AI Exorcist Threshold Test
Can AI assistants actually open your website?
When someone pastes your link into ChatGPT, Claude or Perplexity, the assistant sends its own visitor to fetch the page. The Threshold Test checks whether it gets in, and what to change if it doesn't.
One-off test free if you've claimed your AI Exorcist profile · Threshold Watch $30/year
Request a Threshold Test
- 01We review your request and confirm whether it's free (claimed profile) or paid.
- 02We test your key pages as each assistant.
- 03We email you a private link to your report.
In summary
- Security settings often turn away AI assistants that fetch a page because a real person asked them to.
- Nobody tells you: the person just gets a vague or wrong answer about your business.
- We check each assistant against your key pages and tell you what to change for your setup.
The problem nobody sees
Your security is doing its job, too well. Firewalls, CDN bot protection and WordPress security plugins are built to stop scrapers. Many can't tell those apart from an AI assistant fetching your page because a real person asked it to.
Nobody tells you. When an assistant is blocked, you get no error and no alert. The person asking gets a vague answer, an answer pieced together from other sites, or "I couldn't access that page."
A clean robots.txt doesn't settle it. Blocks usually happen at the firewall, before robots.txt is ever read. A site can score well on our AIO audit and still be unreadable to assistants.
What we test
| Assistant | Visitor we test as |
|---|---|
| ChatGPT | ChatGPT-User (OpenAI) |
| Claude | Claude-User (Anthropic) |
| Perplexity | Perplexity-User (Perplexity) |
We run each one against your key pages: homepage, pricing, product and docs. For every page and assistant you get one of three results, Got in, Turned away or Can't tell, with the reason in plain words. "Can't tell" means our probe server was turned away too, so the block isn't aimed at AI assistants.
How it works
- 01Knock as each assistant. We request your pages the way each assistant identifies itself, and compare that with an ordinary visit to the same page.
- 02Ask the assistant itself. For ChatGPT and Claude, we also use OpenAI's and Anthropic's own APIs to ask the assistant to read one key page (pricing, or your homepage). Those requests come from the assistant's real fetcher on the vendor's own network, so firewalls see exactly what they see when a person asks. We check the answer against what's actually on the page. Perplexity has no API that reads a given page live, so Perplexity gets step 1 only.
- 03Get the fix for your setup. Short instructions for your stack: Cloudflare, WordPress security plugins, Vercel, Netlify or Nginx.
What you get
- A private report link that isn't published on our registry.
- For each page and assistant: pass or fail, the likely cause (firewall block, bot challenge, rate limit, login wall, empty page) and the fix.
- A free re-test after you change your settings.
Threshold Watch · $30 / year
Know the day an assistant gets locked out
Settings change, plugins update, and CDNs roll out new bot rules. With Threshold Watch we re-test your site every month and email you when an assistant that used to get in starts getting turned away. It works on its own, without the Mirror.
Honest limits
- We test the pages we sample, not every URL. A pass covers those pages only.
- Some firewalls check where a request comes from, not just its name. Step 1 catches rules based on the assistant's name. When our probe server is turned away too, we say we can't tell rather than guess, and for ChatGPT and Claude the check through OpenAI's and Anthropic's APIs decides. That check covers one key page, not every sampled page.
- A failure is strong evidence, not proof that every assistant request gets blocked.
- We never try to get past a CAPTCHA or bot challenge. If we hit one, that's the finding.
Threshold Test FAQ
How do you test the real assistants?
For ChatGPT and Claude we ask the assistant, through OpenAI's and Anthropic's official APIs, to read one key page on your site. In our tests those requests came from the same fetcher, with the same identity and the same network addresses, as the ChatGPT and Claude apps. Perplexity's API doesn't read a given page live, so for Perplexity we rely on our own probe.
How much does it cost?
A one-off test is free if you've claimed your company's AI Exorcist profile, and priced on request otherwise. Threshold Watch, which re-tests every month and emails you when something changes, is $30 a year. Neither needs the Mirror.
Will this change my AIO score?
No. The Threshold Test is reported separately from the AIO score and the Ghost Test.
Will your test show up in my logs?
Yes. We send a small number of spaced-out requests, and each one carries an X-AIExorcist-Probe header so you can find it.
Do I have to let every bot in?
No. You can allow assistants that fetch a page for a person and still block crawlers that collect training data. Your report shows which is which.
Isn't robots.txt enough?
Robots.txt is a request, not a lock. Firewalls act before robots.txt is read, and assistants that fetch for a person don't all treat robots.txt the same way: in our tests Claude obeyed it and ChatGPT did not.