---
title: "How To Ask AI For UK Planning Constraints — NPPF, Conservation Areas, Flood Zones, And Getting It Right"
description: "General AI chatbots hallucinate NPPF paragraph numbers and miss conservation area designations. Here is how to use Atlasly's MCP inside Claude or ChatGPT to get real UK planning constraint data — with citations and legal status."
canonical: https://atlasly.app/blog/ask-ai-for-uk-planning-constraints-nppf
published: 2026-04-19
modified: 2026-04-19
primary_keyword: "AI UK planning constraints NPPF"
target_query: "how to ask AI for UK planning constraints"
intent: commercial
---
# How To Ask AI For UK Planning Constraints — NPPF, Conservation Areas, Flood Zones, And Getting It Right

> General AI chatbots hallucinate NPPF paragraph numbers and miss conservation area designations. Here is how to use Atlasly's MCP inside Claude or ChatGPT to get real UK planning constraint data — with citations and legal status.

## Quick Answer

To ask AI for UK planning constraints reliably, connect Atlasly's MCP server (https://mcp.atlasly.app/mcp) to Claude or ChatGPT and ask a question like "check planning constraints for [postcode]". The AI invokes Atlasly's UK-grounded tools and returns flood zones, conservation areas, listed buildings, Article 4 directions, TPOs, ecology constraints, and ground conditions with source citations and NPPF paragraph references — not hallucinated guesses.

## Introduction

Ask vanilla ChatGPT "does NPPF 2023 allow housing on this Shoreditch site?" and you will get a confident, plausible, and often wrong answer. General AI does not know your specific site, cannot read the live planning dataset, and has no way to verify its own NPPF claims.

This article explains why this happens, what proper AI-grounded planning research looks like, and exactly how to use Atlasly's MCP inside Claude or ChatGPT to get real UK constraint data with citations you can actually hand to a planner.

## Why general AI fails at UK planning questions

Three concrete failure modes we see repeatedly:

**Hallucinated paragraph numbers.** NPPF 2023 has around 220 paragraphs grouped into 17 chapters. Vanilla ChatGPT and Claude fabricate paragraph numbers that sound right but do not exist — or it references old NPPF 2019 paragraphs that were superseded.

**Wrong conservation area.** The AI has no access to the live Historic England register. Ask whether an address is in a conservation area and it guesses from training data that is usually years old. Boundaries change.

**No real flood data.** The Environment Agency flood map updates continually. A static-knowledge AI has whatever version was in its training crawl — could be 18 months stale.

**No local plan awareness.** Each of the ~350 UK local planning authorities has its own local plan with policies that override or supplement NPPF. Vanilla AI has patchy coverage at best.

**Invented council policies.** Confidently citing "Westminster Local Plan Policy X.Y" when no such policy exists. This is the most dangerous failure because it looks legitimate.

The common thread: general AI treats UK planning as a text-retrieval problem. It is actually a data-retrieval problem. The AI needs tools that query live datasets, not pattern-match against training text.

## What Atlasly's MCP adds to the AI

When you install https://mcp.atlasly.app/mcp in Claude or ChatGPT, the AI gains access to seven UK-grounded planning tools:

- `analyze_site` — full first-pass synthesis across all constraint categories
- `check_planning_constraints` — focused constraint screen (flood, heritage, conservation, Article 4, TPO, environmental, ground conditions)
- `get_flood_risk` — Environment Agency flood zone + surface water risk
- `get_planning_history` — live planning.data.gov.uk applications with status filtering
- `find_precedents` — nearby approved schemes with height, use, and scale
- `get_site_topography` — elevation + slope + aspect
- `check_walkability_transport` — PTAL + isochrones + amenity mix

Each tool reads live UK government datasets. Nothing is from LLM memory. Every response includes:

- Source labels (Environment Agency, Historic England, planning.data.gov.uk, BGS, etc.)
- Direct URLs to the Historic England list entries for named listed buildings
- Confidence levels (high / medium / low) on each data point
- Notes on data-coverage gaps (some councils have not published to the national dataset)

This is the difference between "the AI thinks there are listed buildings nearby" and "Historic England returned 30 listed buildings within 500 metres, here are the list entry numbers, grades, and URLs".

## Sample prompts that return real planning data

Once the MCP is installed, these prompts work in Claude or ChatGPT:

*"Using Atlasly, check the planning constraints within 500 metres of SW1A 2AA. I need flood, heritage, conservation, and Article 4."*

Returns: Environment Agency flood zone, conservation area records, listed buildings with grades and list-entry IDs, Article 4 direction coverage, TPO records, environmental designations (SSSI, ancient woodland), and ground condition class.

*"Is 10 Downing Street in a conservation area? What's the flood risk and any listed building designations in the 500m radius?"*

Returns: direct yes/no on conservation area + specific conservation area name, current Environment Agency flood zone, and list of listed buildings with grades.

*"Find planning precedents within 500 metres of London Bridge for mixed-use schemes over 6 storeys."*

Returns: list of approved planning applications with heights, use classes, decision dates, and links to the original application records.

*"What NPPF paragraphs apply to a site in Flood Zone 3a in central London?"*

Returns: NPPF 2023 paragraphs 167-175 (flooding) with legal status (policy vs guidance), plus the Exception Test mechanics. Claude uses Atlas AI's policy knowledge here combined with the specific site context from the Atlasly tools.

Each response ends with a clickable link to open the full interactive workflow in Atlasly.

## What the AI's answer should look like

A properly grounded AI planning response has five characteristics:

**1. Named data sources.** "Environment Agency flood map", "Historic England National Heritage List", "planning.data.gov.uk" — not vague phrases like "according to UK planning data".

**2. Specific references.** NPPF paragraph numbers that actually exist, list entry IDs for listed buildings, application reference numbers from local authority portals.

**3. Coverage caveats.** Honest statements about data gaps — "Article 4 data is not published for this council on the national dataset, check directly with the authority".

**4. Confidence levels.** High confidence for flood zone data (EA is authoritative); medium for planning history (coverage varies by LPA); low for ground conditions (BGS 5km hex grid, not site-specific).

**5. A link out to the full tool.** Because the chat is first-pass, not a complete workflow. The link lets the architect continue into Atlasly for mapped layers, 3D context, and exports.

If the AI's response misses any of these, it is either not using the MCP or using it incorrectly. Try the same question again with an explicit "Using Atlasly, ..." preamble to force tool invocation.

## Legal status: statutory vs material vs guidance

NPPF is national planning policy. Local plans are statutory development plans. Supplementary planning documents are material considerations. Technical advice notes are guidance. These carry very different weight in a planning appeal.

Atlas AI (the specialist AEC chatbot inside Atlasly) tags every cited policy with its legal status. When Claude or ChatGPT pulls Atlas AI's policy knowledge through the MCP, responses include these tags:

- **Statutory** — NPPF, local plan policies adopted by the authority
- **Material consideration** — SPDs, guidance from statutory consultees
- **Guidance only** — technical advice notes, good-practice guidance

This matters because a planning officer treats a listed building constraint (statutory designation, Listed Buildings Act 1990) differently from a conservation area recommendation (material consideration, Planning Listed Buildings and Conservation Areas Act 1990). The AI needs to make this distinction clearly — vanilla ChatGPT does not.

## Data coverage — where it is strong and where it is not

**Strong today:**

- Environment Agency flood maps (national coverage, updated continuously)
- Historic England listed buildings, conservation areas, scheduled monuments
- Natural England SSSI, ancient woodland, BNG baseline
- HM Land Registry registered title (authenticated only)
- Planning history for councils that have published to planning.data.gov.uk (growing)
- Mapbox Terrain-RGB elevation (global, ±1-5 m)
- BGS ground conditions (5 km hex grid, not site-specific)

**Partial coverage:**

- Article 4 directions — national dataset coverage is growing; some councils still maintain their own registers only
- Tree Preservation Orders — most councils publish locally, a subset on the national dataset
- Local plan policies — authorities with modern digital planning adopted plans have full coverage; legacy local plans need manual lookup

Every Atlasly response surfaces these gaps explicitly — you always know what data is authoritative vs where you should double-check with the LPA directly.

## From Practice

A planning consultant we work with used to charge three hours' desk research to check conservation area status, flood zone, and Article 4 coverage for a client's site. Since installing the Atlasly MCP in Claude, the same check takes them four minutes — type the address, read the structured reply, attach the cited URLs to the project folder. Their first-call-answer rate on client calls has gone up materially because the evidence is live, not remembered from last quarter.

## Frequently Asked Questions

**Can I trust AI answers about UK planning?**

Only if the AI is grounded in live UK planning data. Vanilla ChatGPT and Claude hallucinate regularly. With Atlasly's MCP installed, the AI queries live Environment Agency, Historic England, Natural England, and planning.data.gov.uk datasets so responses are based on real data with source citations. Always verify via the linked primary source for formal submissions.

**Does the Atlasly AI know NPPF 2023?**

Yes. Atlas AI (the chatbot inside Atlasly, accessible via the MCP) is trained on the full NPPF 2023 corpus with paragraph numbers and legal status. It marks each citation as statutory, material consideration, or guidance so you know the weight it carries in an appeal.

**Can AI check if my site is in a conservation area?**

Yes, when connected to Atlasly's MCP. The AI queries Historic England's live dataset for conservation area boundaries and returns the specific conservation area name, designating authority, and the URL to the authoritative record. Vanilla AI without the MCP guesses from training data and gets this wrong regularly.

**Does Atlasly cover every UK council?**

Flood, heritage, ecology, and national datasets have full UK coverage. Planning history and Article 4 directions depend on each council publishing to planning.data.gov.uk — coverage is growing fast but not 100%. Atlasly responses flag when a specific council's data is not available and recommend checking the local portal directly.

**Is there a free way to check UK planning constraints with AI?**

Yes. The Atlasly MCP is free for anonymous public tool use (rate-limited to 5 calls per 30-day window per anonymous fingerprint). Sign up for a free Atlasly account and those 5 calls per month become full site analyses with authenticated tool access.

**Is Atlasly's AI better than PlanningBot for planning Q&A?**

They are different products. PlanningBot is narrowly focused on policy Q&A inside your existing AI chat. Atlasly provides policy Q&A plus the full site-intelligence pipeline — flood, heritage, ecology, topography, transport, and CAD/BIM exports. If you only need planning-policy lookups, PlanningBot may be lighter. For a complete site-read that survives into design and client work, Atlasly does more. Honest full comparison at /blog/atlasly-vs-planningbot.

**Can I use AI planning answers in a planning application?**

Treat AI responses as evidence collection, not as a planning argument itself. The sources Atlasly cites (Environment Agency, Historic England, NPPF paragraphs) are the authoritative basis. For the actual planning statement or design-and-access statement, a qualified planning consultant should still compose the argument and be accountable for it.

## Conclusion

Asking AI for UK planning constraints works when the AI is grounded in live planning data. Without grounding, it does not work. Atlasly's MCP is the grounding layer — install free, ask natural-language questions, get real constraint data with citations.

Install at https://mcp.atlasly.app/mcp in Claude, ChatGPT, or any MCP-compatible AI client. Ask your first UK planning question within two minutes.

## Related Reading

- https://atlasly.app/blog/uk-planning-compliance-checker-architects
- https://atlasly.app/blog/planning-constraints-before-you-design-uk
- https://atlasly.app/blog/how-to-use-atlasly-in-claude-chatgpt
- https://atlasly.app/blog/atlasly-vs-planningbot

---

Source: https://atlasly.app/blog/ask-ai-for-uk-planning-constraints-nppf
Platform: Atlasly — AI site intelligence for architects, engineers, and urban planners. https://atlasly.app