How Claude reads and recommends a site

Claude can read your page and still get your business wrong, not because it is careless, but because the page never said the thing plainly enough for a machine to lift. That is a readability problem, and it is yours to fix.

Claude reads the content, not the chrome

When Claude answers from the live web, it fetches pages and reads the text it can parse. It is not weighing your visual polish. It is working from what the page actually states in a form a machine can read.

A visitor and Claude are different readers. The visitor scrolls, infers, and forgives a slow build. Claude works from the content it can extract and weights what it reads.

So the gap is concrete. A page can feel complete to a person and read as an ambiguous document to the engine.

How it reads the web and decides what to surface

Claude can answer from what it learned in training, or it can fetch live pages and answer from those. When it fetches, the rough shape matches what most answer engines do, described in how AI engines decide what to recommend.

It gathers candidate pages, reads them, and composes an answer grounded in the ones it could parse into clear statements. This is retrieval feeding generation, not memory alone.

We will not pretend to know Anthropic's exact selection logic. It is proprietary and it evolves. What does not change is that a page Claude cannot read cleanly is a page it cannot use cleanly.

What a page needs to be legible to it

To be usable by Claude, a page has to make its meaning machine-readable, legible without a human reading alongside.

• A clear heading that names what the page is about.
• A direct answer in plain text, high on the page, not stranded far below the fold.
• Structured data that labels what the page is, what it offers, and what it costs.
• Clean, crawlable HTML, so the content is reachable rather than hidden behind script.

Old way: write for the visitor and hope a machine infers the rest. New way: write for the visitor and leave a machine-readable layer so the engine reads facts instead of guessing.

This does not compel a recommendation. It makes you legible enough to be read accurately, which is the only part within your control.

The damaging admission

Legible is not recommended. We measure whether Claude and the other engines can read and name you, and we report it honestly, but we will never promise Claude will pick you.

The model is probabilistic. It can read two clear pages and favor one for reasons no outside tool can see or steer. Anyone guaranteeing a Claude recommendation is selling something we will not.

Readiness is the floor. If Claude skips a page it cannot parse, you can fix that. If it reads you cleanly and recommends a competitor anyway, at least you were eligible.

The automated fix is WordPress-only. On Shopify, Wix, Webflow, or a headless setup, the scan still diagnoses the gaps, but you apply the changes yourself.

Why a clean read does not stay clean

Two things move underneath you. You keep publishing, and new pages rarely get the structure your fixed pages have, so your average readability drifts down as you grow. And Claude itself changes: new models and new ways of pulling and reading context, so the same page can land differently from one release to the next.

That is why readiness is a posture, not a setting you save once. The page that read well last quarter can read poorly now with nothing changed on your end.

It matters more as Claude shows up inside the tools where people already work, answering in place rather than sending them out to look. If the answer happens there, being readable when it is composed is what keeps you in it.

How to check whether Claude can read your page

Pick the page you would most want Claude to quote and find out whether it can read it today.

Run a free scan on the URL to see whether Claude and the other engines can parse it and which structural pieces are missing, or read how Citedon makes a site agent-ready and keeps it that way as the engines change.