A neutral infrastructure layer for Agentic Commerce

This project is the reference implementation of the Agentic E-commerce OpEx Optimization initiative.

AI agents are becoming primary interfaces to commerce. ChatGPT, Claude, and Gemini increasingly answer shopping questions, compare products, and guide purchase decisions. However, the current web stack (HTML + JavaScript + scraping) is fundamentally incompatible with reliable, scalable AI-mediated commerce.

CommerceTXT is an open, read-only, token-efficient context protocol that:

1. Eliminates hallucinations in commerce interactions through deterministic data
2. Reduces inference costs by 95-99% compared to HTML scraping
3. Aligns incentives between merchants, AI platforms, and users
4. Preserves the open web while enabling agent-native commerce

Supporting CommerceTXT is not just beneficial—it's a defensive infrastructure investment for the Agentic Web.

AI agents today must:

• Fetch 2-5 MB HTML pages to extract a single price
• Execute JavaScript (or fail when it's required)
• Infer prices, availability, and return policies probabilistically
• Re-scrape the same data thousands of times per day

For AI Platforms:

• Massive token waste (8,000+ tokens per product page)
• High inference costs
• Hallucinated prices and inventory
• Legal/reputational risk from incorrect information

For Merchants:

• Server overload from AI crawlers
• Loss of control over how products are represented
• No reliable channel to update AI with correct data
• Distrust of AI recommendations

For Users:

• "Let me check that for you" loops
• Incorrect pricing information
• Unreliable availability data
• Broken shopping experiences

This cannot be fixed by better models alone. It requires better data structures.

CommerceTXT is an interface contract for the Agentic Web, not a platform.

Anthropic championed llms.txt for documentation discovery. CommerceTXT applies the same philosophy to commerce:

• llms.txt: "Here's our documentation"
• CommerceTXT: "Here's our commerce data"

Supporting CommerceTXT positions Anthropic as the standard-bearer for AI-native web protocols.

Before CommerceTXT:

User: "What's the price of Sony XM5 headphones?"
Claude: [scrapes 2.5 MB HTML page, 8,500 tokens]
       "They appear to be around $350, but prices may vary..."

With CommerceTXT:

User: "What's the price of Sony XM5 headphones?"
Claude: [reads 5 KB commerce.txt, 380 tokens]
       "Sony XM5 headphones are $348.00 (in stock, ships in 3-5 days)"

Result: 95% token reduction, deterministic accuracy, better UX.

Anthropic's developer community would rally around this:

• Clear value proposition (cost savings + accuracy)
• Aligns with Claude's "helpful, harmless, honest" principles
• Natural fit for MCP (Model Context Protocol) ecosystem

Being first to support CommerceTXT gives Anthropic:

• "Most accurate for shopping" positioning
• Early adopter advantage with merchants
• Leadership in defining Agentic Web standards

Anthropic can move faster than Google/OpenAI on this. No legacy Merchant Center to navigate, no internal politics—just adopt what works.

ChatGPT processes millions of commerce queries daily. Current approach wastes tokens on:

• DOM traversal and layout parsing
• Review scraping (15,000+ tokens)
• Availability inference across multiple pages

CommerceTXT Impact: 95-99% token reduction = millions in inference cost savings annually.

Price, stock, returns, and subscription terms are hard facts—they should never be guessed.

Current risk: ChatGPT quotes wrong price → user makes decision → discovers error → blames OpenAI
With CommerceTXT: Factual data is deterministic, verifiable, and traceable to source

Result: Lower reputational and potential legal risk.

Structured context eliminates "let me check" loops:

• Faster responses
• Fewer follow-up questions
• Higher user satisfaction and retention

Google already invests heavily in structured commerce data (Merchant Center, Shopping Graph, Schema.org).

CommerceTXT Impact:

• Unifies Search, Shopping, and Agent data layers
• Works with existing Schema.org (not against it)
• Optimizes structured data for LLM consumption

This is not disruption—it's evolution.

Merchants are frustrated with scraping overhead and lack of control over AI representations.

CommerceTXT Impact:

• Gives merchants a "single source of truth" they control
• Reduces adversarial scraper load on their servers
• Provides deterministic channel for updates

Result: Better relationship between Google's crawlers and independent merchants.

AI crawling is expensive and energy-intensive.

CommerceTXT Impact:

• Cuts payload sizes by ~99% (2.5 MB → 5 KB)
• Eliminates headless browser JavaScript execution
• Reduces carbon footprint of AI commerce

This aligns with Google's sustainability commitments while delivering business value.

If a single vendor controls the commerce protocol:

• ❌ Merchants resist (fear of lock-in)
• ❌ Competitors reject it
• ❌ Regulators intervene
• ❌ Adoption stalls

An open, CC0 standard like CommerceTXT:

• ✅ Encourages ecosystem adoption
• ✅ Prevents fragmentation
• ✅ Avoids antitrust risk
• ✅ Accelerates innovation

The winning move is to support, not own.

Just as no single company owns robots.txt, sitemap.xml, or Schema.org—CommerceTXT succeeds by being vendor-neutral infrastructure.

AI platforms do not need to:

• Mandate merchant adoption
• Replace existing systems
• Build merchant tooling
• Make large infrastructure investments

Minimal support looks like:

1. Acknowledge CommerceTXT as a supported input format in documentation
2. Experiment with optional agent-side consumption
3. Participate in spec discussions (non-governing role)
4. Publish best practices for reading commerce.txt files

That's it. Low risk, high leverage.

The web evolved through layered standards:

1994: robots.txt (crawler guidance)
2005: sitemap.xml (discovery optimization)  
2011: Schema.org (structured data)
2024: llms.txt (AI context)
2025: CommerceTXT (commerce context)

Each standard emerged to solve a specific infrastructure gap.
Each succeeded because it was open, simple, and beneficial to all parties.

CommerceTXT follows this pattern.

AI agents will discuss products whether merchants like it or not.

The only question is: Will they do it accurately, efficiently, and transparently?

• ✅ Agents quote correct prices (deterministic)
• ✅ Merchants control their representation
• ✅ Platforms reduce costs and liability
• ✅ Users get reliable information

• ❌ Agents hallucinate prices (probabilistic scraping)
• ❌ Merchants distrust AI platforms
• ❌ Platforms absorb legal/reputational risk
• ❌ Users lose confidence in AI shopping

For Anthropic:
You pioneered llms.txt. Pioneer CommerceTXT. Add support to Claude, publish integration docs, and lead the Agentic Commerce movement.

For OpenAI:
Reduce inference costs, eliminate commerce hallucinations, and position ChatGPT as the most reliable shopping assistant.

For Google:
Extend your Shopping Graph leadership into the Agent era. CommerceTXT is the bridge between Search and Agentic Commerce.

• Specification: https://github.com/commercetxt/commercetxt/blob/main/spec/README.md
• GitHub: https://github.com/commercetxt/commercetxt/
• Discussion: https://github.com/commercetxt/commercetxt/discussions/
• Email: hello@commercetxt.org

CommerceTXT is CC0 (Public Domain). No permission needed—just implement and ship.

The Agentic Web needs deterministic commerce infrastructure.

CommerceTXT provides it.

The question is: Who will lead?