ROOFHOUND.
Market Intelligence Report

AI Search & Real Estate Infrastructure
Brazil 2026

An analysis of the Brazilian real estate market, the structural shift in AI-driven search, the MCP ecosystem, the positioning of incumbent portals, and the competitive space for AI-native property discovery infrastructure.

Roofhound — Restricted Use
May 2026
Version 1.0
Confidential
Investors & Team
Do not distribute
Table of Contents
00 — Executive Summary

The silent transition
has already begun.

Buyer behavior has changed. Portals and brokerages haven't noticed. This window closes.

Property searches initiated in AI 61.3%

Of buyer-side searches now start in ChatGPT, Perplexity, Claude, or Gemini, not in portals with filter forms.FlyDragon Benchmark 2026 — 8.2M queries, 192 cities

Brokerages invisible to AI 91%

Of agents and brokerages are completely invisible to the tools their buyers use as the primary entry point.FlyDragon 2026 Benchmark Report

AI referral traffic growth (2025) 527%

YoY growth in AI referral traffic. Converts 4–5× more than traditional organic.Starmorph Research 2025

In the Brazilian real estate market, incumbent players haven't yet noticed what is happening with their customer. In 2026, the majority of property searches begin in AI conversational interfaces, not in portals with filters and forms. This is not projected data: it is observed behavior, measured across 8.2 million queries in 192 American cities.

Brazil is approximately 12–18 months behind the United States in this cycle. In the American market, every major portal (Zillow, Redfin, Realtor.com, Homes.com) launched AI search within a window of just 6 months (October 2025 to March 2026). The race has started; it just hasn't arrived here yet.

Agents who began AI optimization in early 2025 receive 5.7× more citations in LLM responses than agents who started the same strategy 12 months later, even with larger budgets.

FlyDragon 2026 AI Visibility Benchmark Report

The Model Context Protocol (MCP), launched by Anthropic in November 2024, has become the universal standard for integration between LLMs and external data sources, adopted by OpenAI, Google DeepMind, and Microsoft in under a year. In December 2025, it was donated to the Linux Foundation, eliminating vendor lock-in risk. With 97 million monthly SDK downloads, this is not a hypothesis: it is established infrastructure.

No Brazilian player offers a solution equivalent to Roofhound's: a dedicated MCP Server for real estate brokerages, combined with semantic data enrichment and AI-optimized discovery pages. The gap is real, verifiable, and temporary.

ROOFHOUND. 00 — Executive Summary · May 2026 · Confidential
01 — Brazilian Real Estate Market

A USD 128.6 billion
market.

Consistent growth, billion-dollar commissions, and a digital distribution layer that has yet to be challenged.

Sector size and growth

Brazil real estate market (2025) $128.6B

CAGR 2.50% — projected USD 160.6B by 2034. Includes sale and rental transactions.IMARC Group, 2026

Residential segment (2025) ~$108B

84% of the transactional market. CAGR 5.52% — residential base growing faster than the total.IMARC Group × Statista, 2026

Total market value (2026) $8.69T

Broad definition: sales + rentals + real estate asset value. Projected USD 9.42T by 2031.Statista Real Estate Outlook, 2026

The standard broker commission in Brazil is 5–6% of the sale price. Applied to the residential transactional segment of approximately USD 108 billion (2025), this represents a commissions market of USD 5.4–6.5 billion per year (before the commercial segment and other asset classes).

The sector posted strong growth in 2024: new developments advanced 18.6% and sales 20.9%. Demand remains pent up, and government housing programs sustain the pace.

Proptech in Brazil: sustained acceleration

Indicator Value Source
Brazil proptech market (2025) USD 866M Grand View Research
Brazil proptech projection to 2034 USD 2.5B CAGR 12.55%
Recent proptech investment (Brazil) BRL 1.2B (~USD 230M) Period 2024–2025
Global VC in proptech (2025) USD 16.7B +67.9% YoY — Propmodo / Bisnow
Global VC in proptech (Jan/2026) USD 1.7B +176% vs Jan/2025 — Bisnow
Annualized growth, AI proptech 42% p.a. vs 24% non-AI — Propmodo, 2025

The global proptech investment environment is the hottest it has been in the past decade. AI-centric proptech companies are growing at nearly twice the rate of non-AI ones. Brazil is Latin America's largest real estate market with active proptech investors: FJ Labs, Monashees, Igah Ventures, and Better Tomorrow Ventures all have sector portfolios, with seed-to-Series-B tickets.

In 2026, Loft earmarked BRL 100 million for technology, focused on AI and automation. This isn't a thesis: it's a budget line.

Loft — Corporate Announcement, 2026
ROOFHOUND. 01 — Real Estate Market · May 2026 · Confidential
02 — The Shift in Property Search

Buyers stopped
using filters.

Property search has migrated from portals with forms to AI conversational assistants. The data is unambiguous.

The central data point — with context caveat

Property searches starting in AI 61.3%

Base: 8.2M queries across 192 American cities. From 17% (2024) to 67% (2026) in 18 months among buyers who use AI as their primary research tool for agents and properties.

Note: U.S. benchmark. No equivalent data is publicly available for Brazil. Expected lag of 12–18 months.FlyDragon 2026 AI Visibility Benchmark Report — newswire.com

Agents invisible to AI 91%

Percentage of real estate agents completely invisible to the AI tools their buyers use first. Real estate ranked last among all monitored sectors in AI visibility, below healthcare, legal, finance, and every other industry.FlyDragon 2026 / 5WPR + Haute Residence Research, Apr/2026

The American portal race: 6 months, entire market

Between October 2025 and March 2026, every major American real estate portal launched or substantially updated its AI search. This wasn't a coordinated strategic decision: it was market pressure. Buyer behavior forced everyone's hand.

Why AI traffic converts higher

Metric Traditional search (portals) AI traffic
Relative conversion rate Baseline (1×) 4.4–5× higher
YoY growth (2025) Stable / declining +527%
Lead quality Browsing user, variable intent User described specific criteria in a conversation
First-mover advantage — 5.7× more citations (vs. those who started 12 months later)

The lead that arrives via AI is not the lead who typed "2-bedroom apartment Moema" into a search box. It's someone who had a conversation with an assistant, described their life situation, received a contextualized recommendation, and arrived at the property already pre-qualified. The 4.4–5× conversion gap reflects exactly that.

AI visibility is not a ranking you buy, but a reputation you build over time through structured data, semantic quality, and consistent presence. Those who enter in 2026 do not start from the same point as those who enter in 2027.

FlyDragon 2026 Benchmark — Roofhound analysis
ROOFHOUND. 02 — The Shift in Search · May 2026 · Confidential
03 — The MCP Protocol

From Anthropic experiment
to industry standard.

The Model Context Protocol went from zero to 97 million monthly downloads in 13 months. This isn't Roofhound's bet — it's the ecosystem's bet.

What MCP is and why it matters for real estate

The Model Context Protocol is an open standard that defines how LLMs and AI agents connect to external data sources, tools, and systems, without each integration needing to be built from scratch. MCP is to AI agents what HTTP is to browsers: a transport protocol that any system can implement.

For brokerages, MCP means that any AI agent (Claude, ChatGPT, Gemini, or future clients) can query a brokerage's inventory in real time, with semantic filters, without needing a portal as an intermediary. The buyer asks the assistant; the assistant queries the MCP Server; the brokerage's properties appear in the response.

Adoption: from 2M to 97M in 13 months

Launch — Nov 2024 2M

Monthly SDK downloads at Anthropic's launchPento.ai / DigitalApplied

Post-OpenAI adoption — Apr 2025 22M

Monthly SDK downloads after OpenAI's adoptionPento.ai / DigitalApplied

Established standard — Dec 2025 97M

Monthly SDK downloads. 10,000+ servers in active productionAnthropic — AAIF announcement, Dec 2025

The protocol is not a bet on Anthropic. It is open infrastructure maintained by the Linux Foundation, with Anthropic, OpenAI, and Block as co-founders. If Claude is displaced, MCP continues. If ChatGPT dominates, MCP is there too.

Agentic AI Foundation — Linux Foundation, Dec 2025
ROOFHOUND. 03 — The MCP Protocol · May 2026 · Confidential
04 — Brazilian Real Estate Portals

Structural dominance.
Structural conflict of interest.

Portals control Brazilian real estate distribution — and have a concrete financial incentive to not fully solve the AI visibility problem.

Grupo OLX market position

Combined monthly users 34M+

Zap Imóveis + VivaReal + OLX ImóveisOLX Group

Annual revenue BRL 1B+

EBITDA 28%. 50% of group revenue comes from the real estate sectorAIM Group, Dec 2024

Municipalities covered 92%

Of Brazilian territory. 19.2M active listings. 45,000+ professional partnersOLX Group

Player Platforms Monthly reach AI position
Grupo OLX Zap Imóveis, VivaReal, OLX Imóveis ~34M users Basic internal AI feature ("Ideal Property"). No MCP. No inventory exposure to external LLMs.
QuintoAndar QuintoAndar.com.br, ImovelWeb, WiMoveis ~23M combined visits Built the first ChatGPT app, that works quite poorly. Focus on internal portal digital transaction only.

The conflict of interest — the core argument

The revenue model of Zap Imóveis, VivaReal, and OLX Imóveis depends, structurally, on brokerages having to pay to be seen. Brokerages pay thousands of dollars a month to be seen, and portals are the gatekeepers of that visibility.

A product that makes brokerages directly discoverable by LLMs (without going through the portal) is a direct threat to portals' revenue model. They have a concrete financial incentive to not build this solution in any complete form.

Portals can launch "AI," and they will. But the AI they build is closed: it operates inside Zap, recommends Zap properties, for Zap users. MCP is open: it operates inside ChatGPT, Claude, Perplexity, any agent the buyer happens to be using. That difference is the moat.

Roofhound Analysis — May 2026

Grupo OLX's announced BRL 440 million investment plan for 2026 (NeoFeed) may include more sophisticated AI features, including exploration of open protocols like MCP. The structural conflict of interest is real, but not immutable: Zap could decide to sacrifice some brokerage dependency in exchange for a position in the AI ecosystem. Probability assessed as low, but not negligible.

In 2024, Zap went through a public trust crisis with partners, described as "back-stabbing" and "misaligned communication" by the trade press (Online Marketplaces). The relationship between portal and brokerage in Brazil is transactional, and fragile.

ROOFHOUND. 04 — Brazilian Portals · May 2026 · Confidential
05 — Competitive Landscape

The Brazilian gap
is confirmed.

There are GEO agencies, analytics tools, and one American MCP player in real estate. None operate in Brazil with Roofhound's value proposition.

Identified players map

Player Type Product Analysis
First Page SageUSA Agency Pioneer GEO for real estate. Proprietary AI optimization framework. Marketing service, not product. Optimizes individual agents, not infrastructure for brokerages. Does not operate in Brazil.
GenevateUSA Agency GEO + strategic PR for real estate and other sectors. Reputation and AI positioning service. Does not resolve structured data. Does not operate in Brazil.
ProfoundUSA Analytics Visibility monitoring across 10+ AI engines (ChatGPT, Claude, Perplexity, Gemini…). Pure analytics — measures where you appear, doesn't make you appear. Enterprise pricing. Generic, not specialized in real estate.
Semrush AIOGlobal Analytics AI search tracking module integrated into Semrush. Add-on to existing SEO platform. Analytics, not infrastructure. Generic, no real estate specialization.
Homesage.aiUSA Product Investment intelligence via MCP. 140M+ American residential properties for AI agents. Focus on institutional investors, not traditional brokerages. US only. Market data, not consumer-facing inventory exposure. Different use case from Roofhound.
RoofhoundBrazil Infra MCP Server + semantic enrichment + AI-optimized pages. For Brazilian brokerages. Only identified player building MCP infrastructure for Brazilian brokerages. Product, not agency. Infrastructure, not analytics. Brazil, not USA.

Reading the landscape

The global GEO/AEO market for real estate is fragmented between service agencies (First Page Sage, Genevate) and analytics tools (Profound, Semrush AIO). None of these players solve the infrastructure problem: making a specific brokerage's inventory directly queryable by an LLM in real time.

Homesage.ai is the closest analog: it uses MCP to expose property data. But the model is different: aggregated market data for investors, not brokerage inventory for end buyers. Operates only in the US, with over 140M properties mapped.

In Brazil: zero players identified with an equivalent solution. Portals (Zap, VivaReal) have basic, closed AI features: recommendations within the portal, not inventory exposure to external LLMs.

The absence of direct competition in Brazil is not a sign of a small market: it's a sign of timing. The American market, 12–18 months ahead, already has multiple players. In Brazil, whoever arrives first still defines the territory.

Roofhound Analysis — May 2026
ROOFHOUND. 05 — Competitive Landscape · May 2026 · Confidential
06 — TAM / SAM / SOM

Addressable market
with explicit assumptions.

All estimates below are built bottom-up with stated assumptions. These are not audited projections.

TAM Total Addressable Market — Brazil, current model
~BRL 400M/yr

Estimate. Proxy: 19.2M Zap/OLX listings as eligible properties — brokerages will integrate their full portfolio and pay per listing; deduplication is handled by Roofhound in backend. Secondary reference: Statista cites ~54M residential units; assuming 84% residential share and conservative 15% vacancy yields ~9.7M unique properties — but active listings are the operational base. Expanded TAM with LatAm and adjacent services: BRL 1B+/yr.

Eligible properties (proxy: Zap/OLX listings)19.2M
MCP per property/monthBRL 1.55
Annual MCP revenue — 100% captureBRL 357.1M/yr
Enrichment — 30% annual renewalBRL 39.5M/yr
TAM Brazil (current model)~BRL 400M/yr
Expanded TAM (LatAm + adjacencies)BRL 1B+/yr
SAM Serviceable Addressable Market — Years 1 to 3
~BRL 42M

Estimate. There are approximately 70,000 brokerages in Brazil. The SAM (Years 1–3) are the 3,000 operating with inventories between 350 and 5,000 properties, open to technology. Assumption: 3,000 eligible brokerages × 550 average properties = 1,650,000 properties in SAM.

Eligible brokerages (estimate)~3,000
Properties per brokerage — SAM average550
Total properties in SAM1,650,000
Annual MCP revenueBRL 30.7M/yr
Enrichment revenue — initial baseBRL 11.3M
Total SAM~BRL 42M
SOM Serviceable Obtainable Market — Year 1 (pilot)
~BRL 1.1M

Pilot program. 100 brokerages. Target profile: brokerage owners with an average of 550 active properties. 80% discount on initial enrichment for the first 100 brokerages.

Brokerages in pilot100
Properties per brokerage — pilot profile550
MCP revenue — 100 × 550 × BRL 1.55 × 12BRL 1,023,000/yr
Pilot enrichment — 80% discountBRL 75,350 (one-time)
SOM Year 1~BRL 1.1M

The unit economics per brokerage are clear: 550 properties × BRL 1.55 = BRL 852.50/month recurring. With full-price initial enrichment (BRL 6.85/property): BRL 3,767.50 one-time. CAC tends to be low via inbound and pilot network effects.

Calculation based on Roofhound pricing model — mid-sized brokerage profile
ROOFHOUND. 06 — TAM / SAM / SOM · May 2026 · Confidential
07 — Implications & Decision

What the data
determines.

Five conclusions that follow directly from the evidence. Separated from operational assumptions.

ROOFHOUND. 07 — Implications · May 2026 · Confidential
08 — Risks & Caveats

What might
be wrong.

Contrary evidence, data limitations, and downside scenarios. Included for analytical completeness — not as thesis disqualifiers.

ROOFHOUND. 08 — Risks & Caveats · May 2026 · Confidential
09 — Sources

References
and attributions.

All sources verified and accessed in May 2026. Data from 2025–2026 unless otherwise noted.

Brazilian Real Estate Market

AI in Property Search

Model Context Protocol (MCP)

Real Estate Portals and Competitive Landscape

ROOFHOUND. Market Intelligence Report · Brazil 2026 · Restricted Use · Confidential