Lovely to spend Thursday morning with you. Fifteen minutes is nothing like enough time to do justice to AI in property, so here's everything we covered, plus the bits we couldn't get to. In one place, no pressure, dip in when something catches you.
AI in property isn't a tech problem, it's a leadership opportunity. AI isn't a tool, it's a skill. The women getting the most from it aren't learning to prompt fastest, they're treating it like a thinking partner, drowning it in context, and keeping their judgement firmly in the room.
Everything on this page sits under that. The four jobs where AI is already earning its keep in property. The three rules that turn it from a search engine into a thinking partner. The billboard test that keeps you and your clients safe. And the offer from the room: a working AI agent, built with you in about an hour, yours to keep.
Same order we covered them in. Useful if you want to pick up a specific thread, or share it with someone on your team who wasn't in the room.
01 / 13
02 / 13
03 / 13
04 / 13
05 / 13
06 / 13
07 / 13
08 / 13
09 / 13
10 / 13
11 / 13
12 / 13
13 / 13You don't have to become a prompt engineer. You just need AI to write its own brief for you, and to sharpen anything you've already got. These two prompts do exactly that. Paste one in, answer the questions, keep what comes out.
For when you know the job you want AI to do, but not how to ask for it. Paste this into ChatGPT, Claude or Gemini. Answer one question at a time. You'll end with a prompt tailored to your actual work.
Write a prompt for me, for a piece of work I want AI to do. Ask me questions one at a time, maximum of eight, until you have enough. Then draft the prompt. Keep it plain, direct, specific to my situation. Nothing generic. Start with your first question.
For when something you've been using kind-of works but could be sharper. Most people's prompts have two or three things quietly working against them. This surfaces them.
Here's a prompt I've been using. Help me make it sharper. """ [paste your prompt here] """ Ask me questions one at a time, maximum of eight, until you see what's getting in the way. Then rewrite the prompt. Show me the original next to the new version and explain in plain English what you changed and why.
You don't have to transform your business to see the benefit. You have to pick one job, do it properly, then pick another. These are the four areas where the women ahead of the curve are already getting real time back.
First drafts in a fraction of the time, in your voice once you've given it the context. Same quality, far less effort. The tax your team quietly pays on written communication, reduced.
Summarising planning reports, reading local plan documents, extracting the relevant clauses from a 200-page PDF, pulling themes from agent feedback. AI is faster at this than any human, and doesn't get tired at slide 47.
Reviewing leases, flagging unusual clauses, preparing first drafts for legal review. Never a substitute for your solicitor. Always a good pair of hands before you hand it to them.
Summarising regulatory changes (the Building Safety Act, the Renters' Rights Bill, MEES changes, the slow drip of ESG reporting), preparing for ISO audits, turning a thousand-line spreadsheet into a board-ready two-pager.
You were generous with your questions on Thursday, and a few of them deserved a longer answer than I could give in fifteen minutes. Here's the set, with the bits I said I'd send you. Dip in wherever something pulls you. Nothing here is urgent. All of it matters when the right moment comes.
Before you paste anything into a free AI tool, ask yourself: would I be comfortable if this prompt appeared on a billboard outside my office? If the answer is no, don't type it. Free tools often use what you give them to train the model. Your confidential client data, planning correspondence, financial numbers and personal details of tenants or buyers all belong inside enterprise or team plans, not free accounts.
The fastest fix, worth doing this week: move anyone on your team who's using a free ChatGPT, Claude or Gemini account for real work onto a paid business plan. Small cost. Large exposure closed.
Ask the right questions. Don't just ask AI to do something, ask it to help you think through the problem first. The conversation is where the quality lives.
Drown it in context. Who's the client, what's the brief, what have you tried, what does good look like for you. The more you put in, the less generic it sounds on the way back.
Review and improve. Everything AI gives you is a first draft. Never publish, never send, never act on it without a human eye across the top.
AI hallucinates because it predicts probable text, not true text. It fills gaps with confident-sounding nonsense. Three habits keep you out of trouble.
One, ask for sources and actually check them. "Cite your sources and give me the page numbers." Then verify they exist. About a quarter of the time, they don't.
Two, treat every output as a draft. Never the final answer. Your experience is the editor.
Three, give it real data. Upload your actual documents, your numbers, your spreadsheets. AI is far more reliable when it's working from something concrete you've handed it, rather than reaching into its training for a plausible answer.
The single highest-leverage thing most firms can build in a week. Six short documents that give AI the same context you carry in your head. Why you exist, what you do, who your clients are, your tone, your strategy this year, your quality bar. Drop them in a shared folder and every piece of AI work your team produces gets sharper, immediately, without anyone learning anything new.
Paste this into your AI of choice. Answer one question at a time. You'll have the six documents drafted by the end of the afternoon.
Help me build a context library for my business. Ask me the questions you need to write six short documents that will make any AI work I do more accurate and more on-brand. The six documents are: 1. Why we exist (our purpose, what we're really here for) 2. What we do (our products, services, how we deliver) 3. Who we serve (our clients, who they are, what they care about) 4. Our voice (tone, language, what we sound like) 5. Our strategy (where we're going, our priorities this year) 6. Our quality bar (what "done well" looks like for us) Keep each document to under 400 words. Plain English, no jargon, written so anyone on my team could read it once and get it. Ask me one question at a time. Don't move on until I answer. When we've got enough, write all six and show them to me in order.
You need three versions, not one. An internal policy for how your team uses AI day to day. An external policy for clients and partners who will ask how you manage AI. And a water cooler policy, a single side of A4 by the coffee machine, in plain English, that sets the culture in public. The water cooler version is the one most people skip. It's the one that matters most.
Indifference, fear, cheating, confidence, transformation. Most people in property are somewhere between stage two and stage three. Whichever stage you're at, the one thing that moves you forward faster than anything else is practice on a real piece of your own work, not a tutorial. Which brings us neatly to the next thing.
Not what I tell clients to use. What's on my own machine, earning its keep daily, that I think will earn its keep for you too.
You think faster than you type. The emails you put off because writing them feels heavier than thinking them, the long replies you compress into three words, that's a quiet tax on your thinking all day. Wispr Flow takes the tax away. Hold a key, talk, watch clean text appear wherever your cursor is. Brilliant for agents on the move, long client update emails, meeting write-ups, and every piece of a proposal that isn't the numbers.
I wrote most of Thursday's deck with it. I'm writing this with it. If you want a look, it's at wisprflow.ai.
Google's research workspace and, in my view, the most underused AI tool for property right now. Upload up to fifty documents (planning packs, leases, reports, council minutes, site surveys), then ask it questions. It answers only from your documents. No hallucinations from the wider internet, no confident nonsense. It will cite the line and page. For due diligence, for summarising local plans, for getting your head round a new scheme quickly, nothing else comes close.
Free to try, paid plan is modest. notebooklm.google.com.
The deck above was built in Gamma. No designer, no overnight panic. You brief it the way you'd brief a smart junior, it drafts a complete deck, you spend your time editing the thinking not the pixels. For anyone who pitches, presents at Women in Property events, runs client workshops or sends board updates, it's the single cheapest hour back in the week I know of.
It's at gamma.app.
Pick one job you do over and over. The client update email that eats a Friday afternoon. The first-draft lease report. The comparable write-up. The monthly board summary. Bring one golden example and a couple of what-great-looks-like examples. We'll build you a working agent for that job together. About an hour of your time. Yours to keep.
You asked good, sharp questions on Thursday. Here are the ones that kept surfacing, answered a little more fully than I could in the time we had.
The honest analogy: think of it as cloning your best team member for one specific job. The one who already knows how you like proposals written, what a proper client update sounds like, what flags to raise on a dodgy lease. You brief them once, properly. From then on, they just do the work.
Technically, an agent is five short files. One says who it is. One holds what it knows (your precedents, your tone, your templates). One says what it does. One says what it can touch. One says what it must never do. Unlike a prompt, it has judgement. Unlike a workflow, it handles variation. Unlike a person, it doesn't need a coffee break.
Pick one thing. Not ten. The email you've been meaning to write to a landlord client. The report summary that always runs late. The first draft of a brief that keeps slipping to the weekend. Drop it into ChatGPT or Claude with the context from the billboard test comfortably applied, have a proper conversation about what good looks like, iterate twice, save the prompt that finally worked.
That's it. One win this week. Do the same next week. Within a month you'll have a small library of prompts that save you hours, and you'll be ready for the next thing.
Anonymise before you paste. Off-market instructions, personal tenant data, live-deal financials — replace the real names and numbers with placeholders. AI can still help with the structure, the language, the arguments. Your client's data stays out of it.
Pair that with the billboard test and the paid-plan move from the threads above, and you've covered the three things that matter most on data.
Not the ones who learn to lead with it. The shape of junior work changes. Less time typing, more time thinking. Less research grind, more synthesis. The junior surveyor, planner or agent who becomes fluent with AI this year will do work that looks like a middle-career person's work two years ago. That's a gift to them, and a gift to your firm.
The risk isn't AI replacing your team. The risk is your team quietly becoming dependent on it without developing their own judgement. The fix is pairing AI with proper mentoring, and making review of AI output part of how juniors learn the craft.
You are. The director, the agent, the surveyor, whoever signed it off. AI is unpaid staff, not a legal defence. The same duty of care applies as if you'd written every word yourself.
Which is why human review on anything you publish or send, IP awareness on anything you upload, and clear internal quality gates all matter from day one, not from incident one. The three-lens policy above is the cheapest insurance you can buy.
Worth being honest about, usually answered with the wrong numbers. Eating a single beef burger uses more water than thousands of AI prompts. The viral bottles-per-query figure that circulates on LinkedIn doesn't reflect how modern data centres actually cool their hardware. The ecological tax on your next cheeseburger is larger than your ecological tax on AI this year.
Worth knowing. Not worth stopping over. If you want to be rigorous, track AI use alongside your existing ESG metrics rather than treating it as a separate moral question.
It isn't the tool. It isn't the prompt. It's having made AI do one real piece of your own work, well. Once. That's the moment it stops being abstract and starts being useful. Everything else follows.
That's what the agent-build is for. One hour, one job, yours to keep. Whether it's the client update email, the due-diligence summary, the compliance report or the proposal that always runs late, we'll build it together. You'll leave with a working agent and the quiet confidence that nothing we covered on Thursday was theoretical.
Pick your start:
Thank you for having me on Thursday. Your questions sharpened my thinking more than you know.
Mary