Your website now has two visitors

Who reads my website before my buyer does?

Increasingly, an AI assistant does. A purchase manager who once typed "ball valve manufacturers in Gujarat" into Google and opened ten tabs now asks ChatGPT, Gemini or Google AI Mode the same question and reads one answer. The assistant has read your pages, or tried to, and the buyer comes to your site only if that answer named you.

So every factory website has two visitors. The first is software with no patience for pictures, downloads or clever menus. The second is the human who signs the purchase order, and who may never arrive if the first visitor found nothing worth repeating.

What is an AI agent, and how is it different from a chatbot?

An agent goes a step further than an assistant: it opens websites and does things on a person's behalf. Google's own guide to generative AI features in Search defines AI agents as "autonomous systems that can perform tasks on behalf of people", and notes that browser agents may work from screenshots, the page code and the accessibility tree, the structured list of headings, links, buttons and labels that screen readers also use.

In practice that means a buyer can ask an agent to shortlist three suppliers of 50 mm ball valves, compare their pressure ratings and fill in the enquiry forms. Whether your company makes that shortlist depends on whether the agent can find the rating, understand the form and press the right button.

Google's Lighthouse tool now tests for this. Lighthouse 13.3, released on 7 May 2026, moved its Agentic Browsing category, until then experimental, into the default audits. It checks agent accessibility (for example, whether every interactive element has a programmatic name), layout stability, llms.txt and WebMCP integration. It gives no overall score yet, because the standards are still emerging, and it is not a search ranking factor. It does show what Google's own tooling has started to check.

What can an assistant read on a typical factory website?

It reads text in the page itself well, and struggles with almost everything else. On many factory websites, the most useful facts sit in exactly the places software reads worst.

The pattern is simple. If a fact exists only as pixels, inside a file, or behind a form, assume the first visitor cannot use it. Certificates are a common case, covered in putting certificates on the page as text, and PDF catalogues are covered in why ChatGPT struggles with your PDF catalogue.

Why do JavaScript menus and tabs cause trouble?

Because not every reader runs the scripts that build them. Google says in its guide that it can process content in JavaScript as long as it is not blocked. Other AI crawlers are less capable: a December 2024 study by Vercel found that none of the major AI crawlers it tested, including OpenAI's GPTBot, Anthropic's ClaudeBot and PerplexityBot, rendered JavaScript. They fetched the script files but did not run them.

That study is now nearly two years old and crawlers change, but the safe rule has not: anything you want an assistant to read should be in the HTML the server sends, before any script runs. A product range hidden in a tabbed widget, or a "Products" menu that is built by a script, can leave a reader with a home page and no route to your product pages.

How can I check my own site in ten minutes?

You do not need tools beyond a browser. Run these checks on your home page and your two most important product pages.

1. View the source. Press Ctrl+U and search (Ctrl+F) for your production capacity, a pressure rating, or the standard you make to. If it is not there as text, it is in an image or a script.
2. Copy from the PDF. Open your catalogue, try to select and copy a line of the spec table. If nothing highlights, the PDF is a scan and has no text to read.
3. Check the menu links. In the source, search for the address of one product page. If the menu links are missing, crawlers may not find those pages from the home page.
4. Read the form labels. Each field should have a visible label ("Quantity required", "Delivery location"), not just grey placeholder text that vanishes when you type.
5. Ask an assistant. Paste a product page address into ChatGPT or Gemini and ask what sizes, standards and capacity the page states. Compare its answer with what you know is true.

If the assistant answers wrongly or vaguely, the problem is usually one of the formats in the table above, not the assistant.

What should I ask my developer to change?

Ask for the facts to be in the page as text, and for every control to work without guesswork. A short brief you can hand over:

• Every spec table rebuilt as an HTML table, with the image version removed or kept only as a supplement.
• Capacity, standards, materials, size range and export markets written in page text on the relevant pages. The factory facts assistants look for lists what to include.
• Certificate details (standard, number, issuing body, validity) typed out next to each certificate image.
• Main menu and product links present as ordinary links in the HTML.
• Descriptive alt text on product photos and diagrams.
• Visible labels on all form fields and a clearly named submit button.
• AI crawlers not blocked by accident, as explained in the robots.txt guide.
• A Lighthouse run in Chrome, with the Agentic Browsing results shared with you.

None of this changes how the site looks to a human buyer. It changes whether the first visitor leaves with anything to say about you. EbizIndia has a fuller explanation of the ways ChatGPT and Google fail to read a manufacturing website, and the AI Visibility section has the rest of the reading path.