Skip to content
agentoray
Scan your store

agent readiness score

How do AI agents see your store?

Shopping answers get written from whatever an agent can read in twenty seconds. We read your storefront the same way, show you exactly what they miss, and price every fix so you know which one is worth doing first.

https://

Free, no account. One scan takes about 20 seconds.

scanned asChatGPTPerplexityClaudeGeminiCopilot

What the scan actually does

No agent tells you why it skipped your store. This is the closest thing: we walk the same path and write down where it breaks.

  1. 01

    We fetch like an agent

    An honest user agent, robots.txt, your sitemap and product feeds — the same doors ChatGPT and Perplexity knock on. Capped at 25 requests, one store at a time.

  2. 02

    We read what they'd read

    JSON-LD and OpenGraph on real product pages, the page with JavaScript switched off, your descriptions, FAQ, shipping and returns policies.

  3. 03

    You get a score and a fix list

    One 0–100 Agent Readiness Score, every finding backed by the evidence we captured, and the fixes ordered by what actually costs you visibility.

22 checks, five categories

Each check returns pass, warn, fail — or not applicable, which never costs you points. Category weights are fixed, so two stores are always comparable.

weighted 30 / 30 / 15 / 15 / 10

Access

30 pts

Can an agent reach the store at all, and is it allowed to?

  • Homepage responds
  • Served over HTTPS
  • Redirect chain is short
  • Time to first byte
  • Homepage HTML weight
  • robots.txt reachable
  • AI agents allowed in robots.txt
  • No bot wall for AI agents
  • Sitemap present and valid

Feeds & discovery

10 pts

Can an agent enumerate the catalog instead of guessing at it?

  • Product pages discoverable
  • Shopify product feed (/products.json)
  • WooCommerce Store API products
  • RSS/Atom feed
  • Agent guidance file (llms.txt / agents.md)

Structure

30 pts

Is product data machine-readable once the agent lands on a page?

  • Product structured data (JSON-LD)
  • Product schema richness (GTIN/SKU/brand/rating)
  • OpenGraph fallback on product pages
  • Product content visible without JavaScript

Content

15 pts

Is there text worth quoting when an agent recommends you?

  • Product descriptions agents can quote
  • FAQ content for agent Q&A

Policies

15 pts

Shipping and returns — the questions agents get asked constantly.

  • Shipping policy agents can read
  • Returns/refund policy agents can read

What you get back

A link you can send to a developer or an agency — score, category breakdown, and every finding with the evidence behind it, what fixing it is worth, and roughly how long it takes.

sample
agent readiness

northline-supply.com

Reachable and fast, but its catalog is invisible to agents that don't run JavaScript.

Access26.5 / 30
Structure15 / 30
Content7.5 / 15
Policies15 / 15
Feeds & discovery3 / 10

67→89 if every finding below is fixed — roughly 11 hours of work in total.

  • high

    Product content visible without JavaScript

    0 of 3 product pages showed name, price or description with JavaScript off.

    project · 8h

    +7.5
  • high

    Product pages discoverable

    No product URLs in the sitemap; found 4 only by crawling the homepage.

    moderate · 2h

    +4.0
  • medium

    AI agents allowed in robots.txt

    No AI agent is named — all 22 fall through to the wildcard group.

    quick · 10m

    +2.5

A score is where it starts, not where it ends

The scan takes twenty seconds. Everything after it is the part that changes the number — a priced plan of work, proof that a fix landed, the stores you compete with, and what an assistant actually recommends today.

signed in

plan

Every fix is priced in points

We flip each failing check to passing and re-run the real scoring engine, so the plan can tell you what a fix is worth and where it lands the score. Ordered by points per hour of work — which is how a ten-minute robots.txt edit ends up above an eight-hour rebuild.

68 → 100 · about 12h 40m of work

verify

Mark it fixed. We check.

Findings become work you can track. You say fixed; only a scan says verified. That gap is where the useful sentence lives — you marked this done on Tuesday and we still see it failing — and it means you never have to wonder whether a change actually deployed.

3 fixes verified in the last 30 days

compare

The same 22 checks, on your competitors

Add the stores you actually lose customers to. They get real scans, so the comparison names the exact checks they pass and you don't — including the ones every single rival has and you don't. Monitored stores refresh their rivals on the same weekly schedule.

rival.com beats you on 5 checks

visibility

What an assistant says when nobody prompts it

Every other check infers what agents think. This asks one: real shopper questions, built from your own catalog and never naming your store, put to a model — which shops it recommends, in what order, and who it names instead of you. Answers kept verbatim.

named in 2 of 5 answers · #2

Scanning and the report stay free and need no account. The workspace above needs one — it has to remember which store is yours, and what you have already fixed.

Pricing

Scanning is free and stays free, and so is the plan that comes out of it. Watching — the weekly re-scans of you and everyone you compete with — is the paid part, because it is the part that costs us something every week.

Free

$0

Audit any store, as often as you need it.

  • Every check, every time — no partial free report
  • Shareable report link, with every fix priced in points
  • An action plan ordered by return on effort
  • Track up to 3 stores, with history and a work queue
Scan a store now

Pronot live yet

$29/mo

Everything in Free, plus watching — yours and theirs.

  • Weekly re-scans of every store you track, and of their competitors
  • Email, Slack or webhook alerts when something moves
  • Competitor comparison — the exact checks they pass and you don't
  • Where you rank against every store we have ever measured
  • 50 stores instead of 3

Monitoring works today, but there is no checkout yet — billing is the next thing we ship. Run a free scan and you'll see the report Pro would be watching.

Questions people ask first

If yours isn't here, ask us — a person answers.

Is it really free?
Scanning and the report are free, with no account and no card. There is an hourly limit per address so one person cannot monopolise the queue. Paid plans add weekly monitoring, which re-scans a store on a schedule and emails you when something changes.
What do you actually fetch from my store?
Your homepage, robots.txt and the sitemaps it points to, up to five product pages, a product feed if your platform publishes one, your shipping and returns pages when they are linked from the footer, and llms.txt or agents.md if they exist. At most 25 requests, then it stops. We never sign in, submit a form, or touch a checkout.
Will it slow my store down?
No. A scan is about twenty requests spread over roughly twenty seconds, from one address, with a ten-second timeout and a single polite backoff if you rate-limit us. That is less traffic than one shopper browsing. Our fetcher identifies itself as AgentorayBot so you can see it in your logs, and blocking it is one rule at your CDN.
Does it work with my platform?
Yes. Shopify and WooCommerce get platform-specific fix instructions with copyable snippets; BigCommerce, Magento, Salesforce Commerce, Squarespace, Wix and fully custom storefronts are detected and scored the same way, with generic instructions. The checks are about what a machine can read, not about which platform rendered it.
How is this different from an SEO audit?
An SEO audit optimises for a search engine that sends a human to your page. This optimises for an agent that reads your page and answers on your behalf, often without the shopper ever seeing your site. That changes what matters: complete Product JSON-LD, prices and availability visible without JavaScript, discoverable policies, and whether AI crawlers are allowed at all — which most default robots.txt files quietly do not address.
What is the score actually measuring?
Twenty-two checks across four groups — can agents reach the store, can they find the products, is the product data machine-readable, and is the surrounding content complete — weighted so the things that block an agent entirely cost the most. Every check shows the evidence behind its result, so you can disagree with any single one and see exactly why we said it.
It says fix this first — how do you know?
We work it out rather than guess. The scoring engine is deterministic, so we re-run it with each failing check flipped to passing and read the difference — that is what a fix is worth, and the projected score is exactly what a re-scan would report. The plan then orders by points recovered per hour of work, using a hands-on time estimate written for each check, which is why a ten-minute robots.txt edit sits above an eight-hour rebuild worth more points.
How do I know a fix actually worked?
You mark a finding fixed and the next scan grades it. We deliberately keep those as two different states: you can say fixed, but only a scan says verified. If the scan disagrees, the store page says so with the date you marked it — which usually means the change never deployed. A check that passes without anyone claiming it is closed anyway, because stores get fixed by people who never open a dashboard.
Can I compare myself to a competitor?
Yes. Add the stores you actually lose customers to and they get real scans through the same 22 checks, so the comparison names the exact checks they pass and you do not — and flags the ones every rival has and you are missing. Nothing is sent to them and they are never told. On a monitored store their scans refresh weekly alongside yours, so the comparison is never against a six-month-old snapshot.
Do you actually ask ChatGPT about my store?
On plans where it is switched on, yes — and carefully. We build shopper questions from your own product catalog and never name your store in them, because a question that names you guarantees a mention and measures nothing. We record which shops the model recommends, in what order, and keep the answer text verbatim so you can check the claim. It is one model, one sample, one moment, and the page says so — the trend over several runs is worth more than any single answer.
Can I share the report?
That is what it is for. Every report has a public link with its own share card, and it stays live so you can send it to a developer or a client. Reports are not listed anywhere and there is no directory of scanned stores — the link is the only way in, and you can ask us to take one down.

Find out in twenty seconds

No account, no card, no crawl of your customers' data — just your public storefront, read the way a machine reads it.

https://

Free, no account. One scan takes about 20 seconds.