AI Agents Prefer Structured Websites: Portfolio Impacts for Domain Owners

Two researchers ran 300 shopping tasks across three AI browser agents and found that "agent-ready" sites completed 89% of tasks versus 49% on an identical human-oriented build — with 30% fewer steps and 19–40% lower token costs.

What the study actually tested

Said Elnaffar (independent researcher in Canada) and Farzad Rashidi (Université Paris Cité) built two versions of one small online store with the same products, prices, inventory, and checkout flow. The baseline was a conventional site. The "agent-ready" variant added four things: product data exposed as JSON-LD, semantic labels and stable identifiers on interactive controls, explicit trust evidence like reviews and certifications, and timestamps revealing data freshness.

Three agents — GPT-4.1, Gemini 2.5 Flash, and Grok 4 Fast — then ran five shopping tasks, ten runs per task per model, for 300 runs total. The agent-ready design won across every model and every task. Data extraction and product comparison jumped 77 percentage points; partial outcomes collapsed from 43% to 3% because clean, labeled facts let agents finish the job instead of flubbing the details. Steps dropped 30% and token consumption fell 19–40%, so agent-ready sites are cheaper for AI companies to shop at.

What it means for your portfolio

Here's the translation to our world. The same markup that helps a screen reader — semantic HTML, labeled controls, structured data — is now also what an AI subagent parses first when it spawns to compare your listing against roughly 20 competing sites in parallel. Nielsen's framing is blunt: your website now serves two species of visitors, and the second species can't see your pretty pixels.

That changes the calculus on developed domains in three ways I'd actually feel in cash flow.

First, it's a liquidity signal. A name with clean JSON-LD and accessible markup is faster for an end-buyer to deploy, easier for an agent to evaluate, and easier for me to verify during due diligence. When I'm pricing a developed domain for sale, "agent-ready markup" stops being a feature and becomes a friction reducer for the buyer — something I'd put in the listing copy and bake into my asking.

Second, it's a holding-cost issue in disguise. If you're parking a name on a thin lander waiting for the right buyer, the same hygiene applies. Structured data costs nothing to add, and it makes your inbound inquiries higher quality — because the people (and now the agents) reaching out can actually read what's on the page instead of guessing.

Third, token economics are shifting in a way that matters even if you never run an agent yourself. Nielsen notes that while per-token prices are dropping, agentic models eat so many more tokens that budgets are exploding. Sites cheap to parse will get favored the way Google rewards fast pages. That's not a UX trend — it's a new ranking signal waiting for the SEO tooling to catch up.

What I'd check this week

If I were auditing my portfolio right now: run every developed name through Google's Rich Results Test. If JSON-LD is missing or broken, that's not a design choice — it's leaving conversions on the table for both humans and agents. When I'm negotiating on an inbound offer, I price the markup as part of the asset, the same way I'd price a clean backlink profile. And when I'm buying aged domains with residual traffic, semantic markup quality becomes a screening criterion I'd add next to DR and historical anchors.

The paper, "Designing Agent-Ready Websites for AI Web Agents," is accepted at ICEME 2026, which means this conversation is moving from blog speculation into academic literature. Treat it accordingly.