What is GEO.new? How to use it

geo.new is a free audit of one public web page. You give it an address, and it reads that page the way an AI search crawler would: what the crawler is allowed to fetch, what it receives, and what it could quote. Then it prints what it found, with the evidence under every result, and what to fix first.

What geo.new is

GEO in one paragraph

GEO stands for generative engine optimization: making a page easy for AI assistants such as ChatGPT, Perplexity, Claude and Google's AI answers to find, read and cite. Much of it is SEO, because a page that cannot be crawled cannot be cited either. Some of it is new: which AI crawlers your robots.txt lets in, whether the main text is in the HTML or only appears after JavaScript runs, and whether a paragraph still makes sense when it is quoted on its own.

What it reads

An audit reads robots.txt, the page's source HTML exactly as the server sent it, and the few files that decide whether the page can be read at all. One optional browser capture runs beside it and shows what a person would see. It does not crawl the rest of your site, sign in, or click anything.

What you get back

One GEO score from 0 to 100, built from 176 checks in six weighted categories:

• AI Citability (22): whether passages can be lifted and quoted as they stand.
• Content and E-E-A-T (22): experience, expertise, authority, trust and content quality.
• Brand and Entity (20): whether the brand behind the page is a recognisable entity.
• AI Access and Rendering (15): crawler access, responses, and text that arrives without JavaScript.
• Schema and Structured Data (13): what the page declares about itself in JSON-LD.
• Platform Readiness (8): how the same checks map onto individual assistants.

The score is printed only when the HTML was usable and at least 80% of the weighted checks could be measured. Otherwise you get the results without a number, rather than a number built on guesses.

It is free and needs no account. A run costs real money in model calls, browser rendering and storage; why it is free explains the arrangement.

How to run an audit

From the homepage

Paste the address into the field on the homepage and press run.

From any address bar

Put geo.new/ in front of the address you are already looking at:

geo.new/https://example.com/pricing

The audit starts as soon as that page opens.

From a terminal or a coding agent

The same audit runs from a terminal, with nothing to install first:

$ npx -y geo-new audit https://example.com/pricing

To let a coding agent such as Claude Code run audits itself, add geo.new as an MCP server:

$ claude mcp add geo-new -- npx -y geo-new@0 mcp

The agents page covers the HTTP API as well.

Reading the report

Results arrive as they are measured. The HTTP response and the deterministic checks appear on the run screen first, and the report opens once the written interpretation is ready.

Verdict

The verdict leads with the score, its grade, and a short diagnosis a language model wrote from the stored evidence. Every finding it quotes is a literal piece of your page or its headers, and every claim names the checks it rests on; anything that fails either test is dropped before it is saved. The model interprets the results. It never changes the score.

Some failures matter more than any average. If the page answers with an error or a bot challenge, carries a noindex, is blocked for Googlebot in robots.txt, or is an empty shell that only fills in with JavaScript, that failure is printed beside the score, so a good number cannot hide it. Those four are ACC-01, ACC-10, ACC-02 and ACC-14, in the order above, in AI Access and Rendering.

Fix list

Fixes are ranked by measured severity and the effort they take. Each one names the checks it resolves and has a Copy fix button, ready to paste into a ticket or an assistant.

All checks

Every check is listed with its result and the evidence behind it, filtered by Issues, Needs a look, Passed or Not measured. Two views are worth opening on their own. Who can read it shows which search, AI search and training crawlers your robots.txt lets in; blocking training crawlers such as GPTBot never lowers the score. Agent readiness records what an AI agent would find besides the page, such as a sitemap or an llms.txt; it is reported, not scored.

Unknown is not a failure

A check that could not be measured is marked unknown. It counts as neither a pass nor a warning, and it is listed under Not measured, so you can see exactly what the audit could not see.

After you change something

1. 1

   Fix the first item on the list

   The list is ranked by measured severity and the effort each fix takes, so the first item is where to start.

2. 2

   Press Run again on the report

   Submitting the same address from the homepage within the hour shows the recent result instead of measuring again.

3. 3

   Open Progress

   It compares the new run with an earlier one, check by check: fixed, new issue, changed or unchanged.

To hand the rest to whoever does the work, the Full report has a button that copies a fix prompt for your AI assistant, and the whole report can be copied or downloaded as Markdown. To show someone the report, send its address: it opens read-only for anyone who has it. Reports are deleted after seven days.

What geo.new does not tell you

• It reads one page. It says nothing about the rest of the site, so audit each page that matters.
• A pass means a signal was present, not that any assistant cited the page.
• The thresholds are our product behaviour, not ranking rules published by OpenAI, Google or anyone else.
• Some reports include what one sampled answer from ChatGPT or Google AI Mode said about the brand. Those rows are marked experimental, never scored, and the next answer may differ.

If a check got your page wrong, write to [email protected] with the report's address and the check's ID. A check you disagree with is the most useful message you can send.