In February 2026, Citrini Research published The 2028 Global Intelligence Crisis, a thought experiment imagining the macroeconomic consequences of rapid white-collar job displacement by AI.
We thought we’d try a similar exercise, narrower in scope but closer to home. Several of the Fund’s holdings are dominant online classifieds platforms whose share prices have fallen significantly since mid-2025 on fears of AI disruption. As at April 2026, European classifieds’ share prices are 12-17x next years’ consensus EPS, which implies a dire long-term scenario for companies that continue to grow.
What would the world need to look like for these businesses to be displaced? For simplicity, we focus on the fund’s investments in online property portals such as the UK’s Rightmove, Germany’s Scout24, and Sweden’s Hemnet. However, we believe that similar logic applies across classifieds, including Auto Trader, REA and CAR Group.
What follows is a fictional scenario set in 2030, which we believe is consistent with the current share price level for these businesses, followed by an examination of what would need to be true for it to play out, and an alternative path that we think is more likely.
I. The Year is 2030
AI assistants improved rapidly through the mid-2020s, with accelerating consumer adoption as Google and Apple embedded them in their mobile operating systems. By the end of the decade, they had become the default way most people interacted with the internet. This changed how people shopped, booked travel, and ultimately how they searched for high-value purchases, including property and vehicles.
It started subtly. In 2026, AI assistants began responding to natural language property queries with reasonable accuracy. “Find me a two-bedroom apartment within a 20-minute drive of my office, with a balcony, under £500,000, near a good primary school.” Early results were rough and incomplete, often pulling stale data from cached web pages. They were also text-heavy, which was a poor fit for a category where buyers want to browse photos, maps, and floor plans. But they improved quickly.
In 2027, the major consumer AI platforms had negotiated data-sharing agreements directly with real estate agents, bypassing portals to build their own complete, real-time listing databases. Agents, most of whom are small independent businesses that had long resented rising portal fees, were willing to try anything to break the grip of the classifieds platforms.
With improved supply, consumers noticed. Why visit Rightmove or Scout24 when your assistant understood your preferences, pre-filtered results, and alerted you to new listings?
Traffic to the major portals began declining in the second half of 2027. At first, management teams dismissed this as driven by cyclical weakness. By the second half of 2028, the trend was clear: traffic had fallen 20-30% from the peak. Enquiry volumes directed to agents by Scout24 were down 30%.
Real estate agents responded to the weakening value proposition with their feet. If consumers were finding properties through their AI assistant, why pay the high monthly platform fee? A few brave early movers dropped their subscriptions. When the sky didn’t fall, others followed. By the end of 2028, agent membership had declined 15%, and average revenue per agent was falling for the first time in these companies’ histories as agents downgraded to cheaper packages, and portals slowed price increases.
The doom loop had begun. Fewer listings made the portals less comprehensive. Less comprehensive portals drove more consumers to AI assistants, which made agents even less willing to pay portal fees. The flywheel that had powered these businesses for two decades was now spinning in reverse. By 2029, the AI platforms had introduced their own monetisation. Agents could pay for priority placement, or for “verified listing” badges that signalled quality to consumers. Those fees were initially lower than portal subscriptions, but they would likely climb over time.
The portals did not stand still. They invested heavily in AI features of their own in an effort to retain users. But this was a new cost bucket for businesses that had historically required very little ongoing investment. As direct traffic fell, marketing spend increased to maintain visibility with users of AI. Margins compressed from both directions: revenue declined as agents churned, while operating costs rose as the platforms fought to remain relevant. The 60-70% operating margins that had made these companies investor darlings were a distant memory.
But it is not 2030.
II. What Would Need to Be True
The scenario above is internally coherent. Each step follows logically from the last. It is the kind of narrative that, once imagined, feels almost inevitable. Yet we believe it is highly unlikely.
Each logical leap depends on a specific assumption. For the doom story to play out, all must hold. Like a chain, if any single link breaks, the entire narrative unravels.
Assumption 1: AI platforms build a complete, real-time property database.
In order to attract consumers, AI platforms first need listings. This would likely require them to build direct commercial relationships with tens of thousands of real estate agents, which is costly, slow, and inherently local. Rightmove has over 16,000 agency branches. Scout24 has over 26,000. Most are small, independently owned businesses that have historically found it difficult to coordinate, and have embedded portal data and tools into their workflows. This fragmentation is one of the reasons online classifieds enjoy network effects in the first place. While it may be theoretically possible for AI to “scrape” that content without a relationship, inconsistent page structures, lagged or incomplete data, limited access to rich media, and prohibitions in terms of service likely make this infeasible at scale. In the UK, for example, agents often deliberately withhold information from their listing pages as a strategy to retain leads.
Global AI companies appear to have little appetite for this kind of work. In practice, they are doing the opposite, partnering with the portals on mutually beneficial terms that improve the AI’s output while maintaining the portal’s primacy as the core listing database.
Assumption 2: Consumers change entrenched behaviour when making the most financially consequential decision of their lives.
Property is typically the largest purchase most people will ever make. This is not a casual purchase. Buyers browse, compare, and revisit listings over weeks or months. The portals have strong brands built on exhaustive inventory and high-quality data. As a result, consumers tend to go directly to the platforms they already know and trust, with over 80% of traffic going straight to the major property portals. AI assistants will need to earn this trust over time, against incumbents who already have it.
Trust alone is not enough. Consumers also need a reason to switch. Changing this behaviour is harder than it sounds. Google Search’s most reliable antitrust defence has long been the entirely accurate statement that “competition is a click away.” Yet despite Microsoft spending billions and even offering users cash back to use its competing service, Bing, Google has maintained its dominance for 25 years. Consumer habits are difficult to break when a service is good, trusted, and free.
The doom scenario frames property search as a query problem: a buyer describes what they want, and the AI returns results. In reality, a large share of portal usage is browsing. Buyers often don't know precisely what they want until they see it, and change their initial parameters as they browse. They discover preferences by scrolling through listings, comparing floor plans, toggling between suburbs, and revisiting saved properties over weeks or months. Portals are built for this kind of open-ended, visual, repeated exploration, which is fundamentally different from conversational AI interfaces. And because portals observe consumer behaviour across the full decision journey, that data becomes the foundation of a better product, accumulated through engagement rather than scraped from the web.
This behaviour is both deep and persistent. In Sweden, most of the population visits Hemnet monthly (peaking at almost 40 minutes per person in 2020), with the average user returning more than three times per week. In the UK, Rightmove reports that consumers spent 16.4 billion minutes on the platform in 2024, or roughly 20 minutes per person per month across the entire population.
So far, there is little evidence that this behaviour is changing. ChatGPT was released in 2022, but property portals have variously noted that referral traffic from LLMs remains below 0.5%, with Scout24 noting that this share actually declined in 2025 from peak levels in 2024. Meanwhile, the portals are adding AI features to their own platforms, leveraging proprietary data across the full decision journey. Scout24’s AI feature, HeyImmo, is one such example.
Assumption 3: Agents can defect from property portals without negative consequences.
Even if AI platforms succeed in capturing listings and consumer traffic, it is not a given that agents (and property vendors) would churn en masse. Classifieds platforms are governed by a prisoner's dilemma that makes the defection of individual agents difficult to justify. An agent who cancels Rightmove, for example, doesn't just save a subscription fee. They also disappear from the platform, along with their listings. Vendors may avoid such agents, knowing their most important asset won’t be marketed to all prospective buyers. This dynamic is even more pronounced in markets like Sweden (and Australia), where it is the property seller, not the agent, who pays for the listing. The rational choice for any individual agent or seller is consequently to remain on the platform, regardless of how they feel about the fees, which helps explain why many pay duplicate fees to list across multiple portals.
For defection to work at scale, agents would need to coordinate. Thousands of independent, competing businesses would need to simultaneously abandon the dominant portal and trust that a new channel will deliver equivalent demand. This has been attempted. In 2013, a consortium of UK estate agents launched OnTheMarket specifically to challenge Rightmove's dominance. More than a decade later, Rightmove's market share is largely unchanged.
III. We’ve Seen This Movie Before
The idea that a powerful technology company will disintermediate specialist classifieds platforms is not new. Google launched property listings in the US in Google Maps in 2009, and experimented with similar features in markets such as the UK and Australia, creating significant market noise, only to shut the feature down by early 2011. Brian McClendon, VP of Google Earth and Maps, cited low usage and “the proliferation of excellent property-search tools on real estate websites.”
In the US, CoStar, one of the largest real estate data companies in the world with deep industry relationships, has spent an estimated US$3 billion building Homes.com as a challenger to residential market leader, Zillow. In 2025, it generated approximately US$80 million in annual revenue, compared with Zillow’s US$2.1 billion. Two major hedge funds have publicly demanded CoStar abandon the effort. In the UK, Zoopla and OnTheMarket (owned by CoStar) have failed to meaningfully dent Rightmove’s 80% traffic share.
What these precedents show is that traffic, technology, and capital are not sufficient on their own. What matters is a complete, trusted listing database and a monetisable relationship with the supply side. These are assets built over decades, reinforced by network effects, and no challenger has managed to replicate them.
IV. The More Realistic Risk
There is a more plausible variant of the bear case worth taking seriously. In this scenario, AI platforms do not compete directly with the portals but gradually capture a larger share of referral traffic, potentially reducing the portals share of direct traffic. As they do, the platforms must pay more to maintain their visibility.
This is not hypothetical. Google already refers an estimated 10-20% of digital traffic to the major classifieds sites and captures some share of their marketing budgets. After 25 years of Google dominance, it is possible that the top of the funnel fragments as OpenAI and other consumer AI services gain share from Google and potentially from direct traffic to the platforms.
But fragmentation at the top of the funnel is not all bad for the dominant portal. More smaller traffic referrers have less bargaining power than one large one. And the market leader has the greatest capacity to spend on performance marketing, making it best positioned to maintain its position even at a modestly lower margin.
V. The More Likely Future
It is 2030. AI has changed property search, but not in the way many had feared.
AI assistants have become a core part of property search - helping users refine preferences, broaden discovery, and personalise recommendations. Rather than by the AI platforms, these improvements have been underpinned by proprietary and non-replicable data accumulated over decades by the property portals themselves. With the most complete inventory, richest data, and tailored browsing experience, buyers continued to gravitate to the property portals.
The signs were there in 2026 as major AI companies partnered with the leading portals to improve their own services as they fought for to maximise their share of the top of the funnel. Building and maintaining a complete, real-time property database is expensive and unglamorous work. The AI companies had little incentive to replicate it when they could simply partner for access.
Economically, the model proved resilient. The portals absorbed some incremental costs as they invested in AI, but their core value proposition remained intact. Agents continued to pay for access to demand, and generated improved productivity through AI-enhanced tools.
From August 2025 to March 2026, share prices across the sector fell roughly 40%, despite continued sales and earnings growth. For long-term investors, the question was never whether AI would change these businesses. It did. The question was whether that change would be sufficient to undermine the durability of their profits. Four years later, the evidence suggests it has not.
VI. Final word
We could be wrong. Predictions are inherently uncertain. If AI truly solves the chicken-and-egg problem that has defeated Google, CoStar, and every other challenger, these businesses could be in trouble. We are watching the evidence carefully. But for now, we believe the market’s imagination is running ahead of reality, and that imagination, not disruption, is what is creating the opportunity.