Readable is now in the menu, and its code is on npm

Since 17 October, Readable sits in the mador.ai menu, under How it works, and in the footer of every page. It is the free check that tells you how much of a site assistants such as ChatGPT, Claude, Perplexity and Gemini can read. You paste the address and within a few seconds you see whether AI gets in, and how much text it finds. It does not ask for your email.

Two different faults

An assistant can miss a site for two reasons, and Readable keeps them apart because they are fixed in two different places.

In the first, the assistant cannot get in. The robots.txt, the file where a site says who may read it, tells it to stay out, or a protection like Cloudflare's answers automated readers with a challenge page.

In the second, it gets in and finds an empty page. The text is written by JavaScript inside the browser, and most assistant readers do not run JavaScript. Whoever built the site opens it in their own browser and sees it full, so from the inside nothing looks wrong.

If you build sites with Lovable, Bolt, v0, Wix or WordPress

The second fault is common on sites made by the builders that write the site for you, such as Lovable, Bolt and v0. On 30 September Mador opened 18 sites of this kind, and to AI 13 of them were a blank page. The full story is at mador.ai/en/blog/tredici-siti-su-diciotto-pagina-bianca.

Readable recognises what the site is built with, and the advice changes to match. On WordPress it tells you where to look when a plugin writes the robots.txt, or when a box in the settings asks not to be found. On Wix, on Lovable and on the others it tells you what to ask for or change in that place.

After the check, the scan

Below the result there is Mador's free scan. It puts your area's questions to ChatGPT and Perplexity and tells you whether they name you, and who they name instead. There the email is needed, because the report arrives by email. A site that can be read is the condition for showing up in answers, and the scan tells you whether it actually happens.

The code is open

The check runs on readable-by-ai, an open source program under the MIT licence. The code is on GitHub, and since 16 October it is on npm too: anyone who builds sites can run it from the terminal by typing npx readable-by-ai followed by the address, and put it inside their own tools.

The reader introduces itself as what it is, Mador's reader, and does not pose as GPTBot or any other assistant reader. Large networks recognise real readers by where they come from, so a disguise would be stopped anyway and the result would be false. When it meets a challenge page, Readable says the site may stop assistants too, which from the outside is all an honest check can say.