State of Agentic Commerce — Q2 2026
Three months ago, we started scanning 1,006 e-commerce brands daily to answer a simple question: are online stores ready for AI shopping agents? This is our first quarterly report. The short answer is that most brands are wide open, a small but important minority are actively blocking, and a new standard called llms.txt is quietly gaining traction faster than anyone expected.
The numbers at a glance
Of the 1,006 brands in our index, 970 allow every major AI agent—GPTBot, ChatGPT-User, ClaudeBot, Claude-Web, PerplexityBot, Amazonbot, Google-Extended, and Bytespider—full access through their robots.txt. The remaining 36 brands block or restrict at least one agent, typically through explicit disallow rules or WAF-level challenges.
This does not mean 96% of brands are strategically prepared for AI agents. It means they have not yet taken a position. The absence of a block is not the same as a strategy.
Category breakdown
Blocking is not evenly distributed. Certain categories have disproportionately high block rates, usually tied to content sensitivity, competitive intelligence concerns, or legacy security postures that predate the current AI agent wave.
An important nuance: many blocks we detect are not intentional AI-agent policy decisions. They are side effects of WAF configurations—Akamai, Cloudflare, and PerimeterX rules that challenge or reject non-browser user agents. The brand's robots.txtmay say “allowed” while the infrastructure says “blocked.” We track both layers.
Automotive stands out at 18%. Dealer management platforms (CDK Global, Reynolds and Reynolds) ship default configurations that block non-standard user agents. Most dealership websites inherit these settings without deliberate choice. As AI shopping agents become a real acquisition channel for auto purchases, this will create friction.
A new standard, moving fast
llms.txtis a proposed standard that lets websites communicate directly with large language models—describing what the site does, what content is available, and how an LLM should interact with it. Think of it as a cover letter for AI agents, sitting alongside robots.txt but designed for a fundamentally different reader.
~10% adoption in under three months
Roughly 100 of the 1,006 brands in our index now serve a valid llms.txt file. For a standard with no browser requirement and no SEO incentive, this is remarkably fast.
Adoption is led by Shopify brands. Shopify's app ecosystem has made it trivial to publish an llms.txt file, and the DTC brands on Shopify tend to be early adopters of new web standards. Fashion and beauty brands over-index on adoption. Luxury brands, ironically, are both the most likely to block agents and the least likely to publish llms.txt.
The quality of published files varies enormously. Some are comprehensive, multi-section documents with product taxonomy descriptions and explicit agent permissions. Others are a single line. We are beginning to track llms.txt depth as a signal, and plan to publish a quality index in Q3.
If adoption continues at the current pace, we project 25–30% of the index will have llms.txt by Q4 2026. The inflection point will be platform-level defaults. When Shopify or BigCommerce generates llms.txt automatically for new stores, the curve goes vertical.
Strategic implications
Silence is a position, but it won't stay neutral
96% of brands have no explicit AI agent policy. Today, that means they are open by default. But as agent traffic grows and becomes attributable, regulators and consumers will start asking whether “open by omission” was a deliberate choice. Brands should document their position before it is documented for them.
Your WAF may be making decisions you don't know about
We see a recurring pattern: brands whose robots.txt allows an agent, but whose CDN or WAF blocks the request anyway. This creates a contradictory signal. If you are open to AI agents, verify that your infrastructure agrees.
Structured data is your agent-facing storefront
Agents do not browse visually. They parse JSON-LD, Schema.org Product markup, and Open Graph tags. Brands with strong structured data will be recommended more accurately by agents. Brands without it will be invisible or misrepresented.
Publish llms.txt now—first-mover advantage is real
At 10% adoption, brands that publish a comprehensive llms.txt today are shaping how AI agents understand their category. By the time adoption hits 50%, the early publishers will have established the schema.
Practical implications
Access is not the bottleneck—data quality is
96% of sites are technically reachable. The real challenge is that product data quality varies wildly. JSON-LD coverage, feed freshness, and price accuracy are unreliable across large swaths of the index. Agents that can gracefully handle incomplete data will outperform those that assume clean inputs.
Respect the blocks
The 4% that block are doing so intentionally (or at least their infrastructure is). Circumventing these blocks erodes the trust that keeps the other 96% open. The agent ecosystem is in a cooperative equilibrium right now. Breaking it would trigger a wave of preemptive blocking.
Check llms.txt first
If a brand publishes llms.txt, it is an explicit signal that they want to communicate with your model. Read it. Follow its instructions. This is the emerging social contract between AI and commerce—agents that honor it will get better access and data over time.
Use our API to stay current
Agent access policies change daily. We detected 127 signal changes in our most recent scan alone. The ARC Report public API gives you real-time access to every brand's status—no auth required for read endpoints.
ARC Report scans 1,006 e-commerce brands daily using automated probes. For each brand, we check robots.txt rules for 8 major AI agent user-agents, test actual HTTP responses to detect WAF-level blocks, catalog structured data signals (JSON-LD, Schema Product, Open Graph, product feeds), and check for the presence and content of llms.txt files. Brands span 15 categories from fashion and electronics to grocery and automotive. The full methodology is available on request.
Get the next quarterly report
We publish quarterly. Subscribe to get the Q3 report and weekly intelligence digests in your inbox.
Need deeper data?
Pro subscribers get historical trends, CSV exports, higher API limits, and category-level analytics.