Discovery files
robots.txt, sitemap.xml, llms.txt, A2A Agent Card paths, and other agent discovery manifests.
GET /llms.txt
SEO for autonomous agents
SEO was built for search engines and human clicks.
ASO is an open framework that helps organizations navigate and implement emerging agent standards and technologies.
The next visitor who matters may not be human. If an AI shopper, browser agent, research assistant, or buying bot lands on your site tonight, will it understand what you sell, trust the source, and know what to do next? ASO gives you the score, open framework, and free self-assessment to find the gaps before agents skip you.
ASO is an open framework built on emerging agent-readiness standards, discovery conventions, and commerce protocols including A2A, MCP, x402, llms.txt, Agent Readiness, and agent discovery initiatives.
Agent Signal Optimization is the practice of optimizing for agent discovery, trust, invocation, commerce, and memory, so AI shoppers, browser agents, research assistants, and buying bots know what to find, cite, recommend, invoke, pay for, and return to.
Signal pillars
ARI max score
Launch artifacts
Readable manifests
The new traffic alarm
Product pages, booking flows, storefronts, APIs, directories, and service businesses are about to be judged by software that does not browse like a person. Agents look for explicit signals: what you do, what can be trusted, what can be invoked, what costs money, and where the canonical answer lives. If those signals are missing, the agent may skip you before a human ever sees the page.
Definition
SEO ranks pages for people. AEO, GEO, and LLMO improve answer visibility. ASO prepares services for agent selection, invocation, payment, and repeat use.
The easiest way to understand ASO is to start with the job SEO did for the web. SEO made pages legible to search engines so humans could find them. ASO makes services legible to autonomous agents so agents can choose, call, pay for, recommend, and return to them.
That shift changes the optimization target. A page can win SEO by being relevant and authoritative. A service wins ASO by being machine-readable, callable, trustworthy, priced clearly, and stable enough for an agent to remember.
Keywords, links, crawl health, content quality, and human click-through.
Retrievable explanations, citations, source authority, and brand mentions.
Manifests, callable specs, auth clarity, trust evidence, pricing, and durable memory.
SEO to ASO
ASO does not replace SEO. It extends the same basic idea into a world where the visitor may be an autonomous client instead of a human using a browser.
Some web properties carry over directly: `robots.txt` and `sitemap.xml` still matter. Others need agent-native companions. A blog post may help an answer engine cite you, but an agent needs a manifest, a callable contract, and explicit policy boundaries before it can safely use you.
Crawler access rules.
robots.txt plus AI crawler policySeparates indexing, retrieval, training, and agent access.
URL inventory for search engines.
sitemap.xml plus llms.txtPairs page inventory with a curated reading path for agents.
Human-facing result snippets.
agent.json and schema.orgMachine-readable identity, service type, owner, and capabilities.
Entity and content context.
OpenAPI, MCP, and manifestsCallable interface, parameters, auth, responses, and errors.
Authority signals.
registry listings and reputation signalsEvidence an agent can use to compare services and reduce risk.
Forms, CTAs, demos, and sales pages.
agent-safe invocation and payment pathDocumented auth, pricing, x402 or payment manifest, and constraints.
Humans remember and search again.
persistent service memoryAgents return through stable IDs, versioned docs, and consistent endpoints.
Agent Flow
A human can tolerate ambiguity. They can click around, infer intent, fill out a form, or ask sales what an API actually does. An agent has a narrower tolerance. If it cannot identify the service, verify the rules, and find a safe invocation path, it will usually choose another option.
ASO turns that fragile browsing process into a deterministic evaluation path. The agent finds public signals, resolves identity, checks trust and commerce constraints, invokes the service, then stores the service as a future candidate.
Signal Stack
A service becomes agent-readable when its public surface answers three questions without a sales call: what are you, how do I call you, and why should I trust you?
Discovery files are not the whole discipline, but they are where the agent starts. The first layer tells crawlers that a service exists. The next layers tell agents what the service does, how to call it, how risk is handled, and whether money can move without a human in the loop.
robots.txt, sitemap.xml, llms.txt, A2A Agent Card paths, and other agent discovery manifests.
GET /llms.txt
agent.json, schema.org, service type, version, owner, canonical URLs, and compatibility.
GET /agent.json
OpenAPI, MCP, auth schemes, endpoint semantics, input constraints, errors, and examples.
GET /openapi.json
HTTPS, status, provenance, rate limits, governance, uptime, and operational boundaries.
GET /.well-known/status
Pricing, payment manifests, x402 routes, registry state, versioned docs, and stable return paths.
GET /.well-known/payments
Open Framework
The six pillars keep ASO from becoming a checklist of random files. Each pillar answers a decision an autonomous agent must make before it can use a service: can I find it, understand it, trust it, pay for it, compare it, and remember it?
Can agents find the service through crawl rules, maps, manifests, and directories?
Can agents understand the service name, owner, type, version, capabilities, and canonical endpoints?
Can agents verify domain, auth, governance, status, provenance, and operational safety?
Can agents understand pricing, payment requirements, x402 endpoints, and purchase constraints?
Can agents compare uptime, completion rates, citations, directory listings, and third-party evidence as reputation signals mature?
Can agents return through stable URLs, persistent identity, versioned docs, and consistent signals?
ASO Score
The Agent Readiness Index is a 0-100 score that grades how prepared a service is for autonomous agent discovery, evaluation, invocation, payment, recommendation, and reuse.
The score is intentionally concrete. A service should not receive credit for saying it is agent-ready. It receives credit for public signals an agent, auditor, or crawler can verify.
Read the scoring rubricASO Score Badge Beta measured using ASO Audit v0.1 via @forgemeshlabs/aso-audit-mcp.
Reports generated by ASO Audit MCP with a public score and level.
Public, reproducible reports linked to a live badge.
Independent verification against the ASO framework.
ASO Trust Roadmap
ASO starts with a scan, becomes useful as a badge, and becomes valuable when reports, verification, registry records, and monitoring make the score reproducible.
Published open framework and signal stack.
Installable scanner for local MCP clients.
0-100 score mapped to ASO-0 through ASO-5.
Initial scan batch and reproducible method.
Static badges now, report-backed badges next.
Public JSON reports linked from live badges.
Directory of report-backed agent-ready services.
Independent review after the report layer matures.
Recurring checks for score drift and broken signals.
Readiness vs ASO
Agent readiness scanners are a useful first pass. They check whether agents can find a site, read it, authenticate against it, discover protocols, and understand commerce surfaces. ASO turns the same signal hunt into a selection framework: identity, trust, commerce, reputation, memory, traffic, and conversion evidence.
Readiness checks answer: can agents inspect this site without guessing?
The Agent Readiness Index answers: how exposed, credible, and commercially useful are the signals?
The six pillars answer: what should be fixed before agent traffic starts choosing winners?
Verified reports will answer: what was scanned, when, with which scanner version, and whether the result is reproducible?
Can agents technically inspect, read, authenticate, and discover the service?
Will agents understand, trust, compare, choose, pay for, and remember the service?
Are agents actually finding, citing, invoking, converting, and returning?
What agents inspect
ASO treats every public artifact as evidence. Some signals help an agent find the service. Some make the content cheaper to read. Some define access rules, authentication, callable tools, or payment terms. The score matters because agents do not reward intent; they reward surfaces they can verify.
The agent is not browsing like a person. It is reducing risk.
Publish a crawl map before the agent has to guess.
robots.txt with AI crawler rulessitemap.xml with canonical URLsLink headers for machine resourcesGive agents clean text paths instead of forcing HTML scraping.
llms.txt as the agent reading guideAccept: text/markdown/index.md fallbacks for important pagesSeparate search, grounding, training, and signed bot access.
Content-Signal directives in robots.txt/.well-known/http-message-signatures-directoryExpose APIs, tools, auth metadata, and task instructions.
/.well-known/api-catalog for API discoveryauth.md for auth instructionsHuman checkout is not enough for autonomous buying flows.
/.well-known/payments or payment manifestASO turns scattered checks into a repeatable evidence trail.
Accept: text/markdown.
A full crawler should still add automated checks for DNS-AID or
DNS-level agent discovery signals where supported, Web Bot Auth,
API Catalog, OAuth metadata, Auth.md, MCP Server Cards, A2A
Agent Cards, Agent Skills, WebMCP, and agentic commerce protocols.
A static reference site needs discovery, readable source files, manifests, status, and verification. It does not need OAuth, MCP, API Catalog, or x402 until it exposes protected resources, callable tools, APIs, or autonomous payment flows.
Prioritize robots.txt, sitemap.xml, llms.txt, Markdown pages, schema, manifests, status, and citations.
Add OpenAPI, API Catalog, auth docs, OAuth metadata, uptime, rate limits, pricing, and support paths.
Add MCP Server Card, Agent Skills, tool schemas, transport details, auth rules, versioning, and invocation examples.
Add machine-readable pricing, payment manifests, purchase limits, refund rules, x402 or other agentic commerce protocols.
Accept: text/markdown.
Implemented for the homepage with a Cloudflare Pages Function and /index.md fallback.
Content-Signal rules to robots.txt once your policy is clear.
/.well-known/api-catalog once you have callable public APIs.
ASO Scanner
This is the quick gut check for whether your site is visible to the next wave of AI-mediated traffic. Mark the public signals you already expose, get an ASO level, then fix the gaps agents are most likely to punish.
The self-assessment is intentionally transparent: every checkbox maps to a public signal an agent, auditor, or crawler can inspect. That makes the score useful for anyone whose revenue depends on being found, trusted, quoted, booked, bought, or recommended by software before a person clicks.
# Run the automated ASO scanner from any MCP client
# Claude Code
claude mcp add aso -- npx -y @forgemeshlabs/aso-audit-mcp
# Claude Desktop / Cursor / Windsurf
{ "mcpServers": { "aso": {
"command": "npx",
"args": ["-y", "@forgemeshlabs/aso-audit-mcp"] } } }
# Then ask your agent
"Scan example.com for agent readiness"
The checklist below is the manual self-assessment. The scanner runs the same framework automatically: 34 public-signal checks across the six pillars, an ASO Score, and a prioritized fix plan. Source on GitHub | npm package
Google Search still depends on SEO fundamentals for generative AI
features and ignores llms.txt. ASO keeps
llms.txt for non-Google agents and now also checks
Google's browser-agent UX guidance: semantic controls, linked
labels, stable layouts, and accessible action signals.
Namespace variants are published from the same scanner core:
@forgemeshlabs/aso-score-mcp
for ASO Score and
@forgemeshlabs/agent-readiness-mcp
for Agent Readiness.
No agent signals selected yet.
ASO Services
The service ladder follows the evidence trail: audit the public surface, document remediation, monitor signal drift, publish reproducible reports, then only introduce paid human review when there is enough public evidence to make that claim meaningful.
Manual and automated reviews across the six pillars.
Executive summary plus engineering remediation plan.
Signal drift, crawler access, manifest validity, and reputation checks.
Public scan reports with score, level, scanner version, scan date, and methodology.
Future paid review with re-scan, claim verification, and renewal.
Badge Roadmap
Today ASO publishes scores and maturity levels, not seals. A score badge should tell agents and humans what was measured, when it was measured, which scanner produced the result, and where the public report lives.
The first useful badge is simple: ASO Score, level, scan date, and a link to the report. The report should include score, level, scanner version, methodology, and either a signed or hashed JSON artifact. That creates evidence without overclaiming authority.
Verified status can come later when reports are public and reproducible. Paid human review should wait until it includes re-scan, verification of claims, and annual renewal. The score becomes the product first; the badge becomes valuable only after the score is recognized.
Anyone can run the scanner and publish an ASO Score with a maturity level.
The report is public, reproducible, timestamped, and linked from a badge or well-known JSON file.
A human-reviewed audit verifies the claims, confirms fixes, re-scans the site, and sets a renewal window.
Score and level only. No seal and no implied third-party endorsement.
Public report exists, score is reproducible, and the badge links to the exact scan result.
Paid tiers only after human review, re-scan, claim verification, and renewal are real.
Reference Position