Content agents can't see
Prices, opening hours, plans, and product or service details that only appear after JavaScript runs are invisible to many agents.
Free AgentScore test
People now ask AI agents to research, compare, book, sign up and buy for them, on every kind of website. AgentScore shows you how they're let in, what they can read, and what gets in their way. It's free and takes under a minute.
Opening soon
AgentScore is in early access. Leave your email and we'll run the test and send you the report as soon as your spot opens.
You're on the list. We'll email your AgentScore as soon as it's ready.
From the team behind Ghost Inspector, the automated website testing service.
Checks how your site treats ChatGPT, Claude, Perplexity, and Google agents.
The problem
Agents read pages differently from people. A site can look perfect in a browser and still fail every agent that visits, and nobody notices because nobody tests for it.
Prices, opening hours, plans, and product or service details that only appear after JavaScript runs are invisible to many agents.
Icons without labels, clickable boxes that aren't real buttons, and pop-ups with no close button stop agents cold.
Bot protection and robots.txt rules often turn away the AI assistants and agents you actually want.
How it works
No account and no code to install. We only use your public pages.
We request your pages as ChatGPT, Claude, Perplexity and Google agents and as a browser, compare what each gets, and check how your pages read and work for agents, from the home page to your pricing, product or cart pages.
See your score from 0 to 100, what gets in agents' way, and what to change, in order of impact.
What we check
Can agents get in?
Can agents understand your pages?
Can agents find their way around?
Can an agent actually get the job done?
Any kind of website
AgentScore works out whether it's looking at an online store, a software company or another kind of site, and only scores the checks that apply. A hotel, clinic, publisher or law firm isn't marked down for having no cart.
Everything below, including product data, prices, the cart, guest checkout and whether agents are blocked at checkout.
Whether agents can open and read your pricing page, and find and press your sign-up or trial button.
Services, travel, local businesses, publishers, nonprofits and more: access, bot protection, content without JavaScript, structured data, named buttons, labelled forms and pop-ups.
Who it's for
Find out whether AI agents can buy, book, sign up or get in touch on your site, before they send your customers to a competitor that works.
Make sure agents can reach your pricing, services, sign-up and contact pages as search moves into AI answers.
Get specific, reproducible findings with evidence, not vague advice. Rescan to prove each fix worked.
FAQ
Yes. It's free and needs no account. Enter your website and your work email at that website's domain, and you'll see your score, every fix, and get the full report by email to share with your team. To scan other sites, such as clients' or competitors', start a trial.
No. The scanner reads public pages. The only things it presses are your add-to-cart button, once, so it can see your cart and checkout the way a shopper would, and a cookie banner's close or accept button if one covers it. It never submits forms, payments, orders, account sign-ups, or messages, and it never types personal data, so nothing is ever ordered. To test complete journeys such as checkout, use Ghost Agents, which stop before payment.
Not exactly. AgentScore requests your pages with the user agents of ChatGPT, Claude, Perplexity, and Google, opens them in a real browser, and checks them the way today's agents read and use pages. Real products differ in the details, so treat the score as a strong signal, not a guarantee.
Access and Readability are worth 25 points each, Navigability 20, and Task completion 30. Most checks are rule-based, so the same site gets a stable score. Anything we couldn't test is marked “not tested” and doesn't count against you: the score is worked out from the categories we could test. Task completion isn't part of the free scan yet, so today the score covers Access, Readability, and Navigability. How AgentScore works has the details.
It's a rough estimate built on planning assumptions, not on your own data. We assume AI agents (assistants, AI search and agents that shop) influence about 6.8% of online revenue, and that fixing everything a scan finds lifts their conversion by about 10%. We scale that 10% by the share of AgentScore points your site is missing, so a perfect score puts nothing at risk, and split the total across the findings by the points each one costs. Your own figure depends on your audience and how much of your traffic already comes from AI. To measure it, connect your orders in Ghost Agent Lab and it reports the revenue AI actually sends you. To try other numbers, use the value calculator.
If your site is an online store, against our study of 255 online stores, scanned on October 20, 2026. The median store scored 80, and 51% scored 80 or more. We only compare scans made with the same scoring version as the study, because scores from different versions aren't comparable. Other kinds of site don't get a comparison yet, because we haven't published a study of them.
After every fix, and whenever you change your theme, checkout, sign-up or booking flow, or bot protection. To have agents test your key journeys every day and alert you when they break, use Ghost Agents in Ghost Agent Lab.
See the AgentScore bot page for how our scanner identifies itself and how to opt out.
Free. Under a minute. No account needed.