AI for estate agents: where it earns its place, and where it wastes money

The short answer. AI earns its place in an estate or letting agency on the reading and writing nobody enjoys: first drafts of listing copy, replies to portal enquiries, the same tenant question answered for the ninth time this week, file notes, landlord updates. It will not win you instructions and it will not carry your compliance. Treat it as a fast, tireless junior with no memory and no judgement, and the time starts coming back inside a week.

I qualified as a chartered surveyor and worked in commercial property before leaving the profession to build and automate businesses. What follows is the version I give in a training room, not the version a software vendor puts in a brochure.

Where does AI earn its place in an agency?

On any job where the input is a document or a repeat question and the output is words on a page. That is a far bigger share of the week than most agents admit until you sit down and count it.

Listing copy. A first draft built from the measurements, the EPC, the floorplan and your own rough notes, in your firm's tone, in seconds. You then fix it, because you stood in the house and it did not.
Portal enquiries. A drafted reply that names the right property and proposes the right next step, sitting in the outbox waiting for a person to press send.
Repeat tenant and applicant questions. Deposit timings, referencing, what happens on move-in day. Those answers already live in your own documents. Point AI at them and the ninth reply costs you nothing.
Landlord and vendor updates. Viewing numbers, the themes in the feedback, what you recommend doing about the price. Assembled from what you already hold, rather than written from a blank page at five to six on a Friday.
File notes and call summaries. The record you know you should keep and skip when the day gets away from you.
Long documents. Leases, management packs, search results. AI pulls out the clauses that matter so someone reads twenty pages instead of two hundred.

Every one of those ends with a person checking the output. That is not a limitation you design around. That is the design.

Where does it waste your money?

Anywhere the value was the relationship, the local knowledge or the judgement. An AI that writes a beautiful valuation letter has done nothing about the fact that the vendor picked the agent who sat in their kitchen and told them the truth about the price.

Three specific ways firms lose money here:

Buying a tool before fixing the job. Most agencies have a handful of jobs eating the week, and nobody has written them down. Software bought before that exercise gets used by two people for a fortnight and then quietly lapses.

Automating the front door. A chatbot that intercepts enquiries feels like progress and reads to a buyer as a wall. The enquiry is the one moment your competitor would pay for. Draft the reply with AI by all means, then have a human send it fast.

Paying for AI you already own. Firms on Microsoft or Google are frequently paying for a third tool that duplicates something their existing licence covers. Check what you have before you buy.

Who is on the hook when AI writes the description?

You are. That does not shift because a tool produced the first draft.

This is the part vendor content skips, and it is the part that matters most in agency. A property description is a representation you are making to a buyer or a tenant. AI writes fluently and confidently about things it has no knowledge of, so it will happily produce "recently refurbished throughout" or a tenure or a council tax band it inferred from context rather than read from a document. Fluent and wrong is worse than clumsy and right, because nobody proofreads text that reads well.

The working rule I teach: AI drafts anything, AI states no fact. Every number, date, tenure, measurement, band, restriction and material fact in a listing gets its value from your source document, not from the model. Set that habit in week one and it holds when the team gets busy and complacent.

The same applies to complaints and disputes. Consumers now write to agents with AI help too, so complaints arrive longer, better structured and more assertive than they used to be. That changes the tone of your reply, not the substance of your obligations.

Why lettings feels it first

Lettings and property management are more repetitive, more deadline-driven and more document-heavy than sales, which is exactly why AI lands there fastest.

Sales work is lumpy and relationship-led. Lettings work is a conveyor belt: referencing, renewals, notices, certificates, deposit queries, maintenance triage, arrears, the same twelve questions from a hundred tenants. Repetition is what AI is for. If you run both sides of a business and want a first win people notice, start on the lettings desk.

What to do on Monday

Skip the strategy document. Do this instead.

Count the hours. Ask each person to list the three jobs that eat the most of their week. Do not lead them. Patterns show up within about ten replies.
Pick one job, not six. The one that is repetitive, low-judgement and annoying. Arrears chasing, viewing feedback and applicant queries are the usual winners.
Write the brief, not the prompt. Feed the model what you would tell a new starter: your tone, your process, an example of a good output and an example of a bad one. Most disappointing AI output is a briefing failure, not a model failure.
Run it in draft mode for two weeks. A person presses send on everything. You are buying trust and finding the failure modes on cheap ground.
Then decide what to automate. Once the team has seen where it works, the jobs worth building into something that runs on its own pick themselves.

That order matters. Teams trained first adopt what gets built afterwards. Teams handed a finished system they did not ask for route around it.

Frequently asked questions

Will AI replace estate agents? No, and the reason is structural rather than sentimental. The paid work in agency is winning the instruction, pricing it honestly and holding a chain together when it wobbles. All three run on trust and local knowledge. What AI takes is the admin sitting on top of those jobs, which is the part agents complain about anyway.

Is it safe to put client or tenant data into AI tools? It depends entirely on the tool and the licence tier, and the free consumer versions are the ones to be careful with. Business and enterprise tiers of the main tools handle data differently from the free ones. Before anything real goes near a tool, get a straight answer on what happens to what you type, put it in writing, and tell the team which tools are approved and which are not.

What should an agency use, ChatGPT or Copilot? Whichever your firm is allowed to use. If IT has locked things down and cleared Microsoft only, start with Copilot rather than fighting that battle. The habits transfer between tools, so the skill you build on one moves with you to the next.

Do I need a big system, or does the free version do? Start with the free or low-cost version and your own documents. Most agencies get their first visible win without buying anything. Buy software once you know precisely which job you are buying it for.

How long before the team sees a difference? Days, if you pick a genuinely repetitive job and train people on their own work rather than on a generic demo. Weeks, if you start with a tool and go looking for a use for it.

What is the most common mistake? Letting AI state facts. Everything else on this list is recoverable. A confident, wrong detail in a listing or a landlord report is the one that costs you.

If you want your team trained on this, using your own listings, your own tone and your own awkward jobs, book a session.