone has estimated. Every other investor using the same app or the same search sees the same houses. Price drops, time on the market and words like "chain free" or "probate" are clues to how motivated a vendor might be, and AI can spot those for you. It can't tell you why the vendor is selling or what they'd take. You only find that out by talking to them.
What AI won't do for you
This is the part the tools don't put in their adverts. Finding a house that might work is the first of 7 steps. AI helps with the first 2 and can draft the options for one more. The rest is you:
| Step | Can AI do it? |
|---|---|
| Find a house that might work | Yes, it can search and sort listings |
| Check the numbers | It does the slog fast, but you have to check what it gives you |
| View it | No. You have to stand in it, look at the condition and meet the vendor or the agent |
| Find out what the vendor needs | No. That comes from asking questions and listening |
| Decide how to buy it | Partly. It can set out the options, but whether it's a straight purchase or a creative deal, like paying the vendor over time, depends on what the vendor needs and what you can fund |
| Negotiate and get it agreed | No. That's a conversation, often several |
| Put it through the legals | No. That's you, your solicitor and your broker |
A tool that finds you 50 houses a week hasn't moved you forward if you view none of them. The deal is made in the steps AI can't do.
The trap: proactive procrastination
In my experience the tools that find houses might save you a bit of time now and then. The trouble starts when somebody finds an AI tool that does loads of clever things. They spend their evenings getting it to find houses, analyse them and rank them. It feels like progress, because they're busy and it's all about property. Then a month goes by and they haven't viewed a house or made an offer. I see it more than anything else.
That's proactive procrastination. You're working, but on the safe jobs, so you never have to do the ones that make you uncomfortable. Analysing one more house, building one more spreadsheet and trying one more tool are all ways of putting off ringing the agent. Do enough of it and it becomes analysis paralysis: you've looked at every deal so hard that none of them ever feels safe enough to offer on.
3 jobs make you money in property:
- Viewings.
- Offers.
- Raising money.
Everything else, AI included, is there to support those 3. So the measure isn't how many houses you've analysed this week. It's how many you've viewed, how many offers you've made, and how many conversations you've had about money. If AI is helping those numbers go up, it's working. If they've stayed the same while your spreadsheet got bigger, it isn't.
How I use AI to save time
Used for the right jobs, AI is a massive time saver. The time it saves is time I can spend on viewings, offers and raising money. These are the jobs I use it for.
I use Claude. If you use ChatGPT for property investment, it works the same way. The tool matters far less than what you feed it and whether you check what comes back.
1. Pull the sold prices, rents and EPC data
This is the one that saves the most time. Before a viewing, I ask it for the sale history of the house, its EPC rating and floor area, and the sold prices of houses like it nearby. Then I ask for 2 numbers: what it's worth as it stands, and what it'll be worth done up. That second number is the end value, and the whole deal hangs off it.
My rules for a comparable are strict, and I give them to the AI every time:
- Same type, same age, same number of bedrooms.
- Floor area within about 8 square metres (about 90 sq ft).
- Within about 350 metres.
- Sold in the last 9 to 12 months.
- 3 good ones beat 10 loose ones. The closest match on size does most of the work.
The floor area comes from the house's energy certificate on the government's EPC register, and I work out the price per square metre from it. That's the number that lets you compare a big house with a small one.
The EPC matters for another reason too. At the time of writing, a rental in England and Wales needs a rating of at least E unless an exemption is registered, and the government has confirmed that will rise to the equivalent of a C by 2030, with spending capped at £10,000. So I ask AI for the rating and the improvements the certificate recommends, because a house on F or G needs that work in the budget before it can be let.
Here's the prompt I use. It's the nearest thing I have to a ChatGPT prompt for property investment, and it works in Claude or ChatGPT as long as web search is switched on:
I'm looking at a [type], [beds] bed, [floor area] square metres, EPC [rating], at [postcode]. Find sold prices for houses within about 350 metres, sold in the last 12 months, same type and bedrooms, with a floor area within 8 square metres. For each one give me the address, date, price, floor area, price per square metre and a link to the record. Don't use asking prices as evidence. If you can't find 3 that match, say so rather than widening the search without telling me. Then give me an as-is value and a done-up value, each as a range, and tell me which sales each one rests on.
The last 2 lines matter most. Without them it will happily fill the gaps, and you won't know which numbers are real.
I do the same for rent. I ask for houses like it that have let nearby in the last few months, with the rent and a link to each listing, and I check them on Rightmove myself. An asking rent is only what the landlord hopes for, so I'd rather work from the ones marked let agreed, which means the landlord has provisionally accepted a tenant.
2. Work out the most I'll pay
I never start from the asking price. I start from what I want left in the deal and work backwards, the same way I worked out the price on the cracked house I bought.
So I give it the end value, the refurb, the buying costs and the rent, and ask for the top price I can pay if I want all my money back out when I refinance, £10,000 left in, or £20,000 left in. It does 3 versions in a minute. Doing it by hand, I'd do 1 and guess the others.
The useful part is what it flags. Lenders check that the rent covers the mortgage interest by a set margin, called cover, worked out at a higher rate than you'll actually pay. At 125% cover, the rent has to be at least 125% of the interest. The Bank of England's rules for buy to let lenders say that stress rate should be at least 5.5%. On one house, AI worked out that the refinance would pass at 125% cover by about £12 a month. At the tougher 145% some lenders use, it failed. A deal that only works on the kindest lender's test isn't one I'd chase.
If you want to see what a full deal stack looks like once the numbers are in, I've set one out for a 2 bed I turned into a 3 bed.
3. Show the vendor the evidence
When I make an offer below what the vendor hoped for, I don't want them to think I've plucked a number out of the air. So I use AI to build a short sold price report: the 3 comparable sales with links to each record, the houses that are for sale nearby and how long they've sat there, what the works cost from published price guides, and what the market's done.
Every number in it can be checked. That's the point. The vendor can look up any line and land on the same page I did, and the report goes with a short message. In my experience a vendor believes a sold price they can look up themselves far more than a figure an agent gave them at the valuation.
4. Turn a vendor meeting into options
I'd rather listen to a vendor than stand there writing pages of notes, so before a meeting starts I ask whether they mind me recording it. I only ever record with their permission, asked in advance. If they say no, that's the answer, and I take notes instead. Most people say no. A few say yes.
When I do have a recording, I get a transcript and ask AI to pull out what the vendor actually said they need: how much, by when, and why. From that I ask for the ways the deal could be structured, and what each one gives them against an open market sale. What AI gives me is a first draft. I decide what goes in front of the vendor, and I'd rather say it to them in person than in writing.
Before I paste a transcript in, I take out the vendor's name and anything else that would identify them. I don't need it to get the answer. If you buy through a company, a vendor's name, address, debts and reasons for selling are personal data, and the ICO is clear that data protection law still applies when personal data goes into an AI tool, even by accident. I also check the tool's settings on whether my chats are used to train it. If you're not sure where you stand, the ICO's guidance on AI and data protection is the place to start.
5. Get a second opinion, then argue with it
When someone sends me a deal to look at, I'll often ask AI first. Then I argue with what it says, because this is where AI gets things wrong most.
Where AI gets property deals wrong
A mentee sent me a flat and asked whether it was below market value. The first answer AI gave me said yes, comfortably. I didn't believe it, so I pushed back, and it fell apart.
It had compared the price with an asking price it found linked to the address, and treated that asking price as what the flat was worth. Then it added that the flat had been up for sale at that price and hadn't sold. Nobody had said that. It had made it up to fill a gap. Once we worked it through properly, an old sale on the record looked to be for a share of the flat rather than all of it, and the price came out close to what the flat was worth. What looked like a bargain was a vendor who needed a buyer to fund their way out.
That's typical. These are the mistakes I see most:
- It treats asking prices as values. An asking price is what someone hopes for. Only a sold price is evidence.
- It fills gaps with facts nobody gave it, and says them as if they're true. OpenAI's own help pages say ChatGPT can make up facts and sound confident when it's wrong, and tell you to check important information against reliable sources.
- It trusts the property type on the record. The Land Registry only records 4 main types (detached, semi, terrace and flat), with end of terrace counted as a terrace. On one street I know, the records had a house down as a semi when it's a terrace. I only knew because I'd stood outside it.
- It uses sales that are too old or too far away, unless you tell it not to.
- It can't see condition. A sold price doesn't tell you whether the house was done up or falling down, so I check the old listing photos on the sold price page before I trust a comparable.
- It's too keen. Ask it whether something is a good deal and it leans towards yes. So when it says yes, I ask it what's wrong with the deal.
My rule is that every number AI gives me has to have a link to where it came from, and I check the 3 comparables myself before I use them. I never send a vendor a figure I haven't checked.
Can AI value my house?
The mistake that costs the most is trusting the value it gives you. AI can estimate one, but not in a way a lender will accept. Search for an AI property valuation UK tool and what you get is an estimate built from the same sold prices you can look up yourself. A ChatGPT house valuation is the same thing, only less reliable, because unless it's searching live it may be working from old data or guessing.
What decides your refinance is the lender's valuation, done by a surveyor. Surveyors who are members of RICS have had to follow a professional standard on using AI since 9 March 2026, and it keeps the surveyor's judgement, not the AI's output, at the centre of the advice. So use AI to get a good idea of value before you offer. Don't assume the bank's valuer will agree with it.
Where to find sold prices for property for free
You don't need a paid tool to check what AI tells you. These are free, and they're where the AI gets most of its numbers anyway.
| Source | What it gives you | What it doesn't |
|---|---|---|
| HM Land Registry price paid data | Every sale in England and Wales lodged for registration since 1995: price, date, address, type, and whether it's freehold or leasehold | Bedrooms, floor area or condition. New sales take 2 weeks to 2 months to appear, so the last 2 months are always incomplete |
| Rightmove's sold house prices pages | The same Land Registry sales, often with bedrooms and the old listing photos | Bedroom counts and types can be wrong, so check the photos |
| The EPC register | Floor area, energy rating and the recommended improvements, for the house and its comparables | Anything about price |
Use the 3 together and you can check any valuation AI gives you in 10 minutes. The catch is that it's 3 different sites, for every house you look at. That's the bit I pay to shortcut.
What's the best AI for property investment?
Which AI you use matters less than checking what it gives you. The best AI for property investment is the one you'll actually use, with web search switched on. I use Claude because it handles long documents and long conversations well, and I've built my own tools with it. ChatGPT, Gemini and Copilot can do the same jobs.
What matters more than which one:
- Turn on web search, or it'll work from what it already knows, which may be out of date.
- Give it your rules. My comparable rules and my numbers go in every time.
- Ask for links. If it can't link it, I don't use it.
- Ask what's wrong, not whether it's good.
Is an AI property deal analyser worth paying for?
The same goes for the apps. An AI property deal analyser is a calculator with estimated numbers already filled in: the end value, the rent and the refurb. It's AI property analysis with the answers filled in for you. That saves typing, but the estimates are the whole deal. If the end value is wrong by £20,000, every figure underneath it is wrong too.
So whether you pay for one or use a free one, I'd treat its output the same way as ChatGPT's. Check the end value and the rent against sold prices and let agreed rents yourself, then put your own refurb figure in. A good analyser does the maths. Only you can check the inputs. The people I mentor in the Property Accelerator get my BRRR and flip analyser (BRRR is buy, refurbish, refinance, rent) and my Deal-Killer Gate, a set of questions that tells you when to walk away. They work the same way: you bring the checked numbers, and the tool tells you whether the deal stands up.
What is property deal sourcing?
One more thing if you're thinking of using AI to find deals for other investors. Property deal sourcing is finding houses that work as investments, usually below what they're worth or with a vendor open to a deal on terms. Some people do it for themselves. Some do it as a business and sell the deals on to investors for a fee.
If you want to know how to use AI to source property deals for other people, the AI is the easy part. In England, a sourcer who passes deals on for a fee is usually doing estate agency work, so they need anti-money laundering supervision by HMRC and membership of a redress scheme under the Estate Agents Act 1979. Get that in place before you take a fee.
If you want help with the next one
AI gets you to a number fast. It won't book the viewing, make the offer or ask anyone for money, and those are the jobs most people put off. It won't tell you whether to trust the number either, or what to say to a vendor when the deal is close but not quite there.
That's what I help with in the Property Accelerator. Every month, each member tells the group what they're going to do: how many viewings, how many offers, who they're going to talk to about money. The next month they come back and say whether they did it. When they're stuck on a real deal, they bring it to the call and we work through it together. If you've done your training and you know you've been analysing more than you've been offering, the details are below.
If you'd rather check out anyone before you pay them, including me, I've written about how to find a property mentor you can trust, and there's more about how I invest on my About page. If the money is what's holding you back, I've written about how to find private investors for property and how property joint ventures work.
The small print
This post is general education about using AI tools when looking for and checking property deals, how I use them and how I understand the rules. It isn't legal, financial, tax, data protection or investment advice, and it isn't an offer or an invitation to invest in anything. I'm not authorised by the FCA, and neither is my company. AI tools change often and can be wrong, and no AI estimate is a valuation.
Lending rules, energy efficiency rules, estate agency and money laundering rules, and data protection law are as they stood when I wrote this, in November 2026, and are for England. They change, and each lender sets its own criteria. Check the current position on GOV.UK, legislation.gov.uk and the ICO before you rely on it. Always see qualified professionals before you buy: a RICS surveyor for value and condition, a specialist mortgage broker for the lending, a solicitor for the legal side, a tax adviser for tax and, where you need one, an FCA-authorised financial adviser.
Questions people ask
How do you use AI to find property deals?
Use it to search and sort listings, and to pull the sold prices, rents and EPC data for each house in minutes. Then check its links before you trust any number. The time it saves is only worth something if you spend it on viewings, offers and raising money.
Can AI find property deals?
It can find properties and spot clues like price drops, long time on the market or words like "probate". It can't view a house, find out what the vendor needs, negotiate or get a deal agreed, and that's what turns a property into a deal.
Can AI negotiate a property deal for me?
No. It can help you prepare, by pulling the evidence and setting out the ways a deal could be structured. The negotiation is a conversation between you and the vendor or the agent, and it's yours to have.
What is proactive procrastination in property?
Keeping busy on the safe jobs so you never have to do the uncomfortable ones. Analysing one more house, building one more spreadsheet or trying one more AI tool feels like progress, but if you're not viewing, offering or raising money, nothing is moving.
Will an AI tool give me an edge as a property investor?
Not on its own. Everyone using the same tool sees the same houses. The edge comes from what you do next: viewing quickly, making offers with evidence, and having the money lined up.
Is there an AI to find property deals for you?
There are paid AI property deal finder apps that rank listed houses by yield and return, and Rightmove has its own AI search. They all show listed houses with estimated numbers, and everyone else using them sees the same ones. Use them to sort houses, then decide for yourself.
What is AI property sourcing?
Using AI to search listings, read them for clues about the vendor and run the numbers, to find houses worth chasing. Some apps sell it as a service. It speeds up the search, but the deal still comes from the conversation with the vendor.
What's the best AI for property investment?
The one you'll use, with web search switched on. I use Claude. ChatGPT, Gemini and Copilot can do the same jobs. Give it your rules, ask for links to every figure, and ask what's wrong with the deal rather than whether it's good.
Can ChatGPT value a house?
It can estimate one from sold prices, if it's searching live and you tell it which sales to use. It isn't a valuation, and a lender won't accept it. Treat it as a starting point and check the comparables yourself.
Can AI value my house?
It can give you an estimate based on nearby sold prices. Only a surveyor's valuation counts for a mortgage or a refinance. AI's figure is worth having before you offer, as long as you've checked the sales it's based on.
How accurate is an AI property valuation?
Only as accurate as the sales it uses. It goes wrong when it uses asking prices, old sales, sales too far away, the wrong property type, or houses in a very different condition. Make it list every sale with a link, and check the closest match on size yourself.
What EPC rating does a rental property need?
At the time of writing, at least an E in England and Wales, unless an exemption is registered. The government has confirmed the minimum will rise to the equivalent of a C by 2030, with landlords' spending capped at £10,000. Check the rating on the EPC register before you offer.
Can AI find off-market property?
Not really. An off-market house isn't listed, so there's nothing for it to search. Some tools guess at owners who might sell from public records, but you find most off-market houses through letters, leaflets, agents and people you know. AI can help you check one once you've found it.
Where can I find sold prices for property?
For free, from HM Land Registry's price paid data, Rightmove's sold house prices pages, which use the Land Registry's records, and the EPC register for floor areas. The Land Registry doesn't record bedrooms or floor area, and new sales take up to 2 months to appear.
Does Rightmove have AI search?
Yes. Ask Rightmove lets you search in plain English and compare properties, and Rightmove also has an app in ChatGPT. At the time of writing, Rightmove says these are experimental and only shown to some users.
Is an AI property deal analyser worth it?
It saves typing, but its estimates for the end value, rent and refurb are the whole deal. Check those against sold prices and let agreed rents yourself, put your own refurb figure in, and only then trust the result.
Can I put a vendor's details into ChatGPT?
If you're buying through a business, their details are personal data and UK data protection law applies. Take out their name and anything that identifies them before you paste anything in, and check whether the tool uses your chats for training.
How to use AI to source property deals for investors?
The AI side is the same as finding deals for yourself. The rules are different. If you pass deals to investors for a fee in England, you need anti-money laundering supervision from HMRC and membership of a redress scheme before you take a penny.


