Will ChatGPT recommend your site? Check the 12 things it reads.
The audit reads your public pages without running JavaScript, as many AI crawlers do. Each of the 12 checks comes back pass, partial or fail, with the page it looked at and the fix.
MCP: seoaudit.openkrill.app/mcp12 checksup to 10 pages per run
We ran it on our own site first
tools.openkrill.app scored 6.5 of 12 on 2026-10-19. The audit's top 3 fixes for us: compare Agent Tools with 3 named alternatives in one table, add customer quotes that name the person, and publish one number we measured ourselves with its date.
| # | Check | What earns a pass | Our site |
|---|---|---|---|
| 1 | Readable without JavaScript | The homepage HTML holds at least 150 words before any script runs. | pass |
| 2 | Crawlers allowed, homepage indexable | robots.txt lets Googlebot, GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot in, and the homepage answers 200 without noindex. | pass |
| 3 | llms.txt | A text file at /llms.txt with a title and links to the main pages. | pass |
| 4 | What-is sentence at the top | The first 60 words name the product and say what it is. | partial |
| 5 | A page with your prices | A prices page linked from the homepage, with an amount or the word free. | partial |
| 6 | Comparison page | A linked page that compares 3 or more named products. | fail |
| 7 | 10 real questions answered | 10 or more distinct questions in headings, FAQ markup or expandable answers. | partial |
| 8 | Organization and product markup | JSON-LD with a named Organization and a named Product or app. | pass |
| 9 | One name everywhere | The title, og:site_name, the markup and llms.txt use the same name. | partial |
| 10 | Named customer quotes | 2 or more quotes that credit a named person. | fail |
| 11 | Last-updated date on each page | Every page read states when it was updated. | partial |
| 12 | One original number | A number you measured yourself, with a date. | fail |
How it grades
A pass is 1 point and a partial is half a point, out of 12. A row is only graded from what the fetch proved. If robots.txt blocks a page the row needs, or a page is too long to read to the end, the row says unknown and earns nothing.
It does not run JavaScript, query AI engines or measure rankings. The crawler row follows Google Search Essentials. The other eleven, the JavaScript row among them, are conventions from a public checklist, and the answer marks them that way.
Run it on your site
Connect the free SEO Audit MCP server, then ask: Will ChatGPT recommend example.com? Run an AI search audit.
The agent calls audit_ai_search and gets the score, the 3 biggest fixes and all 12 rows with evidence. Step-by-step guides: Claude, ChatGPT, Cursor.
Claude Code
claude mcp add --transport http seo-audit https://seoaudit.openkrill.app/mcp
Cursor
{
"mcpServers": {
"seo-audit": {
"url": "https://seoaudit.openkrill.app/mcp"
}
}
}Any MCP client
https://seoaudit.openkrill.app/mcp
Frequently asked questions
Does it ask ChatGPT or Perplexity about my site?
No. It reads your own public pages the way a crawler that does not run JavaScript sees them. It never queries an AI engine, so a pass is not a promise that any assistant will cite or recommend you.
What does unknown mean?
The audit could not prove the row. For example, robots.txt keeps the tool off your prices page, or a page is too long to read to the end. An unknown row earns no point, and the evidence says why. It never counts as a pass.
Which pages does it read?
robots.txt, the homepage, /llms.txt, then the prices, comparison and FAQ pages the homepage links to, and other linked pages, up to 10 pages in all. robots.txt is respected for every one.
Where do the 12 checks come from?
From a public checklist of what makes a site easy for AI answer engines to read and recommend, posted on X by Suraj Sharma (x.com/suraj_sharma14/status/2112000783797641373, read on 2026-10-19). The crawler row also follows Google Search Essentials. Each row in the answer says whether it is a requirement or a convention.
Is it free?
Yes. No key and no sign-up. Answers for the same site are cached for ten minutes, and nothing about you or the site is stored.