Our Website Was Invisible to AI Assistants: A 3-Week GEO Case Study

Setting the scene

We run an enterprise online training platform serving HR departments in SMBs. Our website has been around for five years, and Baidu SEO has been maintained by a dedicated person. Rankings were stable, traffic was predictable.

Then in late 2025, we noticed an unsettling trend:

More and more new clients told us: “We found you when we asked an AI assistant.”

The problem? We had never found ourselves in any AI engine’s answers.

The cold-shower discovery

In early 2026, I ran a simple test: I asked Doubao (ByteDance’s AI), Qwen (Alibaba), and ChatGPT questions like “recommend an enterprise online training platform” and “which LMS is best for SMBs.”

The result made our SEO lead go quiet — not one of the three AI assistants mentioned our brand.

To make it worse, some answers listed our competitors by name. We were simply invisible.

Our first reaction: “But our Baidu ranking is fine — surely AI engines can find us?”

Turns out, what a search engine can index and what an AI engine can cite are two completely different things.

Why we were invisible — we had no idea

We ran through our own checklist:

• Sitemap present ✅
• Baidu indexing normal ✅
• Google Search Console looked fine ✅
• Content updated regularly ✅

Everything seemed correct. Yet AI engines couldn’t find us.

After searching online for days, I stumbled upon an article — GEO in Practice: Making Your Website Discoverable by AI Assistants — and by the end of the first paragraph, I knew exactly what our problem was:

Our website was built for humans and search engines, but it was practically invisible to AI crawlers.

How AI engines saw our site

That article opened my eyes to several things I had never considered:

First, AI crawlers and search engine crawlers are fundamentally different.

Baidu and Google crawl pages and build their own indices. But AI engines (ChatGPT, Claude, Perplexity, Doubao, Qwen) use dedicated AI crawlers — GPTBot, ClaudeBot, PerplexityBot, CCBot — with completely different crawl logic and priority rules.

When we checked our server logs: GPTBot had visited our site only twice in the past six months. ClaudeBot and PerplexityBot: zero visits.

Second, AI crawlers can’t handle heavy JavaScript rendering.

Some of our landing pages used extensive JavaScript for visual effects. Googlebot has learned to render JS, but AI crawlers lag far behind. They likely crawled our pages, got empty DOM shells, and moved on.

Third, AI engines aren’t after “rankings” — they’re after “understanding.”

Traditional SEO aims to get you to position #1 in search results. GEO aims to make AI engines understand what your site is, what value it provides, and why you deserve to be cited in answers. These share some infrastructure (structured data, sitemap, robots.txt), but GEO adds critical requirements:

• llms.txt — a file that tells AI crawlers which pages matter and how to prioritize
• AI crawler whitelisting — many robots.txt files block unfamiliar crawlers, including AI bots
• BLUF (bottom line up front) — content structure that leads with the conclusion, since AI engines prioritize page openings
• FAQ Schema — Q&A content that AI engines can directly cite in answers

Our 3-week GEO optimization

After reading that article, we decided to bring in a team with real experience rather than a generalist agency — the criteria in How to Find a Reliable Technical Partner guided our selection.

AI Enable Harness first ran a full GEO audit. The report came back in about a week:

Audit findings (GEO Health Score: 32/100)

Item	Status	Note
robots.txt AI crawler access	❌	GPTBot and ClaudeBot were blocked by default
llms.txt	❌	Did not exist
JSON-LD structured data	⚠️ Partial	Organization on homepage, but no Article on content pages, no FAQPage
BLUF content structure	❌	Articles opened with background, conclusions at the end
AI crawler crawl frequency	❌	GPTBot: 2 visits in 6 months; ClaudeBot/PerplexityBot: 0
sitemap	✅	Existed but not priority-grouped
Core Web Vitals	✅	Good

That score was a wake-up call. For years we thought SEO meant “online visibility” — but in the AI channel, we were essentially nonexistent.

The 3-week optimization plan

The work was split into three weekly phases:

Week 1: Infrastructure rebuild

• Modified robots.txt to explicitly allow GPTBot, ClaudeBot, PerplexityBot, and CCBot
• Created llms.txt in the site root, listing core pages with crawl recommendations
• Added full JSON-LD structured data to all content pages (Article, BreadcrumbList, FAQPage)
• Reorganized the sitemap by page priority

Most of this was “configure once and it works” infrastructure — near-zero ongoing cost, but foundational for everything else.

For the specifics of writing and maintaining llms.txt, see this detailed guide: How to Write llms.txt

Week 2: Content structure overhaul

• Rewrote core pages and blog posts using BLUF structure — the first two paragraphs state the conclusion and key takeaway
• Embedded FAQ sections on relevant pages with FAQPage Schema markup
• Pruned or merged low-value pages to concentrate AI crawl budget
• Increased internal link density to help AI crawlers understand content relationships

The hardest part was retraining our writing habits. We were used to “context → analysis → conclusion.” BLUF flips that order. After a few days of adjustment, we found it not only improved AI friendliness but also human readability — readers no longer had to scroll to the bottom to understand the point.

For a step-by-step guide on safely opening your site to AI crawlers: AI Crawler Whitelist Setup

Week 3: Verification and iteration

• Submitted sitemaps to AI crawler endpoints (not all AI engines accept submissions, but it’s worth doing for those that do)
• Set up AI crawler log monitoring with weekly frequency check-ins
• Configured brand keyword tracking on Serper, Brand24, and Google Alerts to watch for AI citation emergence
• Adjusted content priority based on crawl data

Three-month results

GEO optimization doesn’t work overnight, but our timeline was clear:

Time	Change
Week 1	GPTBot began crawling daily — from 2 visits in 6 months to 3-5 per day
Week 2	Doubao and Qwen started mentioning our brand in answers for the first time
Week 4	ChatGPT began citing our content, mainly on training-vendor selection questions
Week 8	Perplexity started citing us, coverage expanded from 3 question types to 12
Week 12	We tracked approximately 17% of new inquiries originating from users who discovered us through AI answers then searched our site

The most important observation: GEO effects didn’t plateau at week 12. AI engines continuously crawl and refresh citations. Sites that keep publishing and updating get higher priority in AI knowledge bases — there’s a clear “freshness signal” at work.

Lessons from the buyer’s seat

1. GEO is not a replacement for SEO — it’s an addition

SEO remains your traffic baseline. But GEO is becoming a channel you cannot afford to skip — when users start asking AI assistants for vendor recommendations, if you’re not in the answer, you’ve lost that traffic entirely.

2. Early movers have a real advantage

In 2026, GEO still has a meaningful first-mover advantage. Most companies haven’t systematically optimized for AI engines. The citation pool is still shallow. Getting your infrastructure right now gives you a much higher probability of being cited than waiting a year when everyone is doing it.

3. Infrastructure matters more than content volume

A common reaction to GEO is “let’s write more content.” But AI crawlers first need to find you and understand you — without llms.txt, structured data, open robots.txt, and a clean sitemap, even great content is invisible.

4. Effects compound

GEO, SEO, and content marketing are not three separate activities. Good structured data helps both search engines and AI engines. BLUF content structure benefits both human readers and AI crawlers. FAQ Schema drives AI citations while also powering search engine rich snippets.

5. Don’t focus only on ChatGPT

User habits are fragmented. In China, Doubao and Qwen have massive user bases. In global markets, Perplexity is preferred by technical audiences, while Claude is strong in professional contexts. Different AI engines have different crawl frequencies and citation behaviors — in our experience, Doubao and Qwen were fastest to cite us, ChatGPT was more stable over time, and Perplexity was most responsive to technical content.

---

The bottom line: AI visibility is not optional anymore

The biggest takeaway from this experience is simple: channels change, but the need to be found by the right people doesn’t.

People who missed the WeChat official account wave three years ago don’t want to miss the AI citation wave today. GEO is not a gimmick — it’s your website’s infrastructure ticket to the AI era. Not expensive, but no longer optional.

If you’re considering GEO, start with an audit — find out what your website looks like through the eyes of an AI engine.

Further reading:

• GEO in Practice: Making Your Website Discoverable by AI Assistants — The complete 7-layer GEO setup guide
• Technical SEO Playbook 2026 — Full optimization checklist from crawl to rank
• How to Write llms.txt — Practical guide to AI crawler navigation files
• AI Crawler Whitelist Setup — Engineering approach to safely opening your site to AI crawlers
• Three Years of SEO, Still Invisible to AI — Another buyer’s story: how to vet a GEO-capable vendor
• How to Measure GEO Results — Four-layer verification: crawler logs, index coverage, brand tests, referral attribution