Over the past three posts, everything was about “fixing your own website”: how to write page titles, two pages competing for the same ranking, and writing your company identity in a machine-readable format.
Today we’re going to talk about something uncomfortable — but the sooner you know it, the better: there is a ceiling to what your own website can do.
This conclusion didn’t come out of thin air. We ran a very simple controlled experiment, and the results surprised even us.
A 10-minute experiment you can run yourself
Find an AI with web search, ask it about the same company in three different ways (use your own company), then look only at the source links it lists — you don’t need to judge how good its answer is.
| Question format | Example | Sources we actually observed |
|---|---|---|
| ① Company name only | “Beijing Such-and-Such Company” | ? The official website ranks first, with the homepage, the About page, and several city pages all present |
| ② Company name + evaluative term | “Is Such-and-Such any good?” “Is Such-and-Such reliable?” | ? The official website disappears entirely, leaving only one third-party directory listing |
| ③ Industry only, no company name | “Which XX company in Beijing is the best?” | ? Not a single company’s official website appears in the top 20 |
Row ① is the good news: the past two weeks of work on the website weren’t wasted — when you search by name, you can find us.
Rows ② and ③ are the real point. Pay special attention to row ③ — it isn’t that we didn’t rank; it’s that no company’s official website ranked. What are the top 20? Question pages on Q&A communities, “Top 10 Companies” roundups on portal sites, B2B directories, and a few industry aggregator sites. Not a single official website.
Why? Because whether you’re any good isn’t up to you
Think of the AI as an assistant running errands for your customer. The customer asks it three kinds of questions, and it goes looking for three kinds of sources:
“Who is this company?” → The official website is the most authoritative source, so it uses that directly.
“What does this company do?” → The official website still counts; after all, no one knows the scope of the business better than the company itself.
“Is this company any good? Which one should I choose?” → The official website doesn’t count. Every company’s website says it’s the best, and that isn’t evidence. It has to go find out what others say about you.
So the conclusion is blunt: your website answers “who you are” and “what you do”; it cannot answer “whether you’re any good.” The answer to that second question is written on other people’s pages.
And the vast majority of companies have never managed those “other people’s pages.”
Our own example: a ten-year-old directory listing is speaking on our behalf
Back to row ②. When someone asks “How is Such-and-Such as a company?”, the only source the AI finds is a company page on a B2B directory site. That page says:
Category: Office Equipment
Business scope: sales of office equipment, office supplies, hardware products, and daily necessities; repair of office equipment…
Legal representative, address, phone: all outdated information from ten years ago
This page was posted years ago and long forgotten. But it’s still there — and it is now the main material the AI can draw on when it introduces us.
The most damaging part isn’t that the content is old; it’s that category.
Get the category wrong, and in the AI’s eyes you’re not in that industry at all. When it assembles answers like “which companies are in such-and-such industry,” it retrieves them by industry classification. If you’re filed under “Office Equipment,” then no matter how many times the “IT Services” category is queried, you’ll never come up — not ranked low, but not on the candidate list at all.
The second surprise: the keyword you think you’re fighting for may not be your battlefield
There was another episode in the experiment. We tried a query closer to how a real user searches, adding a year: “best outsourcing in such-and-such industry 2026.” Of the top 18 results:
- 8 were tender announcement sites (e.g., “Such-and-Such Agency 2025–2026 Annual Outsourcing Services Tender Announcement”)
- A few were brokerage research reports analyzing the performance of listed “human resources outsourcing” companies
- The rest were press releases from industry forum events
In other words, in the world of search, the word “outsourcing” belongs mainly to the context of “staffing outsourcing, labor outsourcing, and bidding,” which is a far cry from what we mean. If we blindly set out to optimize for that word, we’d be pushing hard in a room that isn’t ours.
So before choosing a keyword, search it once to see who currently occupies it. If the context of the top results has nothing to do with you, switch keywords — don’t force it. This step takes only 30 seconds and can save you months of effort.
So what should you actually do? A ranking by cost-effectiveness
Step one: find the pages that speak for you.
The method is the experiment above: use an AI with web access, ask once each with the company name, the company name + “how is it?”, and an industry term + “which is best?”, then copy down every source link it lists. That list is essentially “everything about you” in the AI’s eyes. Most business owners do a double take the first time they see it — half the pages on it are ones they never knew existed.
Step two: fixing what’s wrong takes priority over publishing something new.
| Fix First | Why |
|---|---|
| 1. Industry category | Get this wrong and you are effectively not in that business. It has the biggest impact, and one correction often fixes everything. |
| 2. Business scope / core services | Determines whether you appear in answers like “companies that can do X.” |
| 3. Legal representative, address, phone | If these do not match the official business registration, AI trusts the profile less. |
| 4. One-line summary | On many platforms, this sentence gets quoted in full. |
Updating one old page that already ranks is far more useful than publishing ten new articles nobody reads. That old page has already earned AI’s trust — you are simply correcting what is inside it.
Step 3: prioritize business-registration platforms. Data on these platforms comes from official sources, so it carries inherently higher credibility, and AI cites them far more often than ordinary directories. These platforms usually support company claiming (requiring legal-representative verification or a corporate bank transfer), and once claimed, the information is yours to manage.
Step 4 is getting onto “list-type” pages — industry Q&As, ranking roundups, media coverage. This step is slow and the most expensive, so it comes last. The first three steps repair assets that already exist; this one builds assets from scratch.
One practice to avoid: paying someone to mass-produce “positive reviews.”
A batch of posts with near-identical wording, published in the same period and empty in substance, is easy to detect. What actually works is specific facts that check out — year founded, certification numbers, how many clients you have served, what types of projects you have delivered. These withstand cross-verification, and cross-verification is exactly what AI does best.
Wrapping Up: Three Layers in Order, and the Third Is Beyond Your Website’s Reach
An earlier article on structured data mentioned a sequence; today we can complete it:
Layer 1 · Getting Found — your website can be accessed and indexed normally. (Something your website can handle on its own)
Layer 2 · Getting Understood — titles, business descriptions, and identity information are clearly written. (Something your website can handle on its own)
Layer 3 · Getting Mentioned by Others — how other pages describe you. (Your website cannot help you here)
Many companies stay stuck for a long time because they have already maxed out Layers 1 and 2 but keep tweaking anyway, thinking one more revision will break through. In reality, the bottleneck moved to Layer 3 long ago — and Layer 3 is entirely about work outside your own website.
For now, hold off on spending money on promotion. Take ten minutes, run those three queries, and see who is answering “what is this company like” on your behalf right now. Chances are you will find one or two old pages you had forgotten about, still speaking for you with decade-old information, without your knowledge.
We are Beijing Siwen Lide, with 14 years focused on enterprise IT services, more than 1,000 enterprise clients served to date, and 21 software copyright registrations plus ISO 9001 / ISO 20000 certifications. This series covers the pitfalls we hit along the way — you can copy the approach directly. If you would like to talk about your company’s situation, see About Us, or call 400-686-2011.
