17,000 branches. 80% portal share. No native AI-search layer.

An independent high-street branch runs 5–15 FTE. A regional group office (Hamptons, Romans, regional Connells, KFH satellite) runs 15–35 FTE. A flagship London branch (Foxtons Chelsea, Savills Mayfair, Knight Frank prime) runs 35–80 FTE. Roles cluster around branch manager, sales negotiators, lettings negotiators, property managers, admin, and a centralised marketing function at group level.

There is no dedicated AI-search owner at the branch level, and rarely at group. Marketing and SEO sit with one or two in-house generalists or an outsourced regional digital agency that still treats AI overviews as an experimental channel. The branch principal reads *Property Industry Eye* and *Estate Agent Today*, not US SaaS marketing. The register that wins is operational, fee-anchored, evidence-led.

Cash collection is portal-tilted and increasingly painful. Rightmove charges an average £1,432 per branch per month and posts 70% UK-portal share by visits across the big three (Rightmove 88M monthly visits, Zoopla ~50M, OnTheMarket ~16M per Similarweb 2024).

The portal subscription is the single largest controllable line on most branch P&Ls. Agencies overspend on Featured Property and Premium Listing add-ons hoping to compensate for thin own-channel inbound. That gap is the structural opening AI-search and Foundation work address.

The instruction economy is local and reputational. A seller decides between two or three branded shopfronts and one or two online challengers, then checks reviews on Google, allAgents, Feefo, and Trustpilot before booking the valuation. The agency that gets named first in a ChatGPT or Perplexity answer to "best estate agent in {borough}" lands the valuation appointment before GetAgent or Rightmove Discover intermediates the lead.

Renters' Rights Act 2025 (Royal Assent 27 Oct 2025) abolishes Section 21 no-fault evictions, eliminates fixed-term ASTs in favour of periodic-tenancy default, caps rent rises to once per year via Section 13, introduces a mandatory Private Rented Sector Database, extends Decent Homes Standard and Awaab's-Law-style hazard timelines to the PRS, and stands up a new Ombudsman scheme.

Staged commencement runs through 2026. Every AST template, renewal letter, rent-review notice, and possession-claim communication produced by AI workflows must be re-papered. Plentific, Fixflo, Re-Leased, and bespoke automations all need vendor updates plus internal re-validation before each commencement wave lands.

NTSELAT Material Information Parts A/B/C is mandatory across Rightmove and Zoopla as of 2023/24 guidance. Part A: tenure, council tax band, price. Part B: physical characteristics. Part C: restrictive covenants, easements, rights of way.

AI-generated property descriptions must clear three gates: CPRs 2008 (no misleading actions or omissions), CAP Code truthfulness, and Material Information completeness. Off-the-shelf US description writers fail Parts A/B/C completeness by default. Areza-built description workflows pre-load the Material Information schema and require human sign-off on every listing before publication.

Money Laundering Regulations 2017 + HMRC supervision + Estate Agents Act 1979. Every estate-agency business and every lettings agent handling rents above EUR 10,000 / month must register with HMRC, complete a written risk assessment, conduct customer due-diligence, run politically-exposed-person and source-of-funds checks, and file SARs.

HMRC published £1.6M+ in AML fines against estate-agency businesses across 2023–24. An AI Voice Agent can capture name, contact, property address, and reason-for-sale; it cannot conduct regulated CDD. Source-of-funds, PEP, and identity-verification routes must hand off to the agency's nominated officer.

Building Safety Act 2022. Any agency with block-management revenue inherits Principal Accountable Person duties, golden-thread documentation requirements, and Building Safety Regulator registration for higher-risk residential buildings (18m or 7+ storeys).

Workflow automations covering safety-case submissions, resident engagement, and statutory-notice timelines must produce auditable evidence trails. ASA + CAP Code govern the content layer: property descriptions must not mislead, and AI-generated marketing copy must be substantiated before publication.

Tenant Fees Act 2019 still binds the lettings side. Only permitted payments may be charged: rent, deposit capped at five weeks' rent, holding deposit, default fees, change-of-tenancy admin. AI workflows that auto-issue lettings fees outside the permitted-payments list expose the agency to enforcement action and tenant refund liability.

A UK seller or landlord asking ChatGPT or Perplexity today for "best estate agent in {borough}" gets one of three answer shapes: a GetAgent / allAgents / NetAnAgent citation (comparison portals that monetise outbound leads back to agencies); a Rightmove or Zoopla "find an agent" tool reference (portal-owned funnels that capture intent before the agency sees it); or a "depends on your priorities, here are some questions to ask" non-answer when the engine cannot find authoritative agency-authored content to cite.

None of these outcomes serve the agency, the buyer, or the seller well.

The structural opportunity is four-layered. Borough × intent pages built programmatically. Comparison content for queries AI engines need a non-vendor source to answer ("Foxtons vs KFH for SW London prime", "Hamptons vs Savills for Surrey country houses", "online vs high-street agents in {borough}").

Compliance reference pages: high-intent, low-volume, citation-magnet content on Renters' Rights Act 2025, Material Information Parts A/B/C, AML source-of-funds expectations. And an off-site evidence layer across allAgents, Feefo, Trustpilot, and Google Reviews where ChatGPT and Perplexity actually pull brand-authority signal.

The PropTech-vendor mirror is identical. A regional MD asking ChatGPT for "best lettings management software for a 12-branch UK independent" needs to receive a citable shortlist that names the vendor, its integration footprint, and its compliance posture. Vendors who have not built that named-entity layer get omitted in favour of the four or five SaaS brands ChatGPT can verify at retrievable depth. Areza's wedge for the vendor side is identical to the agency side, one abstraction up.