Your site renders for people. Does it read for agents?

AgentPass fetches a URL the way an AI crawler would, scores what it finds on a 100-point index, and hands you the exact fixes — each one written to be pasted into a coding agent.

By Scott Scheferman · Updated November 9, 2026 · ARI-100 v1.0

Site URL Scan free

Free scan, no account. 32 named crawlers · 120 checks · 8 weighted domains · one request each.

What one scan returns

The eight domains

Weights are fixed and public. A domain is scored only over the checks that apply to your site's shape; the domain score is the weighted share of points earned, and the total is the weight-normalised sum across live domains plus at most 5 points of emerging-signal bonus.

The eight domains, their weight in the 100-point total, and the number of checks each carries.

Domain	Weight	Checks	What it reads
Reachability & Access	20	15	robots.txt policy, live reachability for 32 named AI crawlers, indexability, interstitial and edge-block detection.
Render & Delivery	10	11	Server-rendered text before scripts run, status codes, soft-404 detection, TTFB, payload weight and text-to-markup ratio.
Structure & Semantics	12	14	Title, description, a single H1, heading order, self-canonical, language, social tags and control names.
Structured Data	12	12	JSON-LD that parses, page and entity types, sameAs identity, dates, breadcrumbs and price-vs-page agreement.
Discovery Files	14	18	Sitemap, llms.txt shape, llms-full.txt, Markdown mirrors, Accept negotiation, Link headers and the AGENTS.md build guide.
Agent Interfaces	14	27	OpenAPI, RFC 9727 catalog, RFC 9728 and RFC 8414 metadata, MCP card plus a live MCP handshake, A2A and error shapes.
Operability & Commerce	10	13	Search, cart form semantics, variant controls, price in server HTML, guest checkout and policy links.
Trust & Authority	8	10	Author, visible dates, outbound sources, about/contact reachability, legal name, HTTPS hygiene and broken links.

The crawler matrix

32 crawlers across 28 platforms: 11 AI search indexers, 8 live user agents that fetch on a person's behalf, 11 training crawlers and 2 niche fetchers. 21 of them carry weight in the access score; training crawlers carry none but their policy is still read, because a site that blocks training without saying so looks hostile to every agent that reads robots.txt.

What each family is used for, and what a block actually costs you.

Family	Probed	Failure mode it exposes
AI search indexers	11	A robots group or edge rule hides the page from the index that answers questions.
Live user agents	8	A challenge page or a login wall answers the fetch that a person's assistant makes on their behalf.
Training crawlers	11	No stated training policy, so the default is guessed on your behalf.
Niche fetchers	2	Fetchers outside the big platforms cannot read the page at all.

A block is only counted when it is observable from one server. Verified-identity crawlers that answer differently to a verified IP are marked unverifiable and drop out of the denominator instead of being scored as a failure.

Named agents include OAI-SearchBot, Claude-SearchBot, PerplexityBot for search, and GPTBot, ClaudeBot, Google-Extended, Applebot-Extended for training. The report lists every one with its status code and verdict.

How ARI-100 is computed

Every check returns one of five states. A pass earns its full weight, a partial earns half, a fail earns nothing, and inconclusive or not-applicable checks leave the domain denominator entirely — so a single-page site is not punished for missing a product page.

domain_score = round(1000 * Σ(weight × points) / Σ(weight applied)) / 10
core         = Σ(domain_score × domain_weight) / Σ(domain_weight)
ARI-100      = min(100, core + min(5, emerging_signals_passed))

The bonus is capped at 5 points and cannot lift a site past 100. Grade bands run from A+ at 95 to F below 40, and an A+ also requires an accessibility sub-score of 80 or better. The full methodology lists each band, each state and the limits of a one-shot probe.

What we build on

The engine reads public conventions rather than private heuristics. Where a standard exists, the check cites it.

• RFC 9309 — Robots Exclusion Protocol
• RFC 9727 — API discovery linkset
• RFC 9728 — OAuth protected resource metadata
• llms.txt convention
• Schema.org vocabulary

Questions agents ask

What does the ARI-100 score actually measure?

It measures the affordances an AI agent needs before it can read, trust and operate a site over plain HTTP: reachable robots policy, server-rendered text, stable structure, parseable structured data, discovery files, machine interfaces and a checkout path. Each check returns pass, partial, fail, inconclusive or not applicable, and the domain score is the weight-normalised sum.

Why is a snapshot not a guarantee?

One scan is one request per surface at one moment. A CDN rule change, a bot-manager threshold or a redeploy can flip a verdict minutes later. Re-scan after every deploy, and treat the score as a regression test rather than a certificate.

How do you test crawlers you cannot verify?

Every crawler is probed with its real user-agent string and checked against your robots.txt with RFC 9309 longest-match semantics. Verified-identity crawlers answer differently to a verified IP, so a rejection for them is labelled unverifiable instead of blocked, and it leaves the access denominator rather than being scored as a failure.

Is the free tier a trial?

No. Scout stays free: one scan a day, the score, the domain table, the crawler matrix and the first three fix prompts. Paid tiers unlock the whole fix pack, higher API limits, history and a CI gate.

Can a coding agent use the report directly?

Yes. Every fix carries the check id, weight, observed evidence and a prompt written to be pasted into an agent. GET /report/<host>.md returns the same content as Markdown, and the MCP tools scan_site and get_report return JSON over one POST.