A practical guide to agentic commerce.

Begin with the basics, then dig into bot access, product data, security, and the standards shaping AI shopping.

Choose a path

FEATURED RESEARCH

2026 AI Commerce Readiness Index

500+ retailer scans · 15 min read
Read the Index

Browse by topic

Start Here

Foundations

What is Agentic Commerce?

AI agents are becoming shoppers. Learn what that means for your business and how to prepare.

The Future of Shopping

Emerging standards like UCP, MCP, and WebMCP that will define the next era of e-commerce.

Thirty years of bots: the cat-and-mouse game that was never necessary

From Archie to GPTBot, the web keeps fighting the same access problem.

Security & WAFs

Access, controls, and risk

The WAF paradox: why protection infrastructure is blocking AI agents

WAFs cannot reliably tell a human-backed agent from hostile automation.

WebMCP: when your website starts handing agents the keys

What happens when pages expose callable tools inside the user's browser session.

What AI agents actually see when they visit your website

345+ European retail sites scanned. The agent sees what it can parse, not your brand story.

Product Data

Feeds, GTINs, and per-product readiness

Listed. In stock. The wrong bottle.

Why missing GTINs and weak product identity break agent matching.

A feed tells agents what you sell. Your site tells them if it’s true.

Why feeds, product pages, firewalls, and retailer listings all shape whether agents can buy.

Your product feed is your campaign now

Google AI Max reads Merchant Center attributes like a document. Agents use the same logic.

Per-product, not per-site: why every item needs its own agent score

Site averages hide the exact products that will fail agent discovery or checkout.

AOCF 1.0 — Agent-Optimised Catalogue Feed

An open extension for declaring per-product agent purchase terms.

Standards & Payments

Protocols and infrastructure

The agentic payments map

ACP, UCP, MCP, TAP, VIC, and MAP: what each protocol does and what they still need from you.

UCP Is Not Enough

Every brand listed on checkout.directory as UCP-ready still scored Not Ready for agentic commerce.

The Future of Shopping

How emerging protocols will shape agent identity, payment, and checkout.

Who does a business trust when an AI agent arrives?

Updated for Poppy 1.0: what discovery, UCP and merchant controls establish, and what still needs testing.

Market Research

Reports and analysis

2026 AI Commerce Readiness Index

Data from 500+ retailer scans and the infrastructure gaps blocking autonomous commerce.

Your AI assistant is only as good as your infrastructure

Most retail brands have launched AI agents on systems never designed to support them.

Regional Analysis

Market-specific findings

Latin America's AI Commerce Gap

75 LATAM e-commerce sites scanned. Average transactability: 29.4/100.

El mercado de belleza en México y el muro del comercio agéntico

19 marcas líderes analizadas. Las marcas tienen el mejor contenido y el peor checkout.

News

Company and ecosystem updates

Anthropic just lowered the cost of building a shopping agent

Claude for Commerce makes the agent layer cheaper to build. It does not make a retailer's catalogue, cart, or checkout agent-ready.

Aidō Lighthouse joins the Agentic Commerce Consortium

Why the consortium's market problems map directly to what we see in readiness scans.

Twenty years of keeping bots out. Now we need to let the right ones in.

Bot protection is still necessary. The problem is that AI shopping agents are automated traffic too, and many sites now reject agents acting on behalf of real customers.

May 2026 dataset
486
e-commerce sites scanned and analyzed
unique domains
25.7%
block or challenge AI agent access
125 of 486 sites
40%
of blocked sites use hCaptcha
48 of 119 identified
#1
identified blocker is hCaptcha
Cloudflare doubled month-on-month

Provider breakdown · 119 sites

hCaptcha
48
Cloudflare
26
Google reCAPTCHA
13
Generic WAF
13
DataDome
5

How hard is agent allowlisting?

Rate limitersTrivial
Cloudflare WAF rulesEasy
Imperva / AWS WAF / SucuriModerate
Akamai Bot ManagerHard
DataDome / PerimeterX / HUMANHard
Kasada / Shape / F5Very hard

Data from 486 unique e-commerce sites scanned by Aidō Lighthouse as of May 2026. Bypass difficulty reflects the merchant effort required to allowlist verified AI agent user agents, not malicious circumvention. Learn how protocols like Visa TAP and ACP are creating verifiable agent identity →

How we keep Aidō customers secure

The scanner reads third-party websites on your behalf. Those sites can contain anything. Our job is to analyze them accurately without letting what they contain touch anything it should not. That shapes every security decision we make. Reused verbatim from our full Scanner Policy page below.

Our threat model is not typical

Most SaaS platforms worry about what users submit. We also worry about what scanned URLs submit back. A merchant site could return a page with prompt injection payloads designed to manipulate our AI analysis, or HTML crafted to trigger SSRF against our internal infrastructure. These are not theoretical.

We treat every scanned URL as potentially hostile. Prompt injection defense applies to all AI-processed content. SSRF checks run before any outbound request. Security events are logged in enough detail to tell a legitimate scan from an attempted exploit.

Controls in production

We aim for OWASP ASVS Level 2 compliance. The controls below are implemented and active, not aspirational.

SSRF protection on all outbound requests
Every URL is validated before the scanner touches it. Internal networks, loopback addresses, and cloud metadata endpoints are blocked at DNS resolution time, not just at the input stage. That includes DNS rebinding, where a hostname resolves to a safe address during validation and a private one by the time the request actually fires.
Prompt injection defense
Page content from scanned sites goes through validation before it reaches any AI model. The validator strips known injection patterns, role-override attempts, and instruction-injection payloads. Anything it catches, we investigate.
Credential handling
Passwords are hashed with a memory-hard algorithm. Tokens are short-lived with hard expiry. Passkeys (WebAuthn) are available for passwordless login. Secrets and signing keys live in GCP Secret Manager. Nothing sensitive is in source control or container images.
Security headers
HSTS, Content-Security-Policy, X-Frame-Options, X-Content-Type-Options, and Referrer-Policy go out on every response. Every request gets a correlation ID, which ties together distributed traces and makes incident investigation possible.
Rate limiting, abuse controls, and logging
Rate limiting and brute force protections cover all endpoints. Authentication endpoints have additional controls. Security events are logged centrally.
Infrastructure
The platform runs on Google Cloud Run, which has no persistent compute surface between requests. Container images run without elevated privileges. There are no long-lived server processes to compromise.

Cryptographic scanner identity

The scanner signs every outbound request. The public key and scanner policy are at /.well-known/agent-identity.json. Any site operator can verify a request came from us rather than something claiming to be us. Signing keys rotate via GCP Secret Manager. Every request carries a signed, stable identity.

Agent-Idlighthouse.aido-labs.co
Agent-Purposeproduct-discovery / checkout-assessment / payment-signal-detection
Signature algorithmEd25519
Current key ID2026-05
Request rate≤ 1 request/second per domain
Pages per scan≤ 50 pages per domain per scan
Concurrent scans1 per domain
robots.txt / noindexRespected always
Form submission / purchasesNever

To block the scanner, add this to your robots.txt. The scanner respects it without exception.

User-agent: AidoLighthouse Disallow: /

To remove your site from the scan queue entirely, email info@aido-labs.co. The scanner runs on Google Cloud Run in us-central1 (Iowa); egress IPs come from Google's published Cloud ranges at gstatic.com/ipranges/cloud.json, filtered to scope: us-central1.

What data is collected

Page HTML, HTTP response headers, and structured data (JSON-LD, schema.org markup); protocol endpoint responses (robots.txt, llms.txt, /.well-known/* endpoints, openapi.json); payment method signals detected from public page content; bot detection signals encountered during the scan (WAF provider, challenge type). No credentials, session data, or personal information is collected. Scan results are kept for 90 days and visible only to the account holder who ran the scan.

Reporting a vulnerability

If you find a security issue, email info@aido-labs.co. We aim to respond within 48 hours. No bug bounty program yet, but we will acknowledge responsible disclosure publicly on request.