How to write better property descriptions with AI: a 2026 guide for UK agents

The first AI use case most UK agencies adopt is property descriptions, and it's the right one to start with. Low risk, fast feedback, and the time saving is visible in the first week. The problem is that most of the descriptions agencies are publishing right now read like they were written by an over-caffeinated American real estate blog. This guide walks through what actually works.

The two-prompt system

Stop using one prompt for everything. Build two.

The first prompt is your house style. Write it once, save it somewhere your team can find it, and reuse it on every listing.

You are writing property descriptions for [agency name], a UK [estate/lettings] agency in [region].

Style rules:
- British English (colour, neighbour, organise, optimise)
- Plain, factual tone. No superlatives. No exclamation marks.
- Short sentences. Aim for 12 to 18 words on average.
- Lead with the most distinctive feature of the property, not the location.
- Do not invent features. If a feature isn't in the spec I provide, don't include it.
- End with a one-sentence call to action that matches our brand voice.

Format:
- 180 to 220 words
- One opening paragraph (the hook)
- Three short body paragraphs (rooms, outside space, location)
- One closing paragraph (call to action)

The second prompt is the listing-specific brief. Paste it underneath the style prompt every time:

Write a description for the following property:

Address: [address]
Type: [house/flat/etc]
Bedrooms: [number]
Bathrooms: [number]
Distinctive features: [list]
Outside space: [garden/balcony/none]
Tenure: [freehold/leasehold]
Price: [amount]
Notes: [anything specific you want highlighted]

This two-part structure is the difference between getting useful copy and getting generic AI slop. The style prompt locks the voice. The brief prompt scopes the content.

Which tool to use

For 90 per cent of UK agencies the choice is between ChatGPT, Claude, and Gemini. All three handle this job well in 2026. The differences:

ChatGPT is the easiest to onboard a team onto, has the most third-party integrations, and the £20/month Plus plan is enough for a five-person office.

Claude writes more naturally for British audiences out of the box. The current model handles long context well, which matters when you're pasting in a full property questionnaire.

Gemini is bundled with Google Workspace, so if you're already on it the marginal cost is zero. Quality is now within touching distance of the other two.

If you don't already have a preference, start with Claude on the £20/month plan. Switch later if you outgrow it.

What to check before publishing

Three things break AI descriptions in production:

1. Invented features. The model will sometimes add a "stunning south-facing garden" when the brief said "small enclosed yard". Read every output against the original spec.
2. Bedroom or bathroom counts. This sounds basic but it happens enough to be worth an explicit check. Numbers carry through copy reliably about 90 per cent of the time, which means one in ten goes out wrong unless you check.
3. Tone drift. Even with a style prompt, longer outputs sometimes slip into superlatives. Cut anything that reads like a marketing brochure.

The time saving

A senior negotiator writes a property description in 12 to 20 minutes from scratch. With this two-prompt system, it's two minutes to paste the brief, 30 seconds to read the output, and three to five minutes to edit. Across a busy week that's an afternoon back per person.

For a typical five-person sales team, that compounds to roughly 800 hours a year of recovered time. Roughly £30,000 of cost recovered if you cost a negotiator at £40,000 fully loaded.

What we'd avoid

Auto-publishing without a human read. The risk-to-reward ratio doesn't work. The 30 seconds it takes to scan a description is worth it for the one in 50 cases where the model has invented something.

Wholesale switching to a property-specific AI tool just because it has "AI" in the marketing. Most of them are wrappers around the same models you can use directly for a tenth of the cost. Test the underlying tool before paying for the layer.

If you want a tailored view of where AI should fit your agency first, the AI Opportunity Assessment is £19 during the pilot and delivered within 24 hours.