Three Years of SEO Spend, Still Invisible to AI: A Buyer's Post-Mortem

The bottom line first

We are an industrial equipment after-sales service company serving manufacturing plants. For three years we spent tens of thousands of yuan a year on SEO services — keyword rankings climbed, organic inquiries grew. Then in early 2026 we discovered a bizarre fact:

No AI assistant — ChatGPT, Doubao, Qwen, Kimi — had ever mentioned us.

Not occasionally. Never. Competitors were named in AI answers. We weren’t. The rankings we paid three years to build only seemed to work inside Baidu’s search page. In the AI world, we simply didn’t exist.

This article is the full story of that post-mortem: why traditional SEO services couldn’t fix AI visibility, how we learned to vet for real GEO capability, and what changed after we pivoted.

---

Background: three years of SEO services

We sell after-sales service for industrial equipment. Customers find us on Baidu by searching phrases like “equipment repair”, “after-sales outsourcing”. Three years ago we hired an SEO agency that provided:

• Keyword mapping and content ghostwriting (4-6 articles per month)
• Backlink building (10-20 per month)
• Baidu paid-search ad management
• Monthly ranking reports

To be fair, the service worked: core keywords climbed from page three to page one, and monthly inquiries grew from about 20 to 60. We were satisfied and kept renewing the contract.

Late 2025: the way customers talked began to change

The shift started in the second half of 2025. More and more new clients said something different from “I saw you on Baidu”:

“Our plant was discussing equipment maintenance options, asked an AI assistant, it mentioned a few companies — I checked your website.”

At first we thought, “whatever the source, they came.” Then we ran our own test and stopped being complacent.

One test revealed we were invisible in the AI world

In early 2026 we ran a simple brand test. We asked Doubao, Qwen, ChatGPT and Kimi:

• “How do I choose an industrial equipment after-sales service provider?”
• “Who’s reliable for [our product category] repair?”

The results silenced the whole team:

1. None of the four AI assistants mentioned us;
2. A couple of answers listed similar companies — including our direct competitors;
3. The most painful part: some competitors named by AI ranked below us on Baidu.

Meaning: the three years of rankings we built in search engines were worth nothing in AI engines.

First reaction: we called our SEO agency

We immediately called the agency we’d worked with for three years: “Why don’t AI answers include us?”

Their response — now clearly a tell:

“We can’t control what AI says. We do search engine rankings. Look at the ranking reports — everything is improving.”

It sounded reasonable at the time, but something felt off. We started researching on our own and gradually found the key fact:

AI engines citing websites and search engines ranking websites are two completely different mechanisms.

Our entire mental model of “online visibility” was: make the search engine rank my page higher. But AI engines work differently — they need to understand what your site is, whether the content is trustworthy, and whether it deserves a citation in an answer. This goal has several hard requirements that traditional SEO never touches:

• llms.txt: a site manifest for AI crawlers (we’d never even heard of it)
• AI crawler access: our robots.txt may have been blocking GPTBot by default
• BLUF content structure: conclusion first — AI engines read page openings first
• JSON-LD structured data: helping AI understand who you are and what each page covers

At that point we got it: the agency wasn’t incompetent — we had bought the wrong service. Traditional SEO and GEO are two different systems. Our agency had never provided anything GEO-related, and we had never asked — because we didn’t know GEO existed.

Post-mortem: three misconceptions we held

Misconception 1: “Good rankings = AI recognition”

The biggest trap. Baidu rankings depend on keyword matching, backlinks and page authority; AI citations depend on content comprehensibility, structured data and crawler accessibility. Rankings serve search users; citations serve AI engines. Different audiences entirely.

Misconception 2: “A more expensive SEO agency will fix it”

We briefly considered upgrading to a pricier agency, assuming “more expensive means more capable.” We learned that in 2026 the vast majority of SEO agencies still sell traditional services — content ghostwriting, backlinks, ranking reports. Paying more for traditional SEO does not solve AI visibility.

Misconception 3: “AI citations are voodoo — uncontrollable”

Our agency said “we can’t control what AI says,” and we believed it for a while. We discovered that GEO is an engineering practice — from robots.txt whitelisting to llms.txt to structured data, every item is a deterministic configuration. Not voodoo.

Pivoting: how we vetted for real GEO capability

With that lesson burned in, we refused to hire blindly this time. Using the framework from How to Find a Reliable Technical Partner, we developed three questions specifically for testing GEO expertise:

Question 1: “Have you ever shipped llms.txt? How did you prioritize entries?”

• Doesn’t know what llms.txt is → disqualify
• Says “list core pages first, give AI crawlers a guide” → passable
• Says “group by page importance, partition by content type, sync regularly” → actually shipped it

Question 2: “Do your articles open with the conclusion or the background? Which part does an AI engine read first when truncating?”

• “We focus on content quality” → non-answer, disqualify
• Names BLUF (bottom-line-up-front) structure → passable
• Explains “the first two paragraphs must contain the conclusion, or AI truncation captures background noise” → genuinely understands

Question 3: “Which schema types does your structured data cover?”

• “We did structured data” but can’t name types → never really shipped it
• Names Organization / Article / FAQPage / BreadcrumbList → passable
• Explains which schema belongs on which page type and why → has real delivery experience

With these three questions we eliminated two cheap “full-service” agencies and settled on a team that answered fluently — they started with a full GEO audit.

The audit: GEO health score 25/100

The team (AI Enable Harness) spent a week on an audit. Key findings:

Check	Status	Note
robots.txt allows AI crawlers	❌	GPTBot / ClaudeBot / PerplexityBot blocked by default
llms.txt	❌	Did not exist
JSON-LD structured data	❌	Scattered and unsystematic
BLUF content structure	❌	Everything written “context → analysis → conclusion”
AI crawler crawl frequency	❌	GPTBot visited 3 times in six months
sitemap	⚠️	Existed but not priority-grouped
Core Web Vitals	✅	Good

GEO health score: 25/100.

That score made everything clear — in three years of services, not a single yuan had been spent on AI visibility. The agency hadn’t deceived us; it was a different game, and we never realized we were playing it.

Three months of changes: from 25 points to being cited

After pivoting, we spent about three months on a complete GEO pass (the detailed 7-layer infrastructure is covered in this GEO in Practice guide):

Time	Change
Week 1	Opened robots.txt to AI crawlers + deployed llms.txt + added JSON-LD
Week 2	Doubao and Qwen began mentioning our brand in answers
Week 4	ChatGPT started citing our website articles on after-sales selection queries
Week 8	After BLUF restructuring, the number of cited articles doubled
Week 12	Coverage expanded from 2 question types to 9; about 14% of new inquiries cited “AI recommendation”

The most convincing detail: in week 12, a client said verbatim, “an AI assistant recommended you, we checked your site, and here we are.” That was the first time in three years a client explicitly credited an AI referral — previously they always said “I saw you on Baidu.”

The most painful lessons from the buyer’s seat

1. Your vendor’s KPIs determine their behavior.

Our agency’s KPIs were rankings and indexing, so that’s all they did. Without “AI engine citations” in the contract, no agency will ever invest in GEO. To add the KPI, you first have to know GEO exists.

2. By 2026, vetting a vendor must include GEO questions.

Ask “do you do GEO?” today and almost everyone says yes. But saying you do and having done it are different — the three questions above filter most of them out fast. Those who answer fluently have actually shipped it.

3. Traditional SEO isn’t wasted money — but it isn’t everything either.

Our traditional SEO still brings dozens of search inquiries a month; we can’t cut it. But without GEO, we’re handing every “ask AI for a vendor” customer to our competitors. SEO is the baseline; GEO is the new channel. You need both.

4. Don’t let your vendor define your requirements.

We outsourced “website visibility” wholesale — the vendor defined both requirements and acceptance criteria. That was our biggest management failure. You must understand enough to judge what your vendor is actually doing, or you’re buying results you can’t evaluate.

5. Structured data and content structure are the common language of every AI engine.

Whether ChatGPT, Doubao, Qwen or Kimi — they all understand websites through the same primitives: structured data, semantic HTML, clear page hierarchy, crawlable content. Get the infrastructure right, and you’re present for whatever AI engine wins next.

---

Don’t wait for customers to tell you AI can’t find you

Our biggest regret: why didn’t we run a brand test sooner? If we had discovered the problem in 2025, we wouldn’t have renewed a full year of traditional SEO.

If you haven’t tested whether AI engines can find you, do it now — search your brand name and core business terms in Doubao, Qwen and ChatGPT. If you’re not in the answers, your website is invisible in the AI world.

Invisibility isn’t the disaster; not knowing you’re invisible is. Add GEO now and you’ll see crawler changes in 1-4 weeks and citation changes within 3 months. Wait until your customers are all asking AI, and you’re already behind.

Further reading:

• GEO in Practice: Making Your Website Discoverable by AI Assistants — The complete 7-layer GEO setup guide
• How to Find a Reliable Technical Partner — Six dimensions for evaluating a vendor’s real capability
• Technical SEO Playbook 2026 — Full optimization checklist from crawl to rank
• Invisible to AI Assistants? A 3-Week GEO Case Study — Another buyer’s complete 0-to-1 GEO journey
• How to Measure GEO Results — How to verify whether your GEO vendor actually delivered