I Ran a GEO Audit on My Own Blog. AI Crawlers Gave It Near-Perfect Access. Total Score: 42.

To get a better handle on GEO, I ran some optimization tests on my own blog, adamnote.com. On AI crawler access, it scored 95 out of 100. Top 5% of sites.

Total score: 42. Rating: Weak.

The door is open, the lights are on, the shelves are fully stocked, and every label is in a language the customer can’t read.

That’s the core finding from auditing my own site for GEO, generative engine optimization. It’s also the structural mistake most Chinese brands going global are making right now without realizing it.

What GEO actually measures

GEO is not SEO with a new name.

SEO optimizes for a search engine’s ranking algorithm. GEO optimizes for an AI system’s citation decision: when ChatGPT, Perplexity, or Google AI Overviews answers a user’s question, whose content does it choose to cite?

A few numbers:

• 60% of Google searches end with zero clicks; users get their answer on the results page
• How often a brand gets mentioned correlates with AI search visibility 3x more strongly than backlinks do
• Fewer than 5% of websites currently have an llms.txt file

The old SEO logic of “publish good content and wait to be found” is breaking down. AI isn’t searching your content. It’s judging whether your content is worth citing.

A GEO audit quantifies that “worth citing.” The dimensions: can AI crawlers access the site, is the content structured for citation, is Schema markup complete, are E-E-A-T signals sufficient, and how compatible is the site with each AI platform.

Perfect access to a language black hole

Back to the contradiction between 95 and 42.

The root cause turned out to be embarrassingly simple: my blog was Chinese-only.

A blog written for Chinese go-global practitioners is naturally in Chinese. AI search doesn’t see it that way.

ChatGPT runs on Bing’s index. Google AI Overviews runs on English-first language models. Perplexity’s citation sources are overwhelmingly English. When an English-speaking user asks “What are common ad mistakes for Chinese brands going overseas?”, the AI will almost never cite a Chinese-language article in its answer.

Beyond language, a pile of infrastructure was simply missing. No HTML lang attribute, so AI couldn’t even identify the language. No llms.txt, so AI had no idea what the site was about. No Person or Organization Schema markup, so E-E-A-T signals were zero. No author credentials on the About page. No FAQ Schema, so Google AI Overviews had nothing to trigger on. Not submitted to Bing Webmaster Tools, so completely invisible to ChatGPT Search.

What AI crawlers could access perfectly was a pile of content they couldn’t use.

From 42 to 65

I fixed it in two rounds.

Round one: fundamentals (about 2 hours)

1. Set the HTML lang attribute
2. Created llms.txt describing what the site is and what it covers
3. Added Person + Organization Schema
4. Rewrote the About page with industry experience and credentials
5. Added FAQ Schema to core articles
6. Submitted to Bing Webmaster Tools
7. Built a bilingual Chinese-English architecture and published the first English post (the English section you’re reading exists because of this audit)
8. Injected hreflang tags so the two languages point at each other

Round two: deeper fixes (about 3 hours)

1. Fixed the Ghost theme’s H1 problem (Casper renders post titles as h2 by default; a two-line code change resolved 18 SEO issues)
2. Batch-filled meta descriptions and cover images for 15 tag pages
3. Merged two author accounts, migrating 9 posts
4. Added SpeakableSpecification Schema (tells AI which passages it can quote directly)
5. Fixed the language markup on /en/ pages (English content had been tagged as Chinese)

Re-audit: 65/100. Out of Weak, into the upper end of Moderate.

Per-platform scores:

ChatGPT and Bing Copilot sit at the bottom. They share Bing’s index, and Bing hasn’t crawled most of my pages yet. That’s the easiest gap to close: force a re-submission, add more English content, and the scores come up. The Google side scores highest because Google responds fastest to Schema markup. Perplexity is a different animal. Its bottleneck isn’t technical, it’s external citations: it favors content cross-validated by third-party sources like Reddit, Quora, and academic papers. That takes time and reputation. No code fixes it.

Five hours, 23 points. More important than the points: the time exposed problems that, in my previous mental model, didn’t exist at all.

57 issues, 7 root causes

After the GEO fundamentals were done, I ran an Ahrefs Site Audit as outside verification.

The report showed 57 issues.

I didn’t fix them one by one. I exported three CSVs and cross-referenced them: the 57 “issues” mapped to 22 distinct pages, and behind them sat just 7 root causes.

One typical example: Ghost’s Casper theme rendering post titles as <h2> instead of <h1>. That single bug generated 18 Ahrefs records. Two lines of code cleared all of them.

The other root causes were the same shape, one fix clearing many entries:

• 15 tag pages missing meta descriptions and cover images: one batch script filled them all
• The dual-author problem: merged into one
• Orphan pages under /en/: fixed the routes.yaml routing

Ahrefs issues went from 57 down to 5-8, and what’s left are structural limitations of the Ghost platform itself.

When you’re staring at a long list of problems, find the root causes before you start fixing. One root cause will disguise itself as a dozen different “issues.”

Why brands should care

My blog is a personal site; the stakes are low. Put the same problems on the official site of a brand going global and the picture changes.

Take a Chinese consumer electronics brand with decent SEO and real organic rankings on Google. An American user asks ChatGPT to “recommend a good robot vacuum under $300.” The brand isn’t in the answer.

The product isn’t the problem. The problems are:

• The English content isn’t “citable”: paragraphs too long, no structure that directly answers user questions
• No Schema markup, so AI can’t quickly extract brand information and product specs
• No brand presence on Reddit or Quora, the sources AI weights heavily (Reddit contributes 21% of Google AI Overviews citations, Quora 14.3%)
• No llms.txt

The old go-global formula was “translate into English, do SEO, buy ads.” AI search follows a different logic: your content doesn’t just need to be found, it needs to be cited.

Being found is a ranking problem, and money can push it. Being cited is a content structure problem, and money can’t.

Priorities

Do now:

• Check your HTML lang attribute
• Create an llms.txt file
• Complete your Schema markup (Organization, Person, Product, FAQ)
• Submit to Bing Webmaster Tools
• Add clear English FAQ sections to core product pages

Do next:

• Build brand content on Reddit and Quora
• Shift English content from translation thinking to citation thinking: short paragraphs, direct answers, structure
• Set up brand mention monitoring

Don’t rush:

• You don’t need to rebuild your website
• You don’t need to abandon SEO
• You don’t need to mass-produce English content with AI

GEO adds a layer of AI readability on top of what you already have. It’s like adding mobile responsiveness to your site ten years ago. Not a revolution, an upgrade.

The window for brands here won’t stay open long. While your competitors are still figuring out what llms.txt is, getting the fundamentals right is a first-mover advantage.

42 to 65, in five hours.

FAQ

What’s the difference between GEO and SEO?

SEO optimizes search engine rankings; GEO optimizes AI systems’ citation decisions. AI doesn’t rank you, it judges whether your content is worth citing. The two run in parallel, and GEO is an upgrade built on an SEO foundation.

What is llms.txt and how do I create one?

llms.txt is a text file in your site’s root directory that tells AI, in a structured format, what your site does, what the core content is, and who the author is. Fewer than 5% of websites have one, so creating it puts you ahead of most competitors.

Can a Chinese-language website get cited by AI at all?

Not never, but the odds drop sharply. ChatGPT, Perplexity, and Google AI Overviews cite mostly English sources. A Chinese-language site needs at least English FAQ sections and English summaries to enter the citation pool for English AI queries.

How much time and money does a GEO audit take?

Basic audit and fixes take about 5 hours, mostly technical work: setting up Schema markup, creating llms.txt, submitting to Bing Webmaster Tools. No site rebuild, no extra budget.

Which AI platforms matter most for brands going global?

Google AI Overviews has the widest reach (1.5B monthly users), ChatGPT the strongest influence (900M weekly users), Perplexity the fastest growth. Their citation logic differs, but all three want content that can be extracted structurally and is cross-validated by third parties.

Adam Yang | 10+ years in China-to-global expansion · Independent consultant 中文版: 我给自己的出海博客做了一次 GEO 审计，AI 爬虫接近满分，但总分 42