We've been treating Heiso's own site as an AEO sandbox the past few months — from robots.txt and JSON-LD all the way to FAQ Schema. The technical layer is covered, but what turned out to be most under-rated is the copywriting itself.
Our first About page had sentences like this:
Adhering to the principle of customer-first service, we are dedicated to providing one-stop digital transformation solutions for clients across all industries.
Sounds professional. AI reads it and has no idea who we are or what we do.
Why AI doesn't get it
Simple fact: AI is trained mostly on English.
GPT, Claude, Gemini's training corpora are overwhelmingly English. When they read other languages, they apply English-style expectations — subject first, conclusion first, every sentence stands alone.
How AI-friendly (English-style) writing differs from traditional / corporate writing:
| AI-friendly (English-style) | Traditional / corporate |
|---|---|
| Subject mandatory, opens the sentence | Subject often dropped, inferred from context |
| Conclusion first | Build-up first, conclusion last |
| Each sentence self-contained | Sentences chain by tone, fragments out of context |
| Verbs as backbone | Nouns stacked, verbs nominalized |
Native readers fluently parse the right column. AI reads by the left column's standard.
So writing for AEO basically reduces to one thing: write more like English, regardless of your source language.
Three writing principles
1. Lead with the conclusion
The first sentence of every paragraph IS the conclusion. Reasons and details come after.
Traditional structure is context → buildup → conclusion. AI reads only the first line. Burying the conclusion three paragraphs in = invisible to AI.
2. Each sentence stands on its own
Pull a sentence out. Without context, can the reader still understand it?
Common bad habits:
- Pronoun openers: "This approach has the advantage of..." — what's "this"?
- Conjunction openers: "Therefore, we recommend..." — what's "therefore"?
- Subject drops: "Can reduce maintenance cost." — who does what to what?
Fix: every paragraph's first sentence has its full subject. No pronouns. No connectors.
3. State who does what, plainly
The most common B2B-corporate trap: hiding verbs inside nouns, dropping the actor. For example:
Website design and development are our main services.
Means the same as: We design and build websites.
Same meaning, different shape. The subject moves from buried "our" to leading "we"; "design and development" returns from noun-phrase to verbs. AI reads the second version and understands your company in one parse.
A real before/after
Before
Adhering to the principle of customer-first service, we are dedicated to providing one-stop digital transformation solutions for clients across all industries through forward-looking technology and deep industry insight, helping enterprises achieve sustainable cost-reduction and efficiency goals.
AI reads this: doesn't know who, what, or for whom.
After
Heiso provides software outsourcing, custom system development, and AI Agent deployment services. From corporate websites, LINE Bots, AI workflow automation, RAG knowledge bases, to AI implementation consulting, we help SMBs complete digital transformation with an Agent-First mindset.
Similar word count. AI reads:
- Sentence 1: who + three core service lines (software outsourcing, custom systems, AI Agent build)
- Sentence 2: specific offerings (websites, LINE Bots, workflow automation, RAG, consulting — long-tail keywords for AI to grab)
- Sentence 3: target customers + method (SMBs, Agent-First mindset)
Less effort, too — you just say what you do, no piling on adjectives.
This paragraph is what Heiso's site actually uses. Open heiso.io → view source, and you'll find this exact text in the Organization JSON-LD and meta description. Fully verifiable.
How to test if it worked
After making the copy changes alongside the technical changes covered earlier, the way to verify is simple — open ChatGPT or Perplexity and ask:
- "What does [your company] do?"
- "What services does [your company] offer?"
If the answer references your specific services and cites your domain, the rewrite is doing its job. If you still get vague or "I can't confirm" responses, either the AI hasn't re-crawled yet (give it 1–3 weeks) or the copy still isn't structured enough.
Technical changes let AI crawl in. Copy changes let AI talk back. You need both — most teams stop at the technical layer and wonder why nothing happens.
Closing
If you remember just one thing: when writing in any language, write like English — subject first, conclusion first, each sentence readable alone.
Long, "professional" sentences = AI loses the thread; short, "blunt" sentences = AI knows you. This is the opposite of decades of "polished" corporate writing instinct, but it's the AEO reality.
Next post: How to Verify Your AEO Work.
想把這套用在你的專案？
從盤點問題、設計流程到落地實作——Heiso 顧問團隊一起接住