rielty's AI search, explained: one box, three ways to find a home

Property portals ask you where. A town, a postcode, a radius. Then they hand you a wall of dropdowns and hope you can translate your actual life — the school run, the commute, the quiet street you're after — into min-beds and max-price. Most of what matters to you never makes it into the search at all.

We built rielty the other way round. There is one search box on the homepage, and you use it in plain English: "somewhere quiet with a garden, 40 minutes from Manchester, under £400k". That sentence becomes a real database search — actual filters over noise data, train timetables and sold prices — not a keyword match. Our slogan is "search by life, not location", and the search box is where that stops being a slogan.

One box, three ways to use it

The same box works three ways, depending on how you'd rather think.

• Type (search mode, the default). Write a sentence describing what you want, press enter, get results on a map. This is the fastest route when you already know roughly what you're after.
• Chat. Flip the search / chat toggle on the homepage composer and it becomes a conversation. You add requirements one at a time, ask questions about areas and prices along the way, and watch the results update beside the chat as you go.
• Talk. Inside chat mode there's a microphone. Tap it and speak normally; rielty transcribes, runs the same agent, and speaks its answer back. It's not a separate mode — it's the mic inside chat, for when you're walking the dog rather than sitting at a laptop.

All three run on the same engine. A budget you type, say or mention mid-conversation ends up as the same filter against the same data.

Why a sentence instead of dropdowns

Filter panels can only offer the questions the portal thought of. "Detached, 4 beds, £400k–£500k" fits neatly into dropdowns. "Near a good school and woodland" does not. "A girls' grammar within 2 miles" definitely does not. Yet those are the requirements people actually move house for.

A sentence has no such ceiling. rielty's search understands budgets and ranges, property type, freehold versus leasehold, floor area and plot size as separate things (acres and hectares included), places from a single postcode up to "Kent or Sussex" or "between Oxford and Reading", train commute time to a named city, walking distance to around 28 kinds of place — stations, woodland, the coast, parks, GPs, supermarkets, even castles and theatres — plus schools by Ofsted rating, phase, faith, sixth form or league-table position, flood risk ceilings, "quiet street", "low crime", listed buildings, conservation areas and green belt.

Simple queries like "3 bed in York under 400k" don't even touch the AI — they're parsed instantly by deterministic rules. Richer sentences go to a language model, which turns them into the same kind of structured filters. Either way, what runs against the database is a precise query, not a fuzzy guess.

One box, one sentence. The search / chat toggle sits top-right of the composer.

What the search is grounded in

An AI search is only as good as the data underneath it, so here is what's underneath it. rielty holds 15.8 million sold homes across England and Wales — every recorded sale since 1995 from HM Land Registry's Price Paid data, indexed to today's market using the UK House Price Index, and browsable at /sold-prices. On top of that sit the EPC register for floor areas, GIAS and Ofsted school data with GCSE and A-level league tables, Environment Agency flood risk, DEFRA's strategic noise mapping for road and rail, police.uk crime data going back to 2010, the full GB national rail timetable precomputed into station-to-station journey times, and Ordnance Survey OpenData for greenspace, rivers and addresses. Every source and its licence is published at /data.

So when you ask for "quiet", the answer comes from DEFRA noise mapping. When you ask for "45 minutes to London", it comes from real timetables. Nothing is vibes.

What it is honest about

The most useful thing an AI search can do is tell you what it didn't do. rielty shows you every filter it understood as a row of chips — remove one and the search re-runs. Anything it couldn't turn into a filter is shown back to you by name ("couldn't use: 'wine cellar'") rather than silently dropped. When nothing matches, it suggests specific relaxations, like dropping the bedroom minimum, with the number of homes each change would unlock.

And there are real limits, stated plainly:

Bedroom filtering deserves its own caveat: most sold records don't include bedroom counts, and the interface warns you about it rather than pretending otherwise.

Chat mode mid-conversation: the whole brief — budget, bedrooms, school, woodland — applied at once.

Where it fits in your house hunt

rielty won't book your viewings, arrange your mortgage or value your specific home from a chat (the free instant valuation tool on the site is the right door for that). What it will do is collapse the hours you'd spend cross-referencing school league tables, flood maps, crime stats and train times into one sentence — and show you, on a map, the homes that survive all of it at once. For the research you'd normally do street by street, how to research a house before you view it pairs well with it.

The practical next step is simple: go to the search, and instead of typing a town, type a life. "Family home near a good school and woodland, under £500k" is a perfectly good first search. See what it understood, prune the chips, refine in plain English, and let the map do the arguing.

Over the next articles in this series we'll take each way of using the box in depth — starting with the default.

This is part 1 of our AI search series. Next: Search mode — type what you want to live like.