AI agents read your site before your customers do. Find out what they can actually use.
Start with the free readiness scan — what an agent can reach, read and act on, with the exact fixes. Then the evidence: whether a live assistant cites your domain, and how far a buying agent gets toward your checkout.
Free · no signup · See an example report
- 28 pass
- 7 fail
- 2 not measured
- Crawler access100
- Structured data88
- Readable without JS100
- Buy button in the HTML100
- Performance90
- failText-to-markup ratio
If the weight is markup, the usual cause is deeply nested wrappers, repeated navigation and ad slots: flatten the structure and move presentation into stylesheets.
- failSchema type coverage
Align JSON-LD type to page purpose.
- failCommerce protocol manifest
If you sell to agents, publish the discovery manifest for the commerce protocol you support: /.well-known/x402.json (x402), /.well-known/ucp (UCP), or /.well-known/acp.json (ACP).
37 checks · 26 deterministic · 10 heuristic · 1 llm-assisted
Scored against the agents that fetch your pages to answer questions
We track 24 agents in all. These 9 move your score; the other 15 — 10 training crawlers and 5 regional engines — are reported to you and deliberately not charged against it. Blocking a training crawler is a licensing decision, not a readiness defect. Every agent, and which tier it is in →
After the AI finds you
Being mentioned is the easy half.
Visibility tools tell you an assistant said your name. They cannot tell you what happened when it arrived. These are real failures the free scan detects, and what each one costs.
Your structured data says $49. Your page says $59.
The agent reads the machine-readable price and quotes a number you will not honour. The buyer finds out at checkout.
coherence.jsonld_price_matches_visibleYour buy button is painted by JavaScript after load.
An agent that reads the HTML your server sends never sees a button to press, so the purchase path ends on arrival.
operability.primary_action_in_domYour product copy only appears once the page boots.
The agent summarises an empty shell. Whatever it tells the buyer about you, it did not get from your page.
content.js_required_for_primary_contentA consent or login wall stands in front of the content.
The agent hits the wall instead of your offer, and reports back on the wall.
auth.paywall_or_login_wall_on_landingYour JSON-LD is present but does not validate.
The one part of your page written for machines is the part they discard, so the agent falls back to guessing from marketing copy.
structured_data.validatesYour filters and variants have no addressable URL.
An agent asked for "the blue one in medium" cannot construct a link to it, so it cannot get the buyer to the thing they asked for.
operability.filters_addressable
The whole agent path, in the order it fails
Do agents find and choose you?
The scan grades what discovery depends on: robots policy per agent, structured data, the well-known files agents pre-flight. Then, from your report, a live assistant is asked a real buying question — and you see whether it cited your domain, saw it and chose someone else, or never found it.
Scan my site →Can they buy from you?
Three gates — Find, Read, Buy — answered free from the same scan. The Deep Scan goes further: a read-only, five-step purchase walk from your homepage to checkout that shows the exact step a buying agent loses.
Check my store, free →Will you know the day it breaks?
Sites drift — a robots edit, a CDN default, a template swap — and nothing in your analytics fires. Watches re-run the same rubric on your schedule and alert you the week anything meaningful changes.
See plans →The 0–100 readiness score under all three is the diagnosis layer: it tells you which fix moves which answer.
AI agents are shopping. Can they buy from you?
Three gates, in order. Most stores clear the first two and lose the sale at the third, and nothing in your analytics fires when it happens.
Research preview — inferred from your scan, labeled as such.
Structure is a rendering choice. Agents get the other one.
This is our own homepage as a person reads it, and as an agent receives it with JavaScript switched off. Yours is one scan away.
Free 30-second scan of what AI agents can reach, read and act on — then the evidence: whether a live assistant cites you, how far a buying agent gets, and alerts the week anything breaks.
- AI agents read your site before your customers do. Find out what they can actually use.
- Being mentioned is the easy half.
- The whole agent path, in the order it fails
- Do agents find and choose you?
- Can they buy from you?
agentspeed.Free scanCan agents buy?PricingDocsSign inScan my site →Free scan · no signup · ~30 secondsAI agents read your site before your customers do. Find out what they can actually use.Start with the free readiness scan — what an agent can reach, read and act on, with the exact fixes. Then the evidence: whether a live assistant cites your domain, and how far a buying agent gets toward your checkout.Scan my site →Free · no signup · See an example reportSites running AgentSpeed · scored live, badge on the pageWebsiteDown.comPopulationTracker.ioSolarNaperville.comStatusDetector.comOur own properties, scored by the same public rubric as everyone else — including the scores we would rather we…
Both readings from the scan of 2026-09-02 — re-taken daily.
How agents actually fail on real sites
Detected on public scans of 828 domains in the last 30 days. Every one looks fine in a browser.
Five steps, and the five ways they break
Every check in the rubric attaches to one of these five steps, and every failure is one of them going wrong on a page that looks perfect to you.
Run any step on its own, free, or run all five in one scan.
- Actionability8 checks5 emerging28%
- Discoverability11 checks6 emerging22%
- Readability9 checks2 emerging22%
- Performance3 checks—15%
- Structured data5 checks2 emerging13%
15 of the 36 test agent-native standards whose adoption is still emerging — every one scored, every one listed at /rubric.
37 checks, 36 of them scored. The one we report without scoring is llms.txt, because Google says Search does not use it and a large domain study found no link to AI citations.
Read the full spec, every weight and threshold →
PageSpeed measures humans. Keep it. It measures real things, just not these. See the comparison →
Same site, a different reader
Agent traffic is the fastest-growing source a site has: Adobe put AI-referred visits to US retail up 393% year over year in Q1 2026, across more than a trillion visits.
| PageSpeed, Lighthouse | AgentSpeed | |
|---|---|---|
| What it measures | Loading speed, accessibility, SEO | Whether an agent can read, navigate and transact |
| Who the user is | A browser on a phone | GPTBot, ClaudeBot, PerplexityBot, autonomous agents |
| AI crawler policy | Not measured | 24 agents tracked, verdict per agent |
| Cookie and login walls | Not measured | Vendor fingerprinted, verdict per barrier |
Scan → Monitor → Investigate → Certify: the same rubric at every stage
Know whether AI agents can read you and buy from you — not just find you — and get told the day that changes. Every stage scores against the same published rubric, so upgrading never resets what you learned.
Paste a URL, get the graded report: what agents can reach, read and act on, with the exact fixes.
Scan my site →Scheduled re-checks with alerts on meaningful change. Sites drift — a robots edit, a CDN default, a template swap — and nothing in your analytics fires when it happens.
See plans →When something needs proof: the Deep Scan fix-list report, the agent Transaction Audit, the AI Market Report. One-time purchases, no subscription.
Order a Deep Scan →A signed, machine-verifiable attestation with its own weekly monitoring — revoked if your site stops earning it, which is what makes it worth showing.
How certification works →Building agents rather than a site? The telemetry line — runs, cost, journey canaries, alerts, free for 10,000 events a month, metadata-only by design — lives at /observability, with a working demo.
The first three stages, side by side
| 1 · Scan Free | 2 · Monitor Popular | 3 · Investigate | |
|---|---|---|---|
| Price | Free | From $19/mo | $29 once |
| Fixes shown | Top 6 | Every finding, plus diffs | Every finding, step by step |
| Speed | ~30 seconds | Hourly, daily or weekly | ~1 to 2 minutes |
| Alerts on regression | No | Slack, webhook, email | No |
- ✓ 1 scheduled watch · weekly
- — Competitor benchmarking on paid plans
- — AI visibility monitors on paid plans
- ✓ AI crawler traffic analytics
- ✓ 10,000 agent events/mo
- ✓ 1 project · 3 agents
- ✓ 1 journey canary · manual runs
- — Alert channels on paid plans
- ✓ 1,000 evaluations/mo (API · beta)
- ✓ 7-day retention
- ✓ 3 scheduled watches
- ✓ 3 competitors benchmarked
- ✓ 1 AI visibility monitor · 300 prompt runs/mo across Claude
- ✓ AI crawler traffic analytics
- ✓ 100,000 agent events/mo
- ✓ 3 projects · 10 agents
- ✓ 10 journey canaries · scheduled + alerts
- ✓ Slack / webhook / email alerts
- ✓ 25,000 evaluations/mo (API · beta)
- ✓ 30-day retention
- ✓ 10 scheduled watches
- ✓ 10 competitors benchmarked
- ✓ 3 AI visibility monitors · 3,000 prompt runs/mo across Claude
- ✓ AI crawler traffic analytics
- ✓ 1,000,000 agent events/mo
- ✓ 10 projects · 50 agents
- ✓ 50 journey canaries · scheduled + alerts
- ✓ Slack / webhook / email alerts
- ✓ 250,000 evaluations/mo (API · beta)
- ✓ 90-day retention
One event = one ingested row. Full pricing and FAQ →