# Our English website looks fine. Can AI actually read our supplier information? | Winin



- Canonical: https://winin.ai/en/answers/export-website-ai-readability-diagnosis/

- Published: 2026-10-06

- Updated: 2026-10-06



[← Content & Evidence](/en/blog/topics/content-evidence/)
Answers / informational
Our English website looks fine. Can AI actually read our supplier information?
On this page
[Quick answer](#quick-answer)
[Details](#details)
[Common questions](#common-questions)
[Related links](#related-links)
Quick answer
Check the text a public page actually returns, then investigate access barriers and conflicting versions.
Specifications visible in a browser may depend on images, scripts or login. Appearance alone does not establish machine readability. Schema, Markdown and llms.txt can supplement a page, but should agree with its visible content; crawler visits, citations and supplier recommendations require separate checks.
Details
Start with a page that should answer a real buying question
Choose an English purchasing question and list the facts needed to answer it: product identity, operating conditions, specification units or service boundaries. Locate the public page responsible for those facts, then check whether it provides readable text and a current version. This is a troubleshooting method, not a claim that any model will adopt a particular format.
Layer
What to check
Evidence to retain
Access
Direct access to the article rather than login, a challenge page or an error
URL, test time and actual response
Content
Key English facts in the initial response and visible page
Text locations and content that is absent or script-dependent
Format
Specifications available only in an image or attachment
Comparison with the page's readable text
Version
Agreement across HTML, Markdown, structured markup and languages
Version URLs, current wording and dates
Discovery path
Normal internal links pointing to the page
Entry pages and link destinations
Make industrial specifications easy to verify
Keep images and PDFs where useful, while placing the core specifications, units and application conditions needed for the buying question in the page text. Link to the full attachment. This does not mean AI cannot read PDFs; it removes reliance on a single representation that may be difficult to check.
Each specification should identify the relevant product or model. Mixing values from different models, omitting units or retaining outdated English descriptions can produce ambiguity even when access works. Have a technical owner confirm the current public wording before changing the page.
Check schema and content maps against the page
Compare structured markup with visible product names, specifications and conditions. If a Markdown version exists, check it too: an updated page and an old parallel copy can disagree. A content map should point to existing canonical pages. Listing a page in the map is not evidence that it was cited.
For an access barrier, have the web owner inspect the relevant crawl policies and protection settings. Identify the failing layer before making a scoped change. Platform configuration should follow the current official rules; this page does not provide crawler rosters or allow rules that may change.
Hand readability findings to the next check
A readable page still needs separate observations of indexing, citations, supplier inclusion and factual accuracy. Retain test records and changed versions, then observe visible citations and facts on fixed buyer questions. See the
[structured-content explanation](/en/answers/structured-data-llms-txt/)
and
[discovery layers in the topic guide](/en/guides/topic-authority/)
.
Common questions
Q: Why can buyers read our English site while AI still misses supplier facts?
A: Check whether the returned text identifies the product and its conditions, then investigate access barriers and old versions. A successful browser display does not establish that the model received the relevant facts.
Q: Should manufacturers put specifications in HTML as well as PDFs and images?
A: Put the facts needed for core buying questions into verifiable page text, with units and conditions, and keep links to complete attachments. Both should describe the same valid version; duplicating an entire catalogue is not the objective.
Q: Why has adding schema and llms.txt not produced AI citations?
A: Check readable content, consistency and valid destinations first. Markup and maps do not guarantee adoption or citation. Continue observing the answers and visible sources for the relevant buying questions.
Q: Does an AI crawler visit prove our export GEO is working?
A: It provides an observation of a crawl request. Indexing, citations, shortlist inclusion and accurate descriptions are separate results and cannot be inferred from the visit alone.
Related links
[Writing public B2B information](/en/answers/b2b-geo/)
[Winin's published stage definitions](/en/facts/stages-reach-shortlist-recommended/)
Keep reading
[Content & Evidence](/en/blog/topics/content-evidence/)
[How to build brand fact pages: from conflicting information to verifiable facts](/en/guides/fact-pages/)
[What is AI Recommendation Intelligence?](/en/answers/ai-recommendation-intelligence/)
[What is fact governance for AI answers?](/en/answers/fact-governance-ai-answers/)
FROM READING TO ACTION
Start with website readiness. Find what you can improve.
[Check my website](/en/free-site-audit/)