How the GEO score is calculated.

Why we are publishing this

Our own guides tell business owners to walk away from any agency that sells a proprietary score it will not explain. It would be dishonest to write that and then hand you a number from a black box. So here is the whole thing: the categories, the weights, the checks, the bands, and the parts where the number lies to you.

Two practical consequences. First, you can rebuild your own score without paying anyone. Second, when we tell a client their score moved, they can audit the claim line by line instead of trusting us.

The formula

Composite equals the sum of each category score multiplied by its weight, rounded down to the nearest whole number. Nothing else goes into it. There is no adjustment, no curve and no per-client tuning, because a score you can tune is a score you can flatter.

What each category actually checks

AI citability, 25%

Do your key pages answer the question in the first sentence, or bury it under a paragraph of positioning? Are prices, hours, services and locations present in the raw HTML, or injected by JavaScript and images? Does the same claim appear the same way across pages, or does the home page promise something the terms page contradicts? Are passages self-contained enough that a model can quote one without the rest of the page?

Brand authority, 20%

How many places that a model already trusts mention you at all: reviews, directories, community threads, knowledge bases, industry lists. Whether your brand name returns you in ordinary search. Whether a company entity exists anywhere other than your own website. This is the slowest category to move and the one that decides most outcomes, which is why it carries the second-highest weight even though it makes new sites, including ours, look bad.

Content and E-E-A-T, 20%

Named author with a real profile, visible dates, claims that carry a source, pages with enough substance to be worth quoting, and transparency where it costs something: pricing, guarantee terms, limitations. Thin pages and anonymous copy score low even when the writing is polished.

Technical, 15%

Server-rendered HTML, a 404 that returns 404, one canonical per page, a sitemap that matches reality, security headers, sensible speed, and a robots.txt that lets in the AI crawlers you want while excluding what you do not. Most of this is unglamorous and most sites fail at least one of it.

Schema, 10%

Valid JSON-LD that parses, an entity graph whose references resolve, types that fit the page, and, critically, structured data that matches the visible content. Schema describing things a visitor cannot see is a liability, not a signal.

Platform, 10%

Each engine is scored separately, because being visible to one says little about the others. Bing indexing matters for Microsoft Copilot, Google indexing and community signals matter for AI Overviews, and freshness plus quotable structure matter for Perplexity.

Scoring bands

What the score cannot tell you

It is a proxy. A site can score 80 and still never be named, because being quotable is not the same as being chosen. Authority, competition and how often the engines even trigger an AI answer for your queries all sit outside it.

It is also relative to itself. Comparing your score to another business tells you less than comparing your score to your own score last month, measured the same way. And it is a snapshot: the engines change, and a number from March is not a number from July.

This is why every report we send puts the score next to the measurement that actually matters, which is how often the engines name you across a frozen panel of buying prompts over multiple runs. That method is published in full in our guide on how to track whether ChatGPT recommends your business. If a score ever moves and the prompt panel does not, we say so.

We ran it on ourselves first

On July 24, 2026 this site scored 47 out of 100: citability 78, technical 83, content and E-E-A-T 40, schema 37, platform 37, and brand authority 3. Those six numbers against the weights above sum to 47.95, which rounds down to 47. We first published it as 48 and corrected ourselves the same day, which is the point of showing the arithmetic. We published the numbers before doing any of the work, and the same-day re-scores after shipping fixes came out at 62 and then 69, with brand authority moving only from 3 to 4 because nothing off-site changes in an afternoon. The whole baseline, including the parts that make us look bad, is on our work page.

Questions