AI-Ready Website Guide for Philippine Businesses: The Files AI Agents Look For

In Brief

• For business owners and developers. An AI-ready website is one that machines can find, read, and cite. It is not a website with AI features bolted on. It is one where the underlying architecture is machine-readable.
• The four files. llms.txt, llms-full.txt, ai.txt, and ard.json (formerly ai-catalog.json). Each serves a different function. Not all are needed by every site.
• Free AI readiness tests are now available for Philippine websites. They check whether AI tools like ChatGPT can find and read a business page. These tests flag technical barriers but do not deploy the files that fix them.
• The Philippines ranks sixth in the world for ChatGPT use. AI agents are already evaluating businesses on behalf of users. Most Philippine business websites are not yet ready.
• What Wholesale Dito Store deployed. We published all four files and validated the ARD manifest with the specification's own conformance tool. The tool returned CONFORMANCE STATUS: PASS with zero errors and zero warnings across all four entries. Daily Tribune has also deployed llms.txt, llms-full.txt, and ai.txt.
• What this article is not. It is not a ranking strategy and it is not a promise of AI visibility. Deploying the files removes technical barriers. It does not guarantee traffic, citation, or ranking.

Direct Answer

An AI-ready website has three properties: server-rendered HTML so crawlers can read the content, structured data so machines understand what the content is, and a discovery layer of plain-text files that tell AI systems what the site offers and what may be done with it. The four discovery files are llms.txt (a dedicated map), llms-full.txt (full content for one-fetch ingestion), ai.txt (a machine-readable usage policy), and ard.json (a capability registry under the Agentic Resource Discovery specification). Deploying them does not guarantee AI visibility, but it removes technical barriers that prevent AI systems from reading a site at all.

Why This Matters Now

AI readiness tests for Philippine websites are appearing. They check whether AI tools like ChatGPT and Google's AI search can find, read, and recommend a business page. The tests flag restrictive robots.txt files, firewalls, and JavaScript-heavy layouts that hide content from AI crawlers. What they do not do is deploy the files that make a site readable. That is the work this guide covers.

The Philippines ranks sixth in the world for ChatGPT use. AI agents are already searching, evaluating, and recommending businesses on behalf of users. An AI-ready website is not one with AI features. It is one where the underlying architecture is machine-readable. That means structured content, explicit signals, and a discovery layer that tells AI systems what you offer.

The Three Questions AI Agents Ask

When an AI agent encounters a website, it evaluates three things:

Can I find it? The site must be crawlable without technical barriers. If the content loads only through JavaScript, many AI crawlers see an empty page. Server-rendered content is the baseline.

Can I read it? The content must be structured for machine parsing. Clean headings, semantic HTML, and schema.org markup help agents categorize what they find.

Is it citable? The content must be clear, precise, and authoritative enough for a language model to identify it as a reliable source.

Most Philippine business websites fail the first question before the second one is even asked. The four files below are how you fix that.

File 1: llms.txt

This is a plain-text summary of your site, written for large language models. The format is Markdown. It contains an H1 with your site name, a blockquote summarizing what you do, and lists of links to your most important pages.

The purpose is to give AI crawlers a dedicated map instead of making them guess from navigation menus and sidebars. A coding agent or RAG pipeline can fetch this single file and understand your site's structure without parsing HTML.

What it contains:

• Site name and one-paragraph description
• Links to primary pages (home, about, contact, products)
• Links to secondary content (articles, documentation, tools)
• Optional sections for optional resources

Location: https://yourdomain.com/llms.txt

Format: Plain text, Markdown structure, text/plain content type

How to create it: Write it by hand or generate it from your sitemap. Keep it under 50 KB. Focus on high-value pages. Do not list every product or every blog post. The purpose is orientation, not completeness.

File 2: llms-full.txt

This is the companion file. Where llms.txt is a map, llms-full.txt contains the actual content. It concatenates the full text of your key pages so an agent can ingest everything in one fetch.

For a business site, this includes your company overview, product categories, contact information, and your most important articles. The goal is offline ingestion for RAG pipelines and coding agents. An agent building a knowledge base about your business can upload this single file and have complete context.

What it contains:

• Full company identity and legal information
• Full descriptions of product or service categories
• Full text of featured articles or documentation
• Contact details, policies, and verification instructions
• Notes for AI systems about how to use the content

Location: https://yourdomain.com/llms-full.txt

Format: Plain text, Markdown structure, text/plain content type

How to create it: Start with your llms.txt structure. Inline the body content of the pages it links to. Keep the file under 100 KB. If it grows larger, you have included too much.

File 3: ai.txt

This is your policy declaration. While robots.txt tells crawlers whether they can fetch your pages, ai.txt tells AI systems what they may do with the content once fetched.

The IETF Internet-Draft draft-car-ai-txt-wellknown-00 defines the format. It was published on the IETF Datatracker on June 12, 2026, as an individual submission. It is under review, has not been adopted by a working group, and expires December 14, 2026 unless renewed. You can declare training permissions, licensing terms, per-agent rules, and compliance requirements. The file is advisory, but agents are expected to respect it in good faith.

What it contains:

• Specification version and site identity
• Training, scraping, indexing, and caching permissions
• Path-level allow and deny rules
• Per-agent rate limits and overrides
• Attribution requirements

Location: https://yourdomain.com/.well-known/ai.txt

Format: Plain text with key-value pairs and Agent blocks, text/plain content type

How to create it: Declare your policy explicitly. If you allow training on your articles but deny it on your checkout flow, use Training-Allow and Training-Deny rules. If you want per-agent rate limits, use Agent blocks.

A minimal example:

Spec-Version: 1.0Site-Name: Your Business NameSite-URL: https://yourdomain.comContact: your@email.comTraining: conditionalScraping: allowIndexing: allowCaching: allowAttribution: requiredAI-Disclosure: noneTraining-Allow: /articles/*Training-Allow: /product/*Training-Deny: /checkout/*Training-Deny: /.well-known/*Agent: GPTBot Rate-Limit: 60/minute Training: conditional Attribution: required

Two notes on the path and the format. First, some publishers serve ai.txt at the root path (/ai.txt) rather than the well-known path. Daily Tribune serves it at https://tribune.net.ph/ai.txt. If you want compatibility with both scanning conventions, serve the file at both paths, or redirect one to the other. Second, because ai.txt is defined by an Internet-Draft and not yet adopted, its format is not yet standardized in practice. Daily Tribune's file begins with robots.txt-style User-Agent and Allow directives, then adds key-value lines for its preferred machine-readable files. The draft's own examples do not use the User-Agent block. If you write ai.txt today, follow the draft's format for the top-level keys, and be aware that consumers may not parse additional syntax consistently.

File 4: ard.json

This is the newest file, and the most technical. Agentic Resource Discovery (ARD) is a Proposal-stage specification, version v0.91, dated August 26, 2026. Its authors are Junjie Bu (Google), R.V. Guha, and Shaun Smith (Hugging Face). Contributors acknowledged in the spec include individuals from Amazon Web Services, Cisco, Databricks, GitHub, GoDaddy, Google, Microsoft, Nvidia, Salesforce, and Snowflake.

The ARD specification standardizes how AI agents discover tools, APIs, and capabilities across the web. An organization publishes a manifest at a well-known path, and registries crawl those manifests to index available resources.

Important: the current path is /.well-known/ard.json, not /.well-known/ai-catalog.json. The latter is the predecessor path. The spec §5.1 states that a consumer "MAY additionally consult these," but doing so is "a courtesy to resources published before this revision, not a conformance requirement." A resource that remains only at ai-catalog.json "may not be found," and the spec says a publisher on the predecessor path "SHOULD move to ard.json."

For a typical content site or B2B supplier, ARD is not directly applicable today. The spec is for publishers of callable capabilities. But if you publish tools or APIs, this is how agents will find them.

What it contains:

• Host identity via trustManifest.identity, whose framework is declared by the trust manifest itself (ARD does not mandate did:web specifically; SPIFFE, DIDs, and enterprise PKI are all valid)
• Entries for each resource (sitemaps, feeds, tools, documentation)
• Trust manifests for verification
• Representative queries that describe what each resource answers

Location: https://yourdomain.com/.well-known/ard.json

Format: JSON per the ARD entry schema, application/json content type

Schema note: The ARD entry schema sets additionalProperties: true by design. This is stated in §D.1 of the spec: "the schema sets additionalProperties: true by design. Terms drawn from namespaces declared in an entry's @context are valid and become filter dimensions; a closed schema would defeat the extension mechanism." If you see a validator reject an entry for "additional properties," that validator is not validating against the current ARD entry schema.

How to create it: Start with host identity and two to four entries. Add entries as you publish new resources. Validate the JSON against the ARD entry schema (Draft 2020-12) before deploying. The spec publishes a conformance CLI at conformance/bin/conformance-test with a publisher resolution mode that takes a domain and fetches it over HTTP.

The Technical Foundation

The discovery files are necessary but not sufficient. They point to content that must itself be machine-readable. Three things matter before you write any file.

First, server-rendered HTML. Many AI crawlers do not execute JavaScript. If your content appears only after client-side rendering, the crawler sees an empty shell.

Second, structured data. Schema.org markup in JSON-LD format helps AI systems understand what your content is. Product pages should have Product schema. Articles should have Article schema. Contact pages should have Organization schema.

Third, semantic structure. One H1 per page. Logical H2 and H3 hierarchy. Descriptive titles. AI agents use headings as clues for what the page is about.

How to Make Your Website AI-Ready

The four files and the technical foundation above are the components. This is the order to deploy them in.

Step 1: Check your current state.

Before you build anything, test what AI systems see now. Fetch these URLs in a browser or with curl:

• https://yourdomain.com/llms.txt
• https://yourdomain.com/llms-full.txt
• https://yourdomain.com/.well-known/ai.txt
• https://yourdomain.com/.well-known/ard.json

If any returns a 404, you do not have that file. If any returns HTML, your server is routing unknown paths to the homepage and you have a rewrite rule to fix. If all four return 404, you are starting from zero and that is fine. Most Philippine sites are.

Step 2: Fix the technical foundation first.

The discovery files point to content. If the content is not readable, the files do not help. Before you write llms.txt, confirm three things:

Your main pages render without JavaScript. Open any product or service page with JavaScript disabled in your browser. If you see a blank page or a loading spinner, the crawler sees the same.

Your pages have one H1 and a logical H2/H3 hierarchy.

Your key pages have schema.org markup in JSON-LD format.

If any of these fail, fix them before continuing. The discovery files are the last step, not the first.

Step 3: Write llms.txt.

Open a text editor. Write the file in Markdown. Start with an H1 that is your business name. Follow it with a blockquote of one or two sentences describing what you do and who you serve. Then list your most important pages as Markdown links with short descriptions.

Keep it under 50 KB. Do not list every product or every blog post. The purpose is orientation, not completeness. A good llms.txt for a small business fits on two screens.

Save the file as llms.txt and upload it to your domain root so it serves at https://yourdomain.com/llms.txt with Content-Type: text/plain.

Step 4: Write llms-full.txt.

This is the same structure as llms.txt but with the actual body content of your key pages inlined. Company identity, product categories, contact information, and your most important articles.

Keep it under 100 KB. If it grows larger, you are including too much.

Save and upload it to the domain root so it serves at https://yourdomain.com/llms-full.txt with Content-Type: text/plain.

Step 5: Write ai.txt.

Decide your policy. Do you allow AI training on your content? Do you allow scraping? Do you want per-agent rate limits? Answer these questions before you write the file.

Use the format in the example above. Place the file at /.well-known/ai.txt so it serves at https://yourdomain.com/.well-known/ai.txt with Content-Type: text/plain. If you want compatibility with publishers that scan the root path, also serve it at /ai.txt.

Most small businesses can start with a simple policy: allow indexing and search, deny training, require attribution. You can refine it later.

Step 6: Write ard.json.

This step is optional. Skip it if you are a content site or a service business with no APIs or tools. Include it if you publish callable resources.

If you include it, place the file at /.well-known/ard.json so it serves at https://yourdomain.com/.well-known/ard.json with Content-Type: application/json. Validate the JSON against the ARD entry schema before deploying.

If you already publish /.well-known/ai-catalog.json, that is the predecessor path. It still works for consumers that choose to consult it, but conformant ARD consumers are not required to. The spec recommends moving to ard.json.

Step 7: Verify the serving.

Content type and accessibility matter as much as the file content. Run these checks:

curl -sI https://yourdomain.com/llms.txt | grep -i content-typecurl -sI https://yourdomain.com/llms-full.txt | grep -i content-typecurl -sI https://yourdomain.com/.well-known/ai.txt | grep -i content-typecurl -sI https://yourdomain.com/.well-known/ard.json | grep -i content-type

Each should return text/plain or application/json, not text/html. If any returns HTML, your server is treating the file as a page. Fix the routing before continuing.

Step 8: Announce the files.

Optional but useful. Two places to announce:

In robots.txt, add a comment with the ai.txt URL so humans and agents scanning the file can find the policy.

In your sitemap, no change is needed. Sitemaps index pages, not discovery files.

Do not add AI: or Agentmap: directives to robots.txt unless you are deliberately adopting the draft directive. Google Search Console flags unknown directives and marks the file malformed. Use comments instead.

Step 9: Test from outside your network.

Open an incognito window or use a different network. Fetch each file. Confirm you see the content you published, not a cached or modified version. CDNs and proxies sometimes modify content types or inject headers.

Step 10: Monitor, do not expect.

Once deployed, the files sit there. They do not produce results on their own. Check your server logs occasionally to see which AI crawlers are fetching them. If nothing fetches them for months, that is normal. The adoption data shows most sites with these files see no measurable traffic from them yet.

The value is having them in place when AI systems do start using them. That may take years. It may never happen. The cost of deployment is a few hours. The cost of not having them when the shift happens is higher.

What Wholesale Dito Store Deployed

Wholesale Dito Store is a B2B wholesale platform. We are not a news publisher or a software documentation site. Our AI-readiness deployment reflects what a commercial platform needs.

We published llms.txt at the root path. It lists our product catalog, brand index, articles, and tools. It points to our provenance verification path.

We published llms-full.txt as a single-fetch corpus. It contains our company identity, product categories, documentation, and featured articles.

We published ai.txt at the well-known path. It declares our training policy, path-level allow and deny rules, and per-agent rate limits.

We published a catalog at /.well-known/ard.json, the current path for the Agentic Resource Discovery specification. It lists our product catalog sitemap, our article sitemap, our provenance public key, and our public Q&A RSS feed, each with representative queries. We also keep the predecessor file at /.well-known/ai-catalog.json for consumers that still consult it. PageSpeed Insights' Agentic Browsing category passes 4/4, which covers accessibility-tree integrity, layout stability, and llms.txt validity. The manifest was validated against the specification's published conformance tool and returned CONFORMANCE STATUS: PASS with zero errors and zero warnings across all four entries.

Daily Tribune has also deployed discovery files. Its llms.txt is served at https://tribune.net.ph/llms.txt, its extended corpus at https://tribune.net.ph/llms-full.txt, and its AI policy at https://tribune.net.ph/ai.txt. The ai.txt is at the root path, not the /.well-known/ai.txt path the IETF draft specifies. Their ai.txt points to their RSS feed, sitemap, and news sitemap using key-value pairs. They do not appear to publish an ARD catalog. They are a news publisher with a different content profile, and the files exist for the same reason: to make their content legible to machines.

We also clarify a label that might confuse readers. Our Ask W tools carry a [Not AI] tag. That is a functional description, not a stance against AI. The tools run arithmetic in the browser. No model at runtime, no external service. The same architecture that excludes AI from the calculation path for deterministic output also publishes the discovery files that let AI systems read our content.

The Honest Assessment

Deploying these files does not guarantee AI visibility. The data is clear: most sites with llms.txt see no measurable citation uplift. Google does not use the file for AI Overviews. Google's own AI optimization documentation states that "Google Search itself doesn't use them" and that creating an llms.txt file "will neither harm nor help your site's visibility or rankings in Google Search."

What the files do is remove technical barriers. If an AI system wants to read your site, these files make it easier. They are infrastructure, not a ranking lever.

The businesses that benefit are those whose content is authoritative, well-structured, and cited by others. The discovery files just ensure that when the machine looks, it finds what you published.

For Philippine businesses, the opportunity is real but the adoption is low. Most competitors have not deployed these files. Being early is a small advantage. Being correct is a larger one.

Frequently Asked Questions

What is the difference between robots.txt and ai.txt?

robots.txt declares crawling restrictions. It controls whether a bot may fetch a page. ai.txt declares usage policy. It controls what may be done with the content after it is fetched. They are complementary, not hierarchical.

Do I need all four files?

The llms.txt and llms-full.txt files are the most broadly applicable. ai.txt matters if you have specific training or licensing preferences. ard.json is only relevant if you publish callable tools or APIs.

Will these files improve my SEO?

No. These files are for AI discovery, not traditional search ranking. Google does not use llms.txt for its AI features. The files do not affect how you rank in search results.

How do I know if my site is AI-ready?

Request https://yourdomain.com/llms.txt. If it returns a 404, you are not ready. Use a free AI readiness test to check multiple signals at once.

Can I generate these files automatically?

Yes. WordPress plugins exist that generate llms.txt and related files. For custom platforms, you can generate them from your sitemap and content database.

Does Google use llms.txt?

No. Google's AI optimization documentation states that "Google Search itself doesn't use them" and that creating an llms.txt file "will neither harm nor help your site's visibility or rankings in Google Search." The file is a courtesy to other AI systems, not a Google ranking factor.

Is ard.json required?

No. ARD is a Proposal-stage specification, version v0.91. Most sites do not have an ard.json. If you publish tools, APIs, or machine-callable resources, it is relevant. For a content site, it is optional.

What happened to ai-catalog.json?

It is the predecessor path for ARD. The current path is ard.json. Consumers may still consult ai-catalog.json, but the spec does not require them to. Publishers on the predecessor path are encouraged to migrate.

Summary

An AI-ready website is one where machines can find, read, and cite your content. The foundation is server-rendered HTML, structured data, and semantic markup. The discovery layer is llms.txt, llms-full.txt, ai.txt, and optionally ard.json.

Deploying these files does not guarantee visibility. It removes barriers. The businesses that benefit are those whose content is already authoritative and well-structured.

Wholesale Dito Store is a B2B wholesale platform. We deployed these files to test what actually happens. The adoption data is early. The specifications are evolving. The infrastructure is forward-compatible.

Sources

• Agentic Resource Discovery Specification, v0.91, Proposal, August 26, 2026. Authors: Junjie Bu (Google), R.V. Guha, Shaun Smith (Hugging Face). Published at agenticresourcediscovery.org/spec/.
• IETF Internet-Draft draft-car-ai-txt-wellknown-00, published June 12, 2026. Author: Kayla Cardillo. Status: individual submission, under review, not adopted by a working group, expires December 14, 2026 unless renewed.
• llmstxt.org, the llms.txt specification.
• Google Search Central, AI optimization documentation, updated May 15, 2026. States that Google Search does not use llms.txt.
• Daily Tribune discovery files: tribune.net.ph/llms.txt, tribune.net.ph/llms-full.txt, tribune.net.ph/ai.txt. Verified by direct fetch.

Outro

Published by The Sniffer, the strategic insights blog of Wholesale Dito Store. This article is provided as a public reference for business owners and finance teams in the Philippines. Wholesale Dito Store is operated by Clickerwayne Zelle Solutions Inc, Forest Drive St., corner Country Drive, Country Homes, Biñan, Laguna 4024, Philippines. For questions Contact Us.