"When I asked ChatGPT about our company, it gave incorrect answers for both our pricing model and primary services"—we have seen a surge in such inquiries recently. Now that conversational AI and autonomous agents are becoming the front door for corporate information alongside search engines, this issue is serious. A pricing table obvious to a human at a glance looks to an AI like merely a decorative series of numbers. Business descriptions baked into images are not read at all. In this manner, unintended descriptions of your company take on a life of their own.
In response to this trend, Google announced the Open Knowledge Format (OKF) on June 13, 2026, an open document specification designed to be handled easily by both AI agents and humans. Behind this lies a paradigm shift: while web pages were historically created for human eyes, going forward they must also be designed for AI consumption. While OKF itself is a new specification, the underlying demand for information architecture that AI can read accurately is an actionable initiative right now, regardless of which specifications trend or fade.
Why readable to humans but unreadable to AI happens
The reasons AI misinterprets website information generally boil down to one of the following:
| Symptom | Common cause |
|---|---|
| Pricing and specifications are quoted incorrectly | Tables converted into images / Pseudo-tables styled only with CSS |
| Business services are misunderstood | Conclusions are buried under decorative copy without stating clearly what the company does |
| Outdated information is provided in responses | Update dates are not structured, preventing AI from determining freshness |
| Content is not cited at all in the first place | Critical pages rely on JavaScript rendering, leaving content empty during retrieval |
The common thread is that visual information and machine-readable information are out of sync. While humans can infer meaning from context and imagery, AI agents rely primarily on marked-up structures and text. Prioritizing aesthetics to the point where essential information becomes invisible to machines is the root problem.
The three layers of being read by AI
Even without waiting for new standards like OKF, you can elevate AI readability using existing technologies. When designing websites in client web development, we structure this across three layers:
- Semantic HTML structure: Use heading hierarchies (h1–h3) for meaning rather than decoration, and represent prices and specifications using real
<table>elements. Place conclusions at the beginning, and always accompany in-image text with alt text. - Structured data: Use Schema.org (
Organization,Product,FAQPage, etc.) to clearly express company details, services, pricing, and update dates in machine-readable form. The fundamentals are detailed in Making Sites Discoverable in AI Search with Structured Data (GH Media). - Entry points for AI and agents: Provide documents intended for AI consumption, such as
llms.txtor OKF, allowing machines to view the site's full picture, key pages, and summaries at a glance.
The key point is that these three layers are not independent, but layered upon one another. Adding structured data on top of sloppy HTML structure will only cause AI to become confused by conflicting signals. Building up from a solid HTML foundation is ultimately the most effective approach.
AI readability as the next step beyond SEO
One thing to note is that AI readability is not a replacement for SEO, but rather a distinct layer extending from SEO. With AI-generated summaries (AI Overviews) appearing at the top of search results, the phenomenon of ranking first yet receiving no clicks is spreading. We addressed responding to this shift in Content Strategy in the AI Overview Era (GH Media).
While traditional SEO focused on optimizing to be found by search engines, going forward it will also incorporate optimizing to be accurately cited by the AI that finds you. The two do not conflict. Semantic HTML and structured data represent an investment that benefits both search engines and agents simultaneously. Criteria for checking whether your site is ready from an AI standpoint are also summarized in Site Renewals to Make Your Website Agent-Ready (GH Media).
Our client web development approach — starting with an "ask AI" audit
When undertaking AI readability improvements in client web development, our first step is not flashy implementation, but taking inventory of the current state. In a project for an actual manufacturing client (company name withheld), we proceeded as follows:
- Ask ChatGPT and Gemini questions like "What does this company do?" and "What are the prices of their main products?" to catalog errors
- Identify the source pages causing incorrect answers and categorize causes into image text, JS rendering, missing structured data, etc.
- Starting with high-impact pricing, service, and company profile pages, proceed with HTML structural fixes and adding structured data
- After corrections, ask the AI the same questions again to verify that the explanations have improved
In this engagement, simply converting pricing tables from images to real tables and adding Product structured data enabled conversational AI to answer prices accurately. No major overhaul was needed; results were achieved merely by "making information visible to machines where it had previously been invisible."
Start by asking AI about your own company
Improving AI readability does not start with adopting a brand-new specification. Begin by asking ChatGPT or Gemini about your company to confirm where things are being misrepresented. The discrepancies you discover will point directly to high-priority areas for improvement. Determining whether incorrect responses stem from image-based text or missing structured data, and then implementing the fixes, is work we support through client web development.
