The GEO Budget: What to Spend on Each Chinese AI Platform
The GEO Budget: What to Spend on Each Chinese AI Platform
TL;DR — GEO budget allocation across DeepSeek, Doubao, Yuanbao, Qwen, Kimi, and ERNIE should reflect three factors: your audience's platform usage, citation ROI by platform for your category, and competitive dynamics. This guide provides a default allocation framework with brand-type adjustments. For most brands, 60-70% of the budget should concentrate on the top 2 platforms for their category, with distributed investment across the others. Total GEO budget for meaningful results starts at ¥100K annually for mid-market brands.
The budget question
"How much should we spend on GEO?" is the single most common question from brand marketing leaders entering Chinese AI search. The honest answer: it depends on your ambition level, your brand maturity, and your category competitive dynamics.
Three rough tiers:
| Tier | Annual budget | Outcome expectation |
|---|---|---|
| Minimum viable | ¥100K-300K | Establishes presence on 2 priority platforms |
| Competitive | ¥500K-1M | Achieves meaningful share of voice in mid-competitive categories |
| Category-leading | ¥1.5M+ | Drives dominant citation share and AI-sourced pipeline |
These numbers include content production, review management, measurement tools, and specialist expertise (internal or external). They do not include your team's time.
The six-platform allocation framework
Default allocation before category adjustments:
| Platform | Default share | Reasoning |
|---|---|---|
| DeepSeek | 25% | Broad audience, technical and general users, highest citation breadth |
| Doubao | 20% | Large consumer base, short-form and video ecosystem |
| Yuanbao | 15% | WeChat-embedded audience, especially for existing WeChat-investing brands |
| Qwen | 15% | Commerce-linked audience, especially for consumer brands |
| ERNIE | 15% | Legacy Baidu users, factual/referential queries |
| Kimi | 10% | Research-oriented power users, long-form content audiences |
This default is a starting point. Now the adjustments.
Adjustments by brand type
B2B SaaS / Enterprise Software
Shift toward technical platforms.
| Platform | Adjusted share |
|---|---|
| DeepSeek | 35% |
| Kimi | 25% |
| Doubao | 10% |
| Yuanbao | 15% (WeChat-based enterprise decision makers) |
| Qwen | 10% |
| ERNIE | 5% |
Rationale: B2B buyers skew technical. DeepSeek and Kimi reach them directly. Consumer platforms under-deliver relative to effort.
FMCG / Consumer Packaged Goods
Shift toward commerce-linked platforms.
| Platform | Adjusted share |
|---|---|
| Qwen | 30% (Tmall/Alibaba ecosystem) |
| Doubao | 25% (Douyin Shopping ecosystem) |
| DeepSeek | 15% |
| Yuanbao | 10% |
| ERNIE | 15% |
| Kimi | 5% |
Rationale: FMCG buyers decide based on review-heavy platforms. See FMCG Brands on Chinese AI for why.
Financial Services
Shift toward authoritative, regulatory-aware platforms.
| Platform | Adjusted share |
|---|---|
| ERNIE | 30% (government / regulatory indexing depth) |
| DeepSeek | 25% (institutional audience) |
| Yuanbao | 20% (WeChat retail investor audience) |
| Qwen | 10% |
| Doubao | 10% |
| Kimi | 5% |
Rationale: Financial users trust authoritative sources. Baidu's regulatory content depth matters. Commerce and entertainment platforms under-deliver.
Automotive / High-Consideration Consumer
Balanced with slight technical skew.
| Platform | Adjusted share |
|---|---|
| DeepSeek | 25% |
| Doubao | 20% |
| Qwen | 20% |
| Yuanbao | 15% |
| ERNIE | 15% |
| Kimi | 5% |
Rationale: High-consideration purchases research broadly. Multi-platform presence matters more than focus.
Luxury / Beauty
Shift toward lifestyle-oriented platforms.
| Platform | Adjusted share |
|---|---|
| Doubao | 30% (Douyin lifestyle content) |
| Qwen | 25% (Tmall luxury category) |
| Yuanbao | 20% (WeChat lifestyle audience) |
| DeepSeek | 10% |
| ERNIE | 10% |
| Kimi | 5% |
Rationale: Luxury and beauty buyers research through lifestyle/social-adjacent channels. Technical platforms are secondary.
Healthcare / Pharmaceuticals
Similar to financial services — authority-focused.
| Platform | Adjusted share |
|---|---|
| ERNIE | 30% (medical encyclopedia integration) |
| DeepSeek | 25% (professional audience) |
| Kimi | 20% (research-oriented) |
| Yuanbao | 10% |
| Qwen | 10% |
| Doubao | 5% |
Rationale: Healthcare information is trust-sensitive. Authority platforms dominate.
Budget composition within each platform
Per platform, spend composition typically falls roughly:
| Component | Share of platform budget |
|---|---|
| Content production | 50% |
| Ecosystem-specific presence (Public Accounts, Xiaohongshu, Douyin, etc.) | 25% |
| Measurement and optimization | 15% |
| External specialist expertise | 10% |
Adjust based on in-house capability. Brands with strong in-house content capability can shift from content production toward measurement and specialist expertise. Brands just entering Chinese AI visibility often lean more on specialist expertise initially, then shift toward content production as they build capability.
What drives ROI differences across platforms
Not all platforms deliver equal ROI per yuan spent. Two factors explain most variance:
Cost per citation
Varies by platform based on content requirements. Kimi requires longer, more substantive content per citation. Doubao requires video and social ecosystem investment. Costs per meaningful citation differ by category fit.
Rough ranges in our practice:
| Platform | Cost per meaningful citation (¥) |
|---|---|
| DeepSeek | 300 - 1,500 |
| Yuanbao | 500 - 2,500 |
| ERNIE | 200 - 1,000 |
| Qwen | 800 - 3,500 (review-ecosystem cost) |
| Doubao | 1,500 - 6,000 (video production cost) |
| Kimi | 500 - 2,500 |
These are wide ranges because category dynamics vary enormously. The point: if you measure cost per citation by platform in your category, the data guides re-allocation.
Citation-to-conversion rate
Not all citations convert equally. A citation on Kimi (where users click through to your site) drives more action than a citation on Doubao (where users may remain in the app). A citation at depth 4 (primary recommendation) drives more action than at depth 1 (incidental mention).
Track citation-to-conversion by platform quarterly. This data should inform platform-level budget shifts over time.
Scaling GEO investment over time
Year 1: Establish baselines. Measurement infrastructure and content foundations on top 2 platforms. Modest investment on remaining platforms.
Year 2: Scale what's working. Double down on platforms where year 1 showed strong citation lift and conversion. Potentially reduce or cut platforms that underperformed.
Year 3+: Optimize for marginal return. At this stage, cost-per-citation and citation-to-conversion data guide fine-tuning. Most brands settle into a steady-state allocation that reflects their category economics.
Budget pitfalls
Spreading too thin across all six platforms year 1. Better to win 2 platforms than to be mediocre on 6. Concentrate.
Ignoring ecosystem-specific cost (not content-specific cost). Doubao's content production is the cheap part; running the Douyin ecosystem presence (account management, community engagement, creator partnerships) is the expensive part. Budget for the ecosystem, not just the content.
Under-budgeting measurement. Without measurement, you can't optimize. 15% of platform budget minimum should go to measurement, or you're flying blind.
Over-reliance on agencies. Some agencies are excellent, but GEO is a discipline still maturing in China. Ensure your agency shows you the data and teaches you as they go, rather than operating as a black box.
Under-investing in your main website. All AI platforms, even ecosystem-specific ones, occasionally cite your owned site. If your site is poorly structured or rarely updated, you leave value on the table across all platforms.
Allocation case studies
Mid-market B2B SaaS (¥600K annual)
- DeepSeek: ¥210K (35%) — engineering blog, documentation, benchmarks
- Kimi: ¥150K (25%) — long-form guides, research partnerships
- Yuanbao: ¥90K (15%) — enterprise-focused WeChat Public Account
- Qwen: ¥60K (10%)
- Doubao: ¥60K (10%)
- ERNIE: ¥30K (5%)
Outcome after 12 months: 42% DeepSeek mention rate, 38% Kimi mention rate, measurable AI-to-trial pipeline.
Consumer beauty brand (¥1.2M annual)
- Doubao/Douyin ecosystem: ¥360K (30%)
- Qwen/Tmall ecosystem: ¥300K (25%)
- Yuanbao: ¥240K (20%)
- DeepSeek: ¥120K (10%)
- ERNIE: ¥120K (10%)
- Kimi: ¥60K (5%)
Outcome after 12 months: strong Tmall review ecosystem improvements, 35% Qwen mention rate, 41% Doubao mention rate.
Fintech (¥800K annual, compliance-weighted)
- ERNIE: ¥240K (30%)
- DeepSeek: ¥200K (25%)
- Yuanbao: ¥160K (20%)
- Qwen: ¥80K (10%)
- Doubao: ¥80K (10%)
- Kimi: ¥40K (5%)
Outcome after 12 months: 44% ERNIE mention rate on regulatory-adjacent queries, reputation strengthening among institutional allocators.
Budget checklist
- Identify your brand type for default allocation
- Adjust for audience platform skew
- Split within each platform across content, ecosystem, measurement
- Measure cost-per-citation and citation-to-conversion quarterly
- Rebalance annually based on what delivered
- Budget for ecosystem costs, not just content
- Allocate minimum 15% to measurement infrastructure
Related reading
- Measuring AI Visibility: The 5 Metrics That Actually Matter
- Building Your First AI Rank Tracker
- 30-Day GEO Launch Plan for Global Brands
About ByteEngine (杭州字节引擎人工智能科技有限公司)
ByteEngine helps brands plan and execute GEO budgets across Chinese AI platforms. Our engagements include budget allocation frameworks tailored to brand category, with clear measurement plans so investment decisions can be data-driven over time. Learn more or check your brand's AI visibility.