01 · GEO
GEO score (the main one)
Out of 100. Answers one question: will AI search cite your site? This is the core result.
Public Scoring Methodology
How GEO Scoring gives you a score
We open your website and ask a simple question: can AI search tools like ChatGPT and Perplexity find it, understand it, and mention it in an answer? A higher score means you are easier to cite.
- A higher score means AI search tools like ChatGPT and Perplexity are more likely to mention your site.
- Personal blogs, shops, and company sites have different requirements. You are not penalized for pages you should not have.
- The report also includes Agent Readiness, which asks whether an assistant can get in. It is not mixed into the GEO score.
The rules align with the Princeton-led GEO study (KDD 2024), Stanford’s citation-verifiability research (EMNLP 2023), and public generative-search guidance from Google, Microsoft, OpenAI, and China’s CAICT. See the sources
1
A report can show a few different scores
3 scores
02 · AR
Agent Readiness (separate)
Also out of 100, but it is not added into GEO. It asks whether an automated assistant can find and read your pages.
03
Other products (for comparison only)
The report may show public scores from Ora or Cloudflare so you can compare. They do not change our scoring.
2
Nine things we look at
100 pts
Things that matter more for being cited are worth more. Current version: geo-4.0+ar-1.0.
14pts
Schema Tier-1 Coverage
Does the site have a machine-readable “name card”: who you are, what you sell, what the article is. Without this, AI has a hard time knowing what the page is about.
Show the details
- Personal sites: Person + Organization + Article/BlogPosting. Product, FAQ, HowTo, and ItemList are not required; their weight moves to the remaining types.
- Commerce sites: Organization + Product/Offer + FAQ + Article + HowTo + ItemList.
- SaaS / corporate / blog: Organization plus content schemas relevant to the site (FAQ, Article, HowTo, ItemList), with non-applicable types reallocated.
6pts
Schema Attribute Completeness
Are the important fields on that name card filled in, such as name, author, or price. The more complete it is, the more confidently AI can cite you.
Show the details
- Organization or Person identity fields according to site type.
- Product attributes only for sites that sell products or offers.
- Article freshness and author fields for non-personal content sites.
- JSON-LD id references across separate script blocks and graph containers count as present — multi-script JSON-LD is treated the same as a single inline entity.
5pts
Schema Nesting Structure
Are those facts connected. For example, which company a product belongs to, or who wrote an article. Connected facts are harder for AI to mix up.
Show the details
- Product to Organization links for commerce sites.
- Article to publisher/author links — personal sites are checked against Article → Person.
- Cross-script JSON-LD id references and graph arrays are resolved before scoring.
- Non-applicable nesting checks are removed and points are reallocated.
5pts
Open Graph & Social Metadata
When the page is shared or fetched, are the title, summary, and image clear. Social previews and AI both use these.
Show the details
- og:title / og:description
- Image availability
- Share-ready page metadata
20pts
Content Richness & Freshness
Is there real content that answers questions, and has it been updated lately. Empty shells and stale pages are hard to cite.
Show the details
- FAQ quality and Schema — counted only for site types that publish FAQs (commerce / SaaS / corporate / blog / unknown).
- Core content pages with Article Schema — paths like /columns/, /news/, /press/ also count; pages without a keyword URL fall back to Article-schema detection.
- HowTo / technical documentation / about pages — counted only when applicable.
- datePublished, dateModified, and recent-update signals are measured against Article-class schemas, not every WebSite or WebPage entry.
- Personal sites: FAQ and HowTo points move into Blog (8) and Freshness (12).
15pts
Knowledge Graph Authority
Can outsiders confirm you are a real brand: official profiles, mentions elsewhere, place and time clues. The more checkable you are, the easier you are to trust.
Show the details
- sameAs links and knowsAbout areas
- Founding date and entity history
- External authority references
- Internal entity links and location signals
20pts
Content Citability & Absorption
Does the writing include definitions, data, comparisons, and steps that can be copied into an answer. The more “this is the answer” a paragraph feels, the easier it is to cite.
Show the details
- Self-contained content blocks
- Definition blocks that explain concepts clearly
- Data blocks with percentages, amounts, years, counts, or measurable facts
- Comparison and step blocks
- Direct answers plus summary, table, list, and FAQ answer assets
10pts
E-E-A-T Signals
Can a reader see who wrote this, why they are qualified, how to contact you, and whether privacy or terms pages exist. That makes both people and AI more comfortable citing you.
Show the details
- Author Schema and a visible byline — HTML class/rel/itemprop hints, plus the Schema-resolved author name in page content as a fallback.
- Required trust pages depend on site type: personal = contact only; corporate / blog = privacy + contact; commerce / SaaS / unknown = privacy + contact + terms. Unused trust-page weight is reallocated to expertise/about.
- Experience and expertise cues — credentials, founding history, professional bios.
- About page presence (weighted higher for personal sites).
5pts
On-Page AI Readability
Is the page tidy: clear headings and readable text. A messy page is hard for a machine to take apart.
Show the details
- Semantic heading hierarchy
- Readable HTML structure
- Low-friction parsing
3
Agent Readiness: can an assistant get in
100 pts
This is not the GEO score. It checks whether the signs an assistant needs are in place. Essentials (80 points) are what most sites should have. Recommended (20 points) are extras. If a check does not apply to your site, it is marked N/A and does not cost points.
Think of it as three steps
01
Can it find you
Can an assistant find your site and treat it as something worth recommending.
Show the details
Looks at robots.txt, whether AI crawlers are blocked, llms.txt, and the sitemap.
02
Can it read you
After it arrives, can it read the text and structure without running complex scripts.
Show the details
Looks at HTTPS, structured data, HTML without JavaScript, and canonical.
03
Can it use you
Scored only for products an assistant should operate. Blogs and content sites usually skip this.
Show the details
Looks for MCP and similar machine interfaces. Personal and blog sites are N/A and are not penalized.
Most sites should have these (80 points)
12pts
HTTPS
The site uses an encrypted connection. This is a basic condition for an assistant to visit.
Show the details
Missing HTTPS is not a fixed -10 GEO penalty, but it lowers Agent Readiness.
10pts
robots.txt available
There is a file at the site root that tells crawlers what they may look at.
Show the details
A valid robots.txt needs to be available at the origin.
16pts
Major AI crawlers allowed
Common crawlers from ChatGPT, Claude, and Perplexity are not blocked. If they are, the GEO score also loses 15 points.
Show the details
Retrieval crawlers include GPTBot, ClaudeBot, and PerplexityBot. Blocking also triggers -15 on the GEO score.
14pts
llms.txt
A site guide written for AI, so an assistant can quickly understand how the site is organized.
Show the details
A valid llms.txt should be available at the root.
10pts
Sitemap
A list of pages, so an assistant knows what it can read.
Show the details
A usable XML sitemap should be present.
10pts
Structured data
Machine-readable labels on the page, so an assistant can tell who this is and what it is.
Show the details
JSON-LD / Schema helps assistants build an entity model.
8pts
Crawlable HTML
The main text is visible even with scripts turned off. If the homepage is a blank shell, an assistant may see nothing.
Show the details
The homepage should be readable without JavaScript, not a CSR blank page.
Nice to have (20 points)
6pts
llms-full.txt
A fuller excerpt file. Not required, but more convenient.
Show the details
An extra source file for longer excerpts.
6pts
Canonical
Tells machines which URL is the official one, so duplicate pages are less confusing.
Show the details
Primary pages include self-referencing canonical tags.
8pts
MCP / machine interface
An interface assistants can operate. Personal sites and blogs usually do not need this, so it is marked N/A.
Show the details
Looks for /.well-known/mcp.json or /mcp. Personal and blog sites are N/A.
Two cases to call out separately
If the site only shows content after running scripts
A page that is blank without JavaScript is not a fixed GEO penalty, but it is flagged as a serious issue and lowers content, structured data, and the Agent Readiness “HTML can be read” check.
How Ora / Cloudflare scores are used
The report may show their public results for comparison. Fixes still follow the GEO rules and our own Agent Readiness on this page.
4
Different sites are judged differently
A personal blog, a shop, a software product, and a company site are not supposed to have the same pages. We detect the site type first, then score what that type should have. Missing items that do not apply are not deducted.
Personal / creator
We care more about who you are, what you wrote, and whether it has been updated. You are not forced to have product pages or FAQs.
- Structured labels
- Person + Organization + Article/BlogPosting. Article → Person is the primary nesting check.
- Expertise and trust
- Trust pages: contact only. Author Schema and a visible byline still required, with more weight on expertise (3) and About (2).
- Content and freshness
- FAQ and HowTo do not apply. Their weight moves to Blog (8) and Freshness (12) so the 20-point total is reachable.
Commerce
We care more about product information, FAQs, and trust pages related to buying.
- Structured labels
- Organization, Product/Offer, FAQ, Article, HowTo, ItemList. Product → Organization is checked.
- Expertise and trust
- Trust pages: privacy + contact + terms all counted.
- Content and freshness
- Default split — FAQ 6, Blog 4, Knowledge 4, Freshness 6.
SaaS / software
We care more about what the product is, how to use it, and whether privacy, contact, and terms pages are present.
- Structured labels
- Organization plus SoftwareApplication/WebApplication, FAQ, Article, HowTo, ItemList. Product Schema is required only when the site has a catalog.
- Expertise and trust
- Trust pages: privacy + contact + terms.
- Content and freshness
- Default split — FAQ 6, Blog 4, Knowledge 4, Freshness 6.
Corporate / service
We care more about the company, the service, and how to contact you. Product pages are not required.
- Structured labels
- Organization, Article, FAQ, ItemList. Product and HowTo are skipped unless detected.
- Expertise and trust
- Trust pages: privacy + contact (terms not required). Unused terms weight is reallocated to expertise (up to 3).
- Content and freshness
- Default split — FAQ 6, Blog 4, Knowledge 4, Freshness 6.
Blog / content
We care more about article quality and how often you publish. Product checks follow shop rules only if you actually sell products.
- Structured labels
- Organization, Article/BlogPosting, FAQ, ItemList. Product checks skipped.
- Expertise and trust
- Trust pages: privacy + contact. Terms weight reallocated to expertise (up to 3).
- Content and freshness
- Default split — FAQ 6, Blog 4, Knowledge 4, Freshness 6.
5
How the score is calculated
We score the nine things first, then check whether AI is blocked. Blocking costs extra points. A helpful AI guide can add a little.
- 01
Open the site
We visit the URL you submitted and look at the pages and structure.
- 02
Score by site type
We first decide whether this is a personal site, a shop, or a company site, then score the nine things to produce the GEO score.
- 03
Check if AI is blocked
Blocking common AI crawlers costs points. Giving AI a site guide can add a little. Bonus is capped at +10. The final score stays between 0 and 100.
- 04
Score assistant access separately
The same scan also produces Agent Readiness. That number is not written back into GEO.
- 05
Tell you what to fix first
Issues are listed by urgency, and marked for engineering, content, marketing, or legal when that helps.
What costs points
-15
Common AI crawlers are blocked
If ChatGPT, Claude, or Perplexity cannot read your site, even great content is hard to cite. This is the heaviest penalty.
Show the details
Blocking GPTBot, ClaudeBot, PerplexityBot or similar retrieval crawlers in robots.txt suppresses visibility regardless of content quality.
-3
The unencrypted address does not jump to the encrypted one
If visits still land on the unencrypted address with no automatic jump, systems may not know which URL is official.
Show the details
HTTP does not redirect to HTTPS. Missing HTTPS itself is not a fixed -10 GEO penalty.
-3
Too many jumps
If a link has to bounce too many times before the real page appears, fetching slows down and ownership gets fuzzy.
Show the details
Too many hops to the final URL slows crawlers and dilutes signal attribution.
What adds points
+3
Give AI a site guide
Putting llms.txt at the site root helps AI find the important pages faster.
Show the details
An llms.txt file at /llms.txt makes high-value resources easier for AI systems to discover.
+3
Lists and rankings are labeled
If you have rankings, collections, or lists, structured labels make it easier to show up in “top N” answers.
Show the details
ItemList Schema helps list-style AI answers.
6
Grade bands
Letter grades are a quick read of the GEO score. Agent Readiness uses the same letters, but the number is separate.
A+
90-100AI can already find and cite you easily.A
80-89Strong foundation. A few more fixes would make it more reliable.B
70-79Readable, but there is still clear room to improve.C
60-69Gaps that make AI extraction and citation less stable.D
0-59Weak machine-readable basics. Hard for AI to cite reliably.