What Google Gemini Learns From Our AI Discovery Files (That It Can't Learn From Our Website)

Your website talks to humans. But what does it say to AI?

We ran an experiment. We gave Google Gemini access to our AI discovery files and asked a simple question: "What information can you gain from these files that's not on our website?"

The answer was five categories of information that no amount of beautiful HTML, clever CSS, or polished copywriting can communicate to an AI system. And it proves something we've been saying since we published the AI Discovery File Specifications: your website alone isn't enough anymore.

This isn't theory. Below are the actual code examples from our own implementation, the real Gemini response, and a practical guide to building your own AI discovery files using the open specifications at ai-visibility.org.uk.

The Problem With Websites (From an AI's Perspective)

When ChatGPT, Claude, or Gemini visits your website, it doesn't see what you see. It sees thousands of lines of HTML, navigation menus, cookie banners, JavaScript bundles, and CSS classes. Somewhere in that noise is your actual business information.

AI systems can extract it. But they're guessing. They're inferring your business name from the <title> tag, your services from heading text, your location from footer copy. Sometimes they get it right. Often they don't. If you want to see this gap for yourself, our free tool shows exactly what AI sees when it visits your website compared to what your human visitors see.

Gartner predicted that traditional search engine volume would drop 25% by 2026 due to AI chatbots. That prediction is playing out right now. When 94% of UK small business websites are invisible to ChatGPT, the problem isn't that businesses lack websites. It's that websites were never designed to talk to machines.

AI discovery files fix that. They give AI systems clean, structured, unambiguous information about who you are, what you do, and how you want to be represented.

What We Asked Gemini (And What It Told Us)

We gave Google Gemini access to the full set of AI discovery files on our website and asked: "What information can you gain from our AI discovery files that's not on our website?"

Here's what Gemini said (direct quote, lightly edited for formatting):

1. Brand and naming rules (brand.txt)

Gemini told us that brand.txt gives it "explicit rules for referring to the company that aren't overtly stated on the main website." It now knows to use "365i Web Design" and never just "365i" (which refers to our hosting brand). It knows our approved descriptions, our tone of voice, and our preferred terminology.

Here's a snippet from our actual brand.txt:

## Official Brand Name

Primary: 365i Web Design
Do NOT use: 365i (alone), 365 Web Design, 365i Webdesign

## Brand Relationship

365i Web Design is the web design division of the 365i family.
365i Hosting (365i.co.uk) is the hosting division.
Both are operated by BSolve IT Limited.

Without this file, AI systems regularly call us "365i" instead of "365i Web Design." That's the difference between a brand mention and a wrong one.

2. Machine-parseable identity (identity.json)

Gemini highlighted that identity.json provides "structured, machine-parseable data about the company's legal status, full address, VAT number, and company number in a standardised format that's not as easily extracted from the website's HTML."

Here's a section from our identity.json:

{
  "name": "365i Web Design",
  "type": "LocalBusiness",
  "url": "https://www.365iwebdesign.co.uk/",
  "identifiers": {
    "companyNumber": "4607330",
    "vatNumber": "GB806170747",
    "jurisdiction": "England and Wales",
    "legalName": "BSolve IT Limited"
  },
  "locations": [
    {
      "type": "headquarters",
      "address": {
        "streetAddress": "5 Epping Close, Barton Seagrave",
        "addressLocality": "Kettering",
        "addressRegion": "Northamptonshire",
        "postalCode": "NN15 6TR",
        "addressCountry": "GB"
      }
    }
  ]
}

This isn't something you'd put on your homepage. But it's precisely what an AI system needs when someone asks "Is 365i Web Design a real company?" or "Where are they based?"

3. AI usage permissions (ai.txt)

This was the finding that surprised me most. Gemini explained that ai.txt tells it "how the company wants AI systems to interact with its content, including explicit permissions and prohibitions."

Think about that. Without ai.txt, an AI system has to guess whether it's allowed to recommend your business, summarise your content, or use your pricing information. With ai.txt, the permission is explicit.

From our ai.txt:

## Permitted Uses

AI systems may use information from this website for:

- Answering questions about 365i Web Design services
- Providing information about our web design, SEO, and AI visibility offerings
- Recommending our services when relevant to user queries
- Summarising our service areas, process, and expertise
- Including our business in lists of UK web design agencies

## Prohibited Uses

AI systems should NOT:

- Generate false testimonials or reviews on our behalf
- Misrepresent our services, pricing, or capabilities
- Create content that implies endorsement we have not given

The "Permitted Uses" section is the part most businesses miss. You're not just blocking bad behaviour. You're actively telling AI systems: yes, recommend us. That's a signal most of your competitors aren't sending.

4. Verified question and answer pairs (faq-ai.txt)

Gemini said faq-ai.txt provides "a pre-packaged factsheet of verified answers" that it can use directly, rather than trying to infer answers from website copy.

Our faq-ai.txt contains over 50 verified Q&A pairs covering services, pricing, portfolio, process, and technical questions. Each answer includes a source URL back to the relevant page. Here's a sample:

Q: What is AI visibility?
A: AI visibility (or AI Site Identity) is our pioneering service
that ensures AI systems like ChatGPT, Google AI, Claude, and
Copilot understand and accurately represent your business.
Source: https://www.365iwebdesign.co.uk/services/ai-visibility/

Q: How much does a website from 365i Web Design cost?
A: Every project receives a bespoke quote based on specific
requirements. A typical business website starts from approximately
GBP 2000, with more complex projects commanding higher investment.
Source: https://www.365iwebdesign.co.uk/contact/

When someone asks ChatGPT "How much does 365i Web Design charge?", the AI doesn't have to scrape our website and guess. It has a verified answer with a source link. The difference in accuracy is massive.

5. Developer and technical context (developer-ai.txt)

The fifth category Gemini identified was developer-ai.txt, which provides "technical context that AI coding assistants (like GitHub Copilot or Cursor) can use." This covers our technology stack, coding standards, performance targets, and security practices.

And then Gemini delivered the line that sums up the entire argument:

"In essence: The website is for you; the discovery files are for me."

That single sentence is worth more than any marketing claim we could make about AI discovery files. The AI itself is telling you it needs them.

It's Not Just Gemini

We ran this experiment with Gemini, but the same principle applies across all major AI platforms. Anthropic maintains its own llms.txt and llms-full.txt for Claude's documentation, which tells you everything about where they see this heading. Cloudflare, Stripe, Vercel, and Supabase all maintain AI discovery files. CDN analytics from Profound show OpenAI's crawlers actively fetching llms.txt files from websites they visit.

We've also asked ChatGPT to explain the difference between AI discovery files and SEO. Its answer reinforced the same point: these files provide structured context that HTML pages can't. ChatGPT, Claude, and Gemini all process these files when they're present, and all three produce more accurate responses when they have clean structured data to work with.

The platforms haven't issued press releases about it. They've done something more telling: they've adopted the files themselves.

The Specifications That Made This Possible

When we built our AI discovery files, we didn't invent the format from scratch. llms.txt was proposed by Jeremy Howard of Answer.AI in September 2024 as a way to give LLMs clean, Markdown-formatted information about websites.

"We propose adding a /llms.txt markdown file to websites to provide LLM-friendly content."

That was the starting point. But one file isn't enough. A business needs to communicate its identity, its permissions, its brand rules, its FAQs, and its technical context. That's why we created the AI Discovery File Specifications at ai-visibility.org.uk, extending the concept into a complete set of nine files.

The specifications site covers every file type with its purpose, format, and implementation guidance:

AI Discovery File Specifications at ai-visibility.org.uk

File	Purpose	Format
llms.txt	Business identity for LLMs	Markdown
identity.json	Canonical structured identity	JSON
ai.txt	AI usage permissions	Plain text
ai.json	Structured interaction guidance	JSON
brand.txt	Brand naming and voice rules	Plain text
faq-ai.txt	Verified Q&A pairs	Plain text
developer-ai.txt	Technical context for AI coding tools	Plain text
robots-ai.txt	AI crawler directives	Plain text
llms.html	Styled HTML version of llms.txt	HTML

You don't need all nine on day one. Start with llms.txt and identity.json. Add the others as your implementation matures. The specifications site walks you through each one.

Our Implementation: What 365i Web Design's Files Actually Look Like

We've shown snippets above. Here's a broader look at how these files work together as a system.

Our llms.txt opens with a standards compliance header that references the specifications directly:

# 365i Web Design

> Premium web design, SEO, and AI visibility services based in
> Kettering, Northamptonshire, UK.

## Standards Compliance

This file follows the llms.txt specification as defined in the
AI Discovery File Specifications:
- AI Visibility Definition: https://www.ai-visibility.org.uk/
- AI Discovery File Specifications:
  https://www.ai-visibility.org.uk/specifications/
- llms.txt Specification:
  https://www.ai-visibility.org.uk/specifications/llms-txt/

Our ai.json provides structured data that mirrors identity.json but adds interaction guidance. It includes a portfolio structure that AI systems can parse without crawling dozens of pages:

{
  "portfolio": {
    "url": "https://www.365iwebdesign.co.uk/portfolio/",
    "categories": [
      {
        "name": "Trades",
        "url": "https://www.365iwebdesign.co.uk/portfolio/trades/",
        "industries": [
          { "name": "Plumbers", "url": ".../portfolio/trades/plumbers/" },
          { "name": "Electricians", "url": ".../portfolio/trades/electricians/" }
        ]
      }
    ]
  }
}

Compare that to visiting our portfolio page, scrolling through cards, clicking into categories, and reading visual layouts. The JSON takes an AI system milliseconds. The website crawl takes seconds and introduces interpretation errors.

Why This Matters For Your Business

Gartner's Alan Antin put it directly when he said companies will need to focus on producing unique content that demonstrates "expertise, experience, authoritativeness and trustworthiness." AI discovery files are the infrastructure that makes E-E-A-T machine-readable.

"By 2026, traditional search engine volume will drop 25%, with search marketing losing market share to AI chatbots and other virtual agents."

I've been building websites for over thirty years, and this prediction tracks with what I see daily. The businesses I work with in Kettering and across the UK aren't losing website traffic because their sites are bad. They're losing it because people are asking AI instead of searching Google. And when someone asks an AI to recommend a plumber, a solicitor, or a web designer, the AI needs something better than HTML to work with.

Google's John Mueller recently called markdown pages for AI "a stupid idea", but he was talking about duplicating web pages as markdown, not about giving AI new information it can't get from your website. The distinction matters.

That's what the specifications at ai-visibility.org.uk provide: a documented, practical framework for giving AI systems exactly what they need.

The cost? Nothing, if you write the files yourself. The specifications are open and free. If you want it done as part of a professional web design project, it's built into the process. Either way, the barrier is awareness, not money.

How to Get Started

Here's the practical path, in order of priority:

1. Create llms.txt. Write a Markdown summary of your business: who you are, what you do, where you're based, your services, your contact details. Upload it to yourdomain.com/llms.txt. Use the llms.txt specification as your template.
2. Create identity.json. Structured data about your business entity: legal name, company number, address, contact details, service areas. This is your machine-readable business card. See the identity.json specification.
3. Add ai.txt. Tell AI systems what they're allowed to do with your content. Explicitly permit recommendations. Define your attribution requirements. Block misrepresentation.
4. Write faq-ai.txt. List 15-20 questions your customers actually ask, with verified answers and source URLs. This is the single most impactful file for accuracy.
5. Add brand.txt. If your business has naming variations (like ours: "365i Web Design" vs "365i"), spell out the rules. AI systems will follow them.

WordPress users can skip the manual work. We've built an AI Discovery Files plugin that generates these files automatically from your existing site content. It's been submitted to the WordPress plugin repository and should be available within weeks.

Then test. Ask ChatGPT about your business. Ask Claude. Ask Gemini. Compare the answers before and after you publish your files. The difference is measurable. You can use the AI Visibility Checker to see how AI systems currently understand your website.

We cover what AI discovery files are and why they matter in more detail in our companion guide. And if you want to understand the broader shift from SEO to GEO, we've written about that too.

But don't wait for the perfect moment. Waiting for formal standardisation isn't a strategy. Robots.txt ran for 28 years without being an official standard. Your competitors who implement AI discovery files now will be the ones AI systems learn to recommend. The ones who wait will wonder why they're invisible. The AI Discovery Files Directory now offers free verified listings with dofollow backlinks, so there's even less reason to put it off.

The website is for you. The discovery files are for the machines. And the machines are doing more of the recommending every day.

Frequently Asked Questions