AI Is Eating Real Estate's Attribution Model, and Nobody Clicked a Link

Any performance marketing lead sitting on a seven-figure paid-media budget in real estate, mortgage, or adjacent transactional verticals has a decision to make in the next quarter: whether to keep trusting attribution reports that structurally cannot see the channel now shaping buyer intent. A small usability study out of San Diego just put numbers on the problem. The numbers are ugly, and they generalize far beyond real estate.

What Happened

On September 9, 2026, Targence, an AI Perception Intelligence company headquartered in San Diego, released findings from a home-selling usability study, as EIN News reported. The methodology was straightforward: video-recorded, task-based sessions run through Userbrain, with participants beginning a home-selling scenario using an AI assistant and thinking aloud as they worked.

The headline finding is a single sentence. Not one participant clicked a citation link inside an AI answer to check a source. Some scanned the response, some found the links hard to notice, some simply felt no need to look further once the AI had spoken. The behavior held across the sample.

The influence started well before any specific business was named. AI was already steering the decision path: whether to sell through an agent, take a cash offer, or fix up the property first. It shaped the comparison criteria (ratings, sales history, local experience) and the questions sellers planned to bring to providers (net proceeds, fees, process). By the time a brand name entered the conversation, the frame was set.

What sellers did after the AI answer is where the attribution model collapses. Participants searched Google to locate an agent AI had suggested. They verified reviews of businesses AI had shortlisted. They opened recommended profiles. They asked AI for phone numbers to call. One participant, who used AI to narrow local agents and pull their numbers before planning comparison calls, said flatly: "That was the only way I found them." Sellers described starting with AI first and using Google purely for validation. Asaf Ashkenazi, Targence's co-founder and CEO, framed the business risk as misallocated budget rather than lost credit, arguing leaders read clean numbers, conclude AI is immaterial, and keep funding channels the report can actually see.

Technical Anatomy

To understand why this matters for anyone running paid acquisition, you have to look at how the measurement stack was built. Attribution platforms, whether you're wired into the Google Ads API or piping conversions through the Meta Marketing API, all share one architectural assumption: an identifiable session, initiated by a click, that can be stitched to a downstream conversion event. Referrer headers, UTM parameters, click IDs, view-through pixels. The whole pipeline is a click-graph.

AI assistants break that graph at the root. When a language model synthesizes an answer, the citation link is decorative unless the user clicks it, and this study says the click rate is functionally zero. The user takes the recommendation, opens a new tab, types the agent's name into Google, and lands on the site via what looks like branded organic search or direct traffic. The AI session leaves no fingerprint in any downstream analytics system. There is no referrer. There is no UTM. There is not even a hostname the vendor could add to a referral exclusion list.

Worse, the branded-search visit that follows will be credited to whichever channel the last-click model favors. If the seller then clicks a Google Ads brand-defense keyword, paid search gets the conversion. If they type the URL directly, direct traffic gets it. The AI recommendation, which actually did the work of selecting the vendor, is invisible.

This is a structurally different problem from the cookie-deprecation story the industry has spent three years on. Privacy Sandbox and the shift to aggregated attribution reporting are about degraded fidelity within a still-recognizable click model. AI-mediated discovery isn't degraded fidelity. It's a different funnel entirely, one that begins in a conversational surface no analytics vendor has instrumentation for. There is no equivalent of ads.txt or an IAB spec that covers "the model recommended us." Not yet.

Who Gets Burned

Real estate is the vertical named in the study, but the pattern will replicate anywhere a high-consideration purchase involves a shortlist of local service providers. Mortgage brokers. Insurance agents. Wealth managers. Home services. Legal referrals. Anywhere a buyer used to open Google, type a category term, scan the map pack, and read reviews, they will increasingly start with an AI assistant and only use Google to verify what the AI said.

The teams most exposed are the ones with the cleanest attribution dashboards and the most disciplined ROAS reporting. That sounds counterintuitive, but the logic is direct: the more religiously a CMO optimizes toward measured channels, the faster they will over-fund brand-defense keywords, retargeting, and paid social that are now capturing conversions AI originated. The measured channels will look like they are working harder. Blended CAC will drift up. Nobody will know why, because the report will keep showing green.

The CFO at any consumer marketplace or lead-generation business should be asking the head of growth this week a specific question: what percentage of our branded organic and direct traffic in the last two quarters can we affirmatively attribute to a non-AI first touch, and how would we even test that? If the answer is a shrug, the marketing budget is being allocated on faith.

There is also a hiring implication. The analyst profile that dominated performance marketing for the last decade, someone fluent in GA4, MMM, and click-path analysis, is not the profile equipped to measure AI-mediated discovery. The emerging skill set looks more like SEO circa 2010 crossed with prompt engineering: manually probing how leading models describe your brand, cataloging the evidence they cite, and closing the gaps. Very few candidates have that resume.

Playbook for Performance Marketing

The defensive moves are unglamorous but urgent. First, run the audit Targence recommends: query the major AI assistants with the exact prompts your customers use, log what they say about your brand, your competitors, and the criteria they surface. Do this weekly. Treat it as a monitoring surface, not a one-time exercise.

Second, restructure your site to serve the AI-arrival visitor. The Targence sessions showed one participant struggling to find a way to speak with a human before signing up. If AI is sending pre-qualified sellers to your site expecting to verify credentials and take a next step, the site needs to lead with proof (transaction history, licenses, reviews, named humans) and offer an obvious contact path. The category-introduction homepage is now the wrong page for half your traffic.

Third, stop trusting last-click for budget decisions in categories where AI discovery is plausibly material. Introduce holdout tests. Kill a brand-defense campaign in a controlled market for four weeks and measure the true incrementality. If direct and organic hold up, you were paying to intercept AI-originated demand.

Fourth, invest in the evidence AI models pull from. Third-party review sites, structured data, authoritative press coverage, verifiable transaction records. The models are consensus machines. Whichever brand's story is easiest to corroborate becomes the default answer.

Key Takeaways

• Targence's September 2026 usability study found zero participants clicked AI citation links, meaning AI-driven decisions leave no trace in click-based attribution.
• AI shapes the decision path (route, criteria, questions) before any brand is named, then hands the credit to Google and direct traffic downstream.
• Teams with the cleanest ROAS dashboards are most at risk of misallocating budget toward channels that only appear to be converting.
• Sites need to serve two arrival states: category-introduction visitors and AI-briefed visitors who arrive already knowing who they want to verify.
• The right hire for 2026 growth teams is closer to an SEO analyst with prompt-engineering fluency than a GA4 specialist.