Competitor Keyword Reverse-Engineering for Chinese AI
Competitor Keyword Reverse-Engineering for Chinese AI
TL;DR — Chinese AI platforms cite your competitors in response to specific user queries. If you can identify those queries systematically, you can build content specifically designed to displace the competitor in those citations. This five-step reverse-engineering process takes 2-3 weeks and often surfaces 50-200 high-value queries where you can compete. It is one of the highest-leverage activities in GEO.
What you're actually reverse-engineering
Traditional SEO competitive analysis asks "what keywords rank for my competitor in Baidu". That question, adapted to GEO, becomes: "what user queries cause Chinese AI platforms to cite my competitor as an answer".
These are different questions. A competitor might rank on page 1 of Baidu for a keyword but never get cited by DeepSeek for related queries — or vice versa. AI citation logic differs from SEO ranking logic. The keywords your traditional SEO tool shows don't directly translate to AI citation queries.
What you need is a purpose-built approach that:
- Identifies queries relevant to your category that AI platforms actually answer (not all queries get substantive AI responses)
- Tests each query across multiple platforms to capture competitor mentions
- Classifies which competitor wins each query and at what depth
- Surfaces queries where you have realistic displacement potential
Done well, this process identifies specific, winnable positions — not just generic keyword opportunities.
The five-step process
Step 1: Identify your competitor set
Start with 5-10 named competitors. Include both direct competitors (same product category) and adjacent ones (your buyers sometimes consider them as alternatives). Mix of established market leaders and newer entrants.
Specifically avoid:
- Giant conglomerates that dominate all AI responses regardless of relevance (Alibaba, Tencent, ByteDance themselves) — they're not your real competition
- Dead brands no longer actively maintained
- Brands outside your actual pricing or size tier
Your competitor list should feel like a realistic consideration set for your buyers. If you're a mid-market SaaS, don't compete with enterprise vendors — focus on other mid-market SaaS.
Step 2: Generate candidate queries
From multiple sources, compile a list of 200-500 candidate queries to test:
Source A: Your own competitor's content topics. Scrape their blog post titles, FAQ entries, case study topics. These are queries they've invested in answering. Some are likely earning them citations.
Source B: Your buyer's actual research questions. Pull from your sales team, support tickets, customer interview notes. What do buyers actually ask when evaluating your category? These are high-intent queries.
Source C: Industry association and publication content. What topics are being written about in industry media? These often become user queries.
Source D: Long-tail expansion. Take each core query and expand with modifiers: "how to", "best for", "vs competitor", "in China", "for small business", "enterprise", etc.
The goal is a wide candidate pool. You'll winnow it in later steps.
Step 3: Test queries across platforms
For each candidate query, send it to your target AI platforms (typically 3-5 platforms that match your buyer persona). For each response, capture:
- Which competitors are mentioned
- At what depth (primary recommendation, list item, incidental)
- What sentiment (positive framing, neutral, negative)
- What sources the response cites
This is where a rank tracker pays off. Manual testing of 500 queries across 5 platforms is 2,500 data points — slow to collect manually, straightforward with automation. See Building Your First AI Rank Tracker.
Expect about 30-50% of your candidate queries to produce substantive AI responses (vs. vague "consult an expert"). Those are the queries worth focusing on.
Step 4: Classify queries by opportunity
Now classify each query against two axes:
Axis 1: Competitive density
- High: 3+ competitors mentioned
- Medium: 1-2 competitors mentioned
- Low: No competitors mentioned (you could win these easily if you create content)
Axis 2: Your position
- Dominant: you're the primary recommendation
- Present: you're mentioned but not primary
- Absent: you're not mentioned at all
- Negative: you're mentioned negatively
The matrix tells you where to invest:
| Opportunity | Description | Investment priority |
|---|---|---|
| You absent + low competition | Easy win territory | High — produce content quickly |
| You absent + high competition | Crowded but real demand | Medium-high — produce content with strong differentiation |
| You present + dominant competitor | You're listed, a competitor is winning | High — displace through superior content |
| You dominant | Defend your position | Medium — maintain content freshness |
| You negative | Reputation issue | Urgent — address the negative framing |
| You absent + no substantive AI response | Queries AI doesn't answer well | Low — skip for now |
Step 5: Plan displacement content
For each priority query, design content that can displace the current citation. The playbook depends on what's currently cited:
If the current citation is a competitor's blog post: write a more comprehensive, more authoritative version. Include original data, clearer frameworks, broader competitor coverage.
If the current citation is a news article: publish definitive reference content on your own site that becomes the natural citation source for future queries on the topic.
If the current citation is a directory or association page: become more prominent in that directory/association (invest in the source itself rather than competing with it).
If the current citation is user-generated content (Xiaohongshu, Zhihu, etc.): engage the source ecosystem directly — community-building, genuine expert responses, organic presence.
Track displacement attempts monthly. Good displacement content typically takes 6-12 weeks to start surfacing in AI responses. Don't expect week-one lifts.
What the data typically shows
Our typical client running this process for the first time discovers:
- 20-40% of their "obvious target queries" are already being cited but without them
- 30-50% of queries they thought were competitive are actually low-competition (their assumption of competitor dominance was wrong)
- 10-20% of queries show surprising negative sentiment about their brand (requires immediate attention)
- 5-15% of queries produce no substantive AI response (opportunity to be the first)
The 10-20% with negative sentiment is often the most urgent finding. Brands routinely discover they have AI perception issues they were unaware of.
Tools and cost
DIY: a Python script running 500 queries across 5 platforms = ~$40-100 in API fees, plus 20-40 hours of analysis time.
Commercial tools: ChinaRankAI and similar platforms offer competitor reverse-engineering modules with pre-built query generation, classification, and displacement planning. Typical cost: ¥2,000-5,000/month.
For a one-time analysis, DIY is cheaper. For ongoing monitoring as you execute displacement content, commercial tools save substantial operational overhead.
Common mistakes
Over-broad competitor sets. Trying to compete with every brand in your category dilutes focus. Narrow to true realistic competitors.
Skipping the classification step. Without classifying queries by opportunity, you end up building content randomly. The classification is where the strategy emerges.
Expecting instant displacement. AI models update their retrieval slowly. New content takes weeks-to-months to surface. Don't quit if week 2 shows no lift.
Ignoring the negative sentiment findings. These often surface from one or two widely-cited negative articles. A counter-narrative content strategy can mitigate these more effectively than any amount of positive content.
Treating it as one-time. Competitor positioning shifts. Do this quarterly for active monitoring, annually for full re-analysis.
Case study: project management SaaS
A mid-market project management SaaS ran this process in Q2 2025. Results:
- 340 candidate queries tested, 187 produced substantive AI responses
- Competitor A (the market leader) cited in 62% of those queries
- Our client cited in 19%
- 28 queries had competitor A as primary citation but our client absent — identified as high-priority displacement targets
- 12 queries had negative sentiment about our client from a single outdated 2023 review — identified for counter-narrative content
Over the next 6 months, they produced displacement content for the 28 priority queries and counter-narrative content for the 12 negative-sentiment issues. By Q1 2026, their citation rate on the priority 28 queries rose to 47% (from 0%), and negative sentiment on the 12 queries decreased from -0.6 average to +0.1.
Total investment: approximately ¥240K in content plus ¥15K in measurement tooling over 9 months. Business impact: meaningful AI-sourced pipeline shift, particularly in enterprise-tier lead flow.
Reverse-engineering checklist
- Competitor set defined (5-10 realistic competitors)
- Candidate query list generated (200-500 queries)
- Queries tested across 3-5 target platforms
- Classification matrix completed
- Priority displacement queries identified
- Content production plan for top 20-40 queries
- Monthly tracking of displacement progress
- Quarterly re-analysis cycle established
Related reading
- Building Your First AI Rank Tracker
- Measuring AI Visibility: The 5 Metrics That Actually Matter
- 8 Content Formats Chinese AI Platforms Cite Most
About ByteEngine (杭州字节引擎人工智能科技有限公司)
ByteEngine provides competitor reverse-engineering as a core service. Our methodology, honed across 100+ brand engagements, identifies displacement opportunities that brands can realistically win with focused content investment. Learn more or check your brand's AI visibility.