Your website is ready for AI when an assistant can parse who you are, quote what you claim, and complete a purchase without a human decoding anything. This checklist turns that into 24 concrete checks, grouped under the eight dimensions the Legible Readiness Index scores. Each check is pass/fail on purpose — work top to bottom, the highest-weight dimensions come first.
1. Structured data
- Ship Organization JSON-LD on your homepage. It's the machine-readable answer to "who is this?" — the question every model asks first.
- Add Product or Service schema with price and availability to every offer page. Agents compare structured fields, not prose.
- Add FAQPage schema to pages that answer real questions. Q&A pairs are the most liftable format an answer engine sees.
- Validate all of it and fix every error. Broken JSON-LD is worse than none; a parser that fails once may skip the block entirely.
2. Entity & identity
- Write one canonical brand description and use it verbatim everywhere. Models build consensus from repetition; five variants read as five uncertainties.
- Publish a
sameAsgraph linking your official profiles. LinkedIn, Wikidata, and social profiles tied together let models confirm you're one entity. - Ground the brand in Wikidata (and Wikipedia where you qualify). Third-party grounding is what turns "a company claims" into "a known entity."
3. Citability
- Restructure key pages to one idea per heading. Models extract sections, not pages; a heading that matches a question gets lifted.
- Convert marketing paragraphs into short, data-bearing sentences. "Cuts onboarding from 14 days to 3" is quotable; "world-class solutions" is filtered out.
- Attribute every statistic to a named source. Unsourced numbers lower model trust in the whole page.
4. Commerce readiness
- Put final price, shipping cost, delivery window, and return terms on the product page in machine-readable form. nShift's research on agent legibility is blunt: when delivery and return terms are unclear, the agent skips the offer without a human ever seeing it.
- Keep stock status accurate and in schema. An agent that gets burned by a phantom "in stock" deprioritizes the whole domain.
- Move policies out of PDFs and JavaScript-only FAQ widgets. If the terms only exist inside a PDF or a rendered widget, they don't exist to most agents.
- Give agents a path to transact — or at least to a clean handoff. "Contact sales" with no structured next step is where agent journeys die.
5. Machine access
- Serve your core content as server-rendered HTML. Content that appears only after JavaScript runs never reaches many crawlers.
- Allow AI search and fetcher bots in robots.txt. Blocking OAI-SearchBot or PerplexityBot removes you from AI answers outright.
- Publish an llms.txt file. It's the machine-readable front door: what you are, where to look.
- Offer a lean, low-token version of key pages. When Cloudflare shipped markdown conversion for agent requests, its example page dropped from 16,180 tokens as HTML to 3,150 as markdown — roughly 80% less for the agent to wade through.
- Keep latency low and your sitemap current. Slow, unmapped sites get sampled, not read.
6. Multilingual
- Add
hreflangfor every locale you sell in. AI answers differ sharply by language; models need the map between versions. - Check parity: your schema and key claims exist in every locale, not just English. A German buyer's assistant reads your German pages.
7. Verification
- Back every claim with evidence a machine can follow. A linked source or named study; unsupported superlatives read as noise.
- Make the publisher unambiguous. Clear authorship, contact details, and legal identity separate you from the content farms models learn to discount.
8. Freshness
- Emit accurate
dateModifiedon every page. Models prefer dated, maintained content and discount the stale. - Audit your top-cited pages quarterly for dead facts. Old prices and discontinued products actively damage trust once an agent catches one.
What this checklist can't cover
Passing all 24 checks makes you legible — it doesn't make you complete. For most companies, roughly 20% of what makes a product the right choice lives in structured data; the rest is tribal knowledge sitting in your best salesperson's head, and no validator will flag its absence.
Rather than self-scoring 24 boxes, you can measure them in one pass: run a free Legible report and get the failing checks ranked by how much score they're costing you.