How the Inspection Works
QuickRankAI makes up to four small requests, runs every check against what came back, and reports only things it actually observed. There is no model guessing at what an algorithm might prefer, and there is no staged progress bar pretending to do work it is not doing.
1. The URL is validated before anything is fetched
The hostname is resolved and every address it answers with is checked against private, loopback, link-local, and reserved ranges. Credentials in the URL are refused. Non-standard ports are refused. Schemes other than http and https are refused.
Then the connection is pinned to exactly the addresses that passed, so the name cannot resolve to something different a moment later. Redirects are followed one hop at a time and each hop gets the same treatment.
This is not paranoia for its own sake. A tool that fetches arbitrary URLs from a server is a way into that server's network if nobody thought about it, and plenty of people did not.
2. The page is fetched once, as raw HTML
One request, up to 1.5 MB, with a nine second ceiling. No JavaScript is executed. What comes back is what an AI retriever gets, and that is the point.
Then up to three small companion requests: robots.txt, llms.txt or llm.txt, and sitemap.xml if robots.txt did not already declare one.
3. Thirty-eight checks run against what came back
Each check is a plain observation with a weight. Does a canonical exist. How many H1 elements are there. Is there a noindex directive in the headers or the markup. How many words survived when the markup was stripped. Is there an Organization node in the structured data.
Nothing here is inferred from a model. Every finding names the evidence it is based on, so you can go and look at the same thing.
4. Each finding is classified
| Class | What it means | Score |
|---|---|---|
| Running clean | The signal is present and correct | Full weight |
| Quick fix | Repairable at page, template, or configuration level | No weight |
| Investigate | Not enough external evidence to call it either way | Half weight |
| Structural | The cause sits above the page | No weight |
The half weight on an investigate finding is deliberate. Scoring an unknown as a pass would hide it. Scoring it as a failure would punish sites that are doing the thing correctly. Half is the honest position.
5. The recommendation is deterministic
Nothing weights the recommendation toward the biggest thing on offer. The rules are fixed and worth stating plainly:
- Three or more structural findings, or one structural finding on a page scoring under 55, recommends looking at the build.
- Any structural finding, or weak entity and AI readiness scores together, or a discovery score under 55, recommends working on site-wide signals rather than this page.
- A score of 90 or better with two or fewer quick fixes recommends monitoring, because there is nothing useful left to sell you.
- Everything else is a tune-up.
What is deliberately not measured
Rankings, traffic, backlinks, keyword difficulty, and whether an AI system has ever cited you. None of those are visible in a page, and a tool that claims to read them from your markup is telling you something untrue.
Rendered performance is also out of scope. QuickRankAI measures one thing about speed: how long the server took to answer, because past a certain point that is a hosting problem worth naming. Everything else about performance belongs in a tool built for it.