Guide

The AI formula: model, interface, tools and context

Guillaume Black Chief Executive, Property Filter Last updated 1 October 2026

Ask people in property which AI they use and everyone has a favourite: ChatGPT, Claude, Gemini, Grok, DeepSeek. The argument about which one is best is a losing battle, because the model is only one of four things that decide what you get back. This guide explains all four, with the property jobs that show each one at work.

What decides what AI can do for you?

Four things: the model, the interface, the tools and the context. Picture a new colleague: the model is the person, the interface is the desk they sit at, the tools are the computer and its plugs, and the context is what is in the drawers.

Written as a formula, it reads:

Model × interface × tools × context

• Model: which colleague does the work, the quick one or the deep one.
• Interface: where it works, in the lobby or at a desk.
• Tools: the plugs into your inbox, your diary, your CRM.
• Context: the briefing only you can write.

The sign between them is a multiplication, and that is the point. Use the model on its own, in a chat window, with nothing plugged in and nothing about you, and you do not stand a chance. You get busier and busier, not more and more capable. Why that happens has its own guide.

Which AI model is the best one?

That is the wrong question. Intelligence is now a commodity: you rent the same models as everyone else, so the edge is what you give the model and which one you pick for which job.

The Artificial Analysis Intelligence Index, an independent ranking, puts the leading models from several different labs close together, all in the same race, and the order keeps changing. Whichever one you choose, a competitor can rent the same one tomorrow.

So the model will not be your advantage. The briefing can be, because nobody else can write it for you. Everyone rents the same model. Only you can write the briefing.

Should I use the quick, medium or deep model?

Pick per job, not per brand. The quick one for a short task with the facts in front of it, the medium one for one document and one judgement, the deep one for many files and a long result.

Think of them as three colleagues with different strengths.

Every AI app offers a fast option, a standard one and a deeper thinking one, in the model menu at the top of the chat, whatever it calls them this month. The names change every few months. The rule does not.

When an answer disappoints, read the symptom before you switch.

What is the difference between the lobby and the desk?

The lobby is a chat window in your browser: it talks, you copy and paste, and tomorrow it is gone. The desk is the AI app on your computer, allowed into a folder: it opens, reads and saves your files, so it works.

Same colleague. The desk is what changes. The interface decides whether the AI can touch your work or only talk about it.

A long briefing pasted into a chat does not turn the lobby into a desk. It is still the lobby until the AI can open, read and save the files in a folder on your computer. Look for the words "desktop app" and "folder" at your AI provider. The Claude desktop app, allowed into one folder on a laptop, is one example of a desk.

The job: spotting motivated sellers

Before, in the lobby. You ask a chat window to spot motivated sellers. It reads page one of Rightmove and tells you the auction lots stack up at their guide price. You dig, you ask again, and you fall into another rabbit hole. An hour in, nothing is done.

After, at a desk. You wake up to a hitlist, every morning: motivated sellers, qualified and stacked against your criteria, each with the questions to put to the agent or the vendor. Same model. One gives you a rabbit hole. The other gives you a hitlist.

The job: the investor deal pack

Before, in the lobby. You paste the numbers into a chat. You get text back. You copy it into your template. Tomorrow it is gone.

After, at a desk. It reads the deal file and your pack template in the folder. It writes the pack in your format and saves it next to the deal. Next quarter it starts from that file. Same model, same subscription. The desk is what changed.

"One hour of dictating. A branded investor pack that used to take me a full day."

Michael C., runs 20+ properties in Yorkshire

What is a connector, and which ones does a property business need?

A connector is the plug that lets your AI reach a tool you already use, such as your email, your diary or your CRM. Without a plug it chats inside the chat box; with one, it does the work.

The plugs a property business tends to use:

• Email: it reads, and drafts replies you send.
• Calendar: it reads your diary and flags dates.
• Calls and call notes (Zoom, Granola): your calls recorded, and the notes from them.
• Tasks and projects (Asana): your task list.
• Your CRM (GoHighLevel, for example): your leads and pipeline.
• Payments (Stripe): what has come in.
• Slack: where it sends notifications and reports back to you.
• A browser: it reads a listing itself.
• Property Filter: the property data.

The starter stack we recommend for a property business is short: Claude Pro as your AI colleague, Gmail on Google Workspace with your own domain, Google Calendar, Google Drive for your files, Slack for where it reports, Granola for your call notes, and Property Filter for deal-making data and niche knowledge.

The job: lead triage, every morning

• It reads. The new leads in Property Filter and the overnight enquiries in your inbox, through the Property Filter, email and CRM plugs.
• It routes. Each lead against your lead rules, hot, warm or wait, with the first reply drafted. The rules live in your context.
• You call. Three names on your desk at 8am, with the reason each one is hot. You keep the conversation.

A plug is not permission to act on your behalf. Connect it to read and to draft. You still press send.

What does context mean in AI, and why does it matter most?

Context is the briefing: what the AI knows about you, and what it knows about your niche. It is the one factor nobody else can rent, because only you can write it.

Half one: what it knows about you

• Your goal: what you want to earn, and by when.
• Your money: what you can deploy, and what you cannot lose.
• Your patch: your postcodes, and why you ruled out the ones next door.
• Your floor: the least you will accept on a deal.
• Your never-do list: what you will not touch, whatever the numbers say.
• Your history: every offer you have made, and what came back.

Half two: what it knows about your niche

• What is working: which strategies are getting offers accepted this quarter.
• Motivated sellers: what one actually looks like on a listing.
• Back and Reduced: a listing back on the market at a lower price, and what that is worth to you at the table.
• The six Ds: why a seller is really selling (defect, divorce, downsizing, death, debt, departure).
• Real sold prices: last quarter's, not asking prices.
• The Renters' Rights Act: what it changes for you from May 2026.

Half one is written once, in a file, and read on every job. Half two is what the internet does not know about your market this quarter, which is why a general assistant cannot supply it from memory.

One current file per subject beats a hundred chats. Keep one version of your criteria, move the old ones out of the way, and it reads the right one tomorrow.

The job: the viewing brief

Before, without the briefing. You ask a chat window for viewing questions. You get ten sensible ones anyone could ask: why are the owners selling, how long has it been on the market, is there a chain, have there been any offers. Nothing about your yield floor or the chimney breast. Gone when you close the tab.

After, with the briefing. It reads your buying criteria, the agent's call notes and the property's history. The questions are yours, built for this house. Can you share the building regulations certificate for the loft conversion? When will the revised EPC be available? It was with another agent before: do you know why it did not sell? After several price reductions, has the seller found their next property? Plus the damp patch, the board type, and the number to confirm before you leave. Saved for tomorrow.

What are the common mistakes?

Most of them come from fixing the wrong factor. Here are the usual ones, and what to do instead.

• Changing brand when an answer is poor. Fix the briefing first, then change colleague.
• Using one model for everything. Quick for the short job, medium for one document, deep for many files.
• Giving it your summary instead of the document. Give it the source, and have it point to the line it relied on.
• Mistaking a long chat for a desk. It is still the lobby until the AI can open, read and save files in a folder on your computer.
• Letting a plug send. Connect your tools for reading and drafting. Sending stays with you.
• Keeping five versions of your criteria. One current file per subject, the old ones moved out of the way.
• Leaving half two to the internet. A general assistant does not know this quarter's market. Give it real sold prices, not asking prices.

What the Property AI Brain does for you here

This guide explains the four factors. The Property AI Brain sets all four up for a UK property business. Built by Property Filter, it works with the AI you already use, ChatGPT, Claude or Gemini, with your notes and documents as plain files in a folder on your own computer, and an optional Property Filter account key connects the live data. It interviews you once about your patch, your numbers, your rules and your power team, so you brief it once, not every time. And every job in a property business already has its command, built on Property Filter's playbooks, so there is no prompt to write: you press the command, answer its questions, and the finished work comes back to you.