How to Make Your Website Legible to AI Agents Before Your Competitors Do

As of August 2026, three unrelated Hacker News threads landed within 48 hours of each other and pointed at the same shift. A proposal called WebMCP — a way for websites to expose actions directly to AI agents — pulled 53 points and 50 comments, an unusually high comment-to-point ratio that means the room is still arguing about how this should work, not whether it's coming. A Show HN for a Raspberry Pi running a 35B-parameter Qwen model as a local car assistant hit 144 points and 46 comments. And a new "web search API for AI agents," Keenable, launched with a self-reported 100-billion-page index and sub-250ms p95 latency, built specifically because general search wasn't built for machine consumers. None of these are AlphaForge news. All three are the same signal: AI agents are becoming a distinct category of visitor to your business, and most business websites are still built for only two audiences — humans and Google's crawler.

Why this matters

When a customer asks ChatGPT, Claude, or Perplexity for the "best HVAC company near me" or "who can build me a website this month," the model doesn't browse your site the way a person does. It pulls from indexes like the one Keenable just built, from structured data embedded in your pages, and increasingly — per the WebMCP thread — from actions your site explicitly exposes for agents to call. If your hours, pricing, service area, or booking flow only exist as text baked into a hero image or rendered client-side after three seconds of JavaScript, an agent may never see them. You don't lose the sale by being outranked. You lose it by being invisible to the thing doing the asking.

A related thread in the same 48-hour window reinforces the stakes: a principal engineer's "Is AI slowing you down?" post (16 points, 12 comments) found that agents perform well only when given precise, low-level specs — vague or ambiguous input degrades results fast. The same is true when an AI answer engine tries to describe your business. Ambiguous, inconsistent, or missing facts about what you do and where you operate don't just confuse humans skimming your site — they cause the agent summarizing you to guess, and guesses rarely favor you.

None of this requires betting on any single protocol. The WebMCP debate is exactly that — a debate, with 50 comments disagreeing on implementation before anything is finalized. What's not in debate is the underlying need: agents need clean, structured, current facts to represent a business correctly, the same way the Raspberry Pi car assistant needed a full manufacturer manual and live vehicle data before it could answer a question about tire pressure instead of guessing. A local business doesn't need to build an MCP server this month. It needs its basic facts to be true, consistent, and readable by something that isn't a person.

The playbook

1. Audit what's machine-extractable, not just human-visible. Open your homepage, services page, and pricing page with JavaScript disabled. If your hours, address, pricing, or service list disappear, an agent crawling your site sees the same blank page. This is the single most common gap AlphaForge finds in local business sites.
2. Add structured data for the facts that matter most. Schema.org markup (LocalBusiness, Service, FAQPage, Offer) turns prose into fields a machine can parse without interpretation. This is the low-tech version of what WebMCP is trying to standardize at the protocol level — you don't need to wait for the spec to fight to reach agreement to get the benefit today.
3. Reconcile your business facts across every surface. Name, address, phone, hours, and service area should read identically on your site, Google Business Profile, and any directory listing. Agents cross-reference; a mismatch reads as unreliable data and gets discounted or dropped, the same way the low-level-spec problem plays out for coding agents — inconsistent input produces unreliable output.
4. Test how the major answer engines currently describe you. Ask ChatGPT, Claude, and Perplexity your own category query — "best [your category] in [your city]" — and read what comes back. If you're missing, wrong, or describing a service you dropped two years ago, that's your current baseline, not a hypothetical risk.
5. Expose your core actions in plain, unambiguous text. Booking, quote requests, service areas, and pricing tiers should exist as clear, crawlable text somewhere on your site — not only inside a chat widget or a form that requires five clicks to reveal what it does. Treat this the way the Keenable team treated general search: purpose-built for the machine reader, not repurposed from what worked for humans in 2019.
6. Re-check monthly, not once. The WebMCP debate alone tells you this space is still being negotiated in public — 50 comments arguing implementation details in two days. Answer engines update how they weight and cite sources on a similar cadence. A one-time audit goes stale within a quarter.

Step four is worth doing before anything else on this list, because it tells you whether you're starting from zero or from a bad baseline. Owners are frequently surprised either way — some find they're recommended accurately and just need to reinforce it, others find an agent confidently describing a service line they discontinued two years ago, sourced from a stale directory listing nobody remembered existed. You can't fix representation you haven't measured.

Common pitfalls

• Treating this as solved by existing SEO work — machine legibility for AI agents and traditional search ranking overlap but are not the same discipline.
• JavaScript-only rendering of critical facts (pricing, hours, service area) that leaves agents with nothing to read.
• Inconsistent NAP (name, address, phone) data across your site, directories, and Google Business Profile.
• No structured data at all, forcing every agent to infer facts from unstructured paragraphs — the exact failure mode the "is AI slowing you down" thread describes for under-specified coding tasks.
• Assuming a single fix ships permanently correct results, instead of monitoring how you're actually being represented.

Quick self-check

• Can you view your hours, pricing, and service area with JavaScript disabled?
• Does your site carry Schema.org markup for your business type and services?
• Do your name, address, and phone number match exactly across your site, Google Business Profile, and top directories?
• Have you asked ChatGPT, Claude, and Perplexity how they currently describe your business, this month?
• Is there a plain-text page listing what you do, where you serve, and how to book — independent of any chat widget or image?

This is the same underlying problem AlphaForge has been tracking since our piece on why a shared agent brain is usually the real AI bottleneck — agents, whether they're reading your website or running your internal workflows, only perform as well as the facts you hand them. A business that gets its structured facts right once and keeps them current has a real, compounding advantage over one that treats this as a website redesign line item.

Most owners don't know what ChatGPT, Claude, or Perplexity currently say about their business — or whether they show up at all. AlphaForge runs a free 24-hour audit that checks exactly that. Get your free AI Visibility Report.