Website scorecard · Method

How the Orvane website scorecard works

The scorecard reads your homepage and up to four key pages linked from it. It runs Google’s mobile speed test and asks an AI model with web search ten questions a customer might ask. It then scores four areas out of 25: speed and reliability, search foundations, content clarity and AI discovery. Every score links to the evidence behind it, and nothing is scored without evidence.

What the scorecard reads

  • Public pages only. It starts from the homepage you enter and follows links to up to four key pages, such as your services and contact pages. That makes five HTML pages at most. It doesn’t log in, fill in forms or crawl the whole site.
  • The HTML your server sends. The scanner reads page code without running JavaScript. Text that only appears after scripts run may not be seen, and the report says when pages couldn’t be read.
  • Your crawl rules. It reads robots.txt first and respects general rules and any rules for OrvaneScorecard, the name it identifies itself with. If robots.txt blocks it or can’t be read, you get a limited report instead of a guess.
  • Google’s mobile speed test. It runs PageSpeed Insights on the homepage in mobile mode. It shows Chrome’s real-visitor Core Web Vitals when Google has enough data for the page.
  • AI answers. It asks an AI model with web search ten questions and checks whether the answers mention, cite or recommend you.

How the score is built

The total is out of 100: four areas, each worth 25 points. A total appears only when every area finished. If a required check fails to run, such as Google’s speed test timing out, we withhold the total rather than show a misleading number. The areas that did finish are still shown.

Scoring rubric growth-1
Area (25 points)ChecksHow each check scores
Speed and reliabilityHomepage delivery (5), HTTPS (5), mobile performance (15)Delivery and HTTPS pass or fail. Mobile performance is the PageSpeed Insights lab score scaled to 15 points; 80 or more counts as passing.
Search foundationsGooglebot access, listing permission, page titles, canonical tags, links to key pages (5 each)Averaged across the pages reviewed; details below.
Content clarityWhat you offer, who you help, where and how you deliver, reasons to trust you, the next step (5 each)Clear 1, unclear 0.5, missing 0, each backed by a quoted passage from your site.
AI discoveryTen sampled answers (25)Recommended 1, mentioned or cited 0.5, absent 0, averaged.

Content clarity in detail

An AI model reads the text of the reviewed pages and rates five questions a new customer has: what you offer, who it is for, where and how you deliver, why they should trust you and what to do next. It rates each as clear, unclear or missing.

Each rating must quote a passage from your site, and our code checks that the quote really appears on the page before the rating is accepted. The report shows those quotes, so you can judge the rating yourself. Menus and run-together headings are never quoted. Website text is treated as data, never as instructions to the model.

AI discovery in detail

  • Questions. Five questions are written from the services and the place your own pages describe, for example “Which providers are worth shortlisting for website development in Kenya?”. They never include your business name, because customers ask about a need, not a brand.
  • Answers. Each question is asked twice, giving ten answers from one AI model with web search. Answers vary from one attempt to the next, which is why each question is asked twice.
  • Scoring. An answer that recommends you scores 1, one that mentions or cites you scores 0.5, and one that leaves you out scores 0.
  • Identity checks. A mention counts only when the answer cites your domain and the quoted text supports it, so a business with a similar name isn’t mistaken for you. A mention we can’t verify earns no points and is shown as unverified, not as a confirmed absence. Answers about a different country are excluded from the score.

This is a dated sample from one system, not a ranking and not a measure of every assistant. The report shows every question, both outcomes, and the businesses named instead. Read more in how to get recommended by AI assistants.

How fixes are chosen

A report lists up to three fixes. Problems that can stop enquiries altogether come first: a homepage that doesn’t load, or no clear next step. The rest are ordered by the scorecard points each fix is worth. Points measure this scorecard, not search rankings or sales.

Each fix shows what we found, why it matters, what to change and how to check that it worked, with links to the affected pages.

What a report can’t tell you

  • Your Google rankings, how much traffic you get or how many sales you make.
  • Whether Google has listed your pages.
  • Whether your forms deliver messages or your WhatsApp link reaches the right person. We never submit forms.
  • Anything behind a login, or pages beyond the five reviewed.
  • What will happen after a change. A better score shows the website improved on these checks; it doesn’t promise more enquiries.

Reports, privacy and cost

Your first completed report for each website is free, with no account or email. Reports stay online for 30 days, and anyone with the link can open them. Download the PDF to keep a copy.

Page text is processed by OpenAI to rate clarity and sample AI answers. Google PageSpeed Insights tests speed. Our privacy page explains what is kept and for how long. You can run the same assessment from an AI assistant or your own code through our API and MCP server.

Sources

  1. Google: About PageSpeed Insights, checked 21 November 2026
  2. web.dev: Web Vitals, checked 21 November 2026
  3. Google Search Central: Introduction to robots.txt, checked 21 November 2026
  4. Google Search Central: How to specify a canonical URL, checked 21 November 2026
WhatsApp Liplan