一、From the Perspective of GPL Licensing
Core Legal Structure
WordPress is GPL v2+, and all plugin code must be GPL-compatible. However, GPL has a crucial boundary:
GPL Scope of Constraint: Distributed software code (PHP/JS/CSS)
GPL Does Not Constrain: Network services (SaaS/API)
This is the commercial plugin’s “Open Core” model:
┌───────────────┬───────────────┬───────────────────────┐
│ Layer │ GPL-Constrained? │ Explanation │
├───────────────┼───────────────┼───────────────────────┤
│ Plugin PHP Code │ 
Yes │ Can be freely modified, distributed │
├───────────────┼───────────────┼───────────────────────┤
│ Plugin JS/CSS │ Yes │ Same as above │
├───────────────┼───────────────┼───────────────────────┤
│ Backend AI Service │ 
No │ SaaS is not “distributed software” │
├───────────────┼───────────────┼───────────────────────┤
│ API Protocol/Format │ No │ Belongs to service interface │
└───────────────┴───────────────┴───────────────────────┘
Key Inferences:
- Users have the right to modify GPL plugin code to connect it to different AI backends — this is a legal right granted by the GPL.
- However, a commercial plugin’s ToS (Terms of Service) may prohibit reverse engineering and bypassing the service — this is a constraint at the contract law level.
- The conflict between GPL and ToS is a gray area, and the WordPress community has ongoing debates about this.
Impact on AI Gateway
- Layer 1 (OpenAI-Compatible Layer): Intercepts requests sent to api.openai.com; these are requests using the user’s own API Key, completely legal.
- Layer 2 (Override Forwarding): Intercepts requests sent to api.yoast.com / api.rankmath.com; legally permissible under GPL, but may violate ToS.
二、From the Perspective of Commercial Plugin Companies
Why Bind AI as SaaS
┌──────────┬─────────────────────────────────────────┐
│ Motivation │ Explanation │
├──────────┼─────────────────────────────────────────┤
│ Anti-Piracy │ Code can be copied under GPL, but SaaS service cannot │
├──────────┼─────────────────────────────────────────┤
│ Recurring Revenue │ Subscription + Credit System = Predictable Cash Flow │
├──────────┼─────────────────────────────────────────┤
│ Quality Control │ Own prompt tuning, model selection ensures output quality │
├──────────┼─────────────────────────────────────────┤
│ Data Barrier │ User usage data helps optimize product │
├──────────┼─────────────────────────────────────────┤
│ Cost Transfer │ AI API costs included in subscription fee, invisible to users │
└──────────┴─────────────────────────────────────────┘
AI Business Models of Various Plugins
Rank Math Content AI:
- Credit-based system, free users get a small amount of credits (~5000 credits/month)
- Pro users get more credits (~15000 credits/month)
- Additional credits need to be purchased separately
- Backend uses OpenAI, users cannot choose
Elementor One (newly launched in Jan 2026):
- Unified subscription system, includes AI credits
- AI features cover copywriting, image generation, code generation
- When credits run out, wait for next month or purchase additional packs
- Deeply integrated with Elementor Connect account system
Yoast SEO Premium:
- AI features built into Premium subscription
- Generates SEO title, description suggestions
- Backend routes through Yoast’s own API servers
Defense Strategies of These Companies
- Account Verification — Every AI request verifies the License.
- Credit System — Credits deducted server-side, cannot be forged client-side.
- Response Signing — Some plugins verify response source.
- Format Privatization — Use non-standard request/response formats.
- Frequent Updates — Changing API formats to break adapters.
三、From the Perspective of Paying Users
Current Experience Pros and Cons
Pros:
- Zero configuration, works out of the box.
- No need to understand API Key concepts.
- Deep integration with plugin UI, smooth experience.
Cons:
┌────────────┬──────────┬───────────────────────────────────┐
│ Pain Point │ Severity │ Explanation │
├────────────┼──────────┼───────────────────────────────────┤
│ Double Payment │ 
│ Pay plugin subscription, AI requires extra credit fees │
├────────────┼──────────┼───────────────────────────────────┤
│ Credit Anxiety │ │ Run out and it’s gone, limits creative freedom │
├────────────┼──────────┼───────────────────────────────────┤
│ Lock-in Effect │ │ Cannot choose cheaper/better AI services │
├────────────┼──────────┼───────────────────────────────────┤
│ Opaque Quality │ │ Don’t know what backend model is used, cannot optimize │
├────────────┼──────────┼───────────────────────────────────┤
│ No Customization │ │ Prompt templates are fixed │
└────────────┴──────────┴───────────────────────────────────┘
What Users Truly Want
Ideal State:
- Pay once, use AI features unlimitedly.
- Can choose their own trusted AI service.
- Fast response, no lag.
- High generation quality, matches their own language and style.
四、From the Perspective of Chinese Users (Most Critical)
Real-World Dilemma
The experience of Chinese users using WordPress AI plugins:
Rank Math Content AI → Request sent to api.rankmath.com → May timeout/be blocked
↓
Backend calls OpenAI → Cannot be directly accessed from China
↓
Result: Paid for it, function unusable
Elementor AI → Request sent to my.elementor.com → High latency
↓
Backend calls OpenAI/Anthropic → Same as above
↓
Result: Generation extremely slow or times out
BYOK Plugins (AI Engine) → Requires OpenAI API Key → Cannot register from China
↓
Result: Function completely unavailable
Core Needs of Chinese Users
┌────────┬────────┬────────────────────────────────┐
│ Need │ Priority │ Explanation │
├────────┼────────┼────────────────────────────────┤
│ Usable │ P0 │ Most basic — AI features must work normally │
├────────┼────────┼────────────────────────────────┤
│ Fast Speed │ P1 │ Domestic servers, low latency │
├────────┼────────┼────────────────────────────────┤
│ Good Chinese │ P1 │ Domestic models’ Chinese capability far exceeds GPT │
├────────┼────────┼────────────────────────────────┤
│ Cheap │ P2 │ DeepSeek cost is 1/50 of GPT-4 │
├────────┼────────┼────────────────────────────────┤
│ Compliant │ P2 │ Data doesn’t leave the country, complies with domestic regulations │
└────────┴────────┴────────────────────────────────┘
Advantages of Domestic AI
DeepSeek V3:
- Chinese comprehension ≈ GPT-4o
- Cost: ¥1/million tokens (1/50 of GPT-4o)
- Servers in China, latency < 200ms
- No need for VPN, direct access
Tongyi Qianwen:
- Alibaba Cloud infrastructure, extremely high stability
- Strong multimodal capabilities
- 128K context
Compared to OpenAI:
- Cannot be directly accessed from China
- Requires proxy, increases latency and cost
- Chinese capability not as good as domestic models
- Price 50 times more expensive
五、Final Judgment on the AI Gateway Three-Layer Plan
┌───────────────────────┬──────────┬──────────┬──────────┬──────────┬─────────────┬─────────────┐
│ Layer │ Legal Risk │ Ethical Risk │ Technical Risk │ Maintenance Cost │ User Value │ Conclusion │
├───────────────────────┼──────────┼──────────┼──────────┼──────────┼─────────────┼─────────────┤
│ Layer 1 OpenAI Compatible Layer │ 
None │ None │ Low │ Low │ Extremely High │ Do │
├───────────────────────┼──────────┼──────────┼──────────┼──────────┼─────────────┼─────────────┤
│ Layer 2 Override Forwarding │ 
Medium │ 
High │ High │ Extremely High │ Limited │ Don’t do │
├───────────────────────┼──────────┼──────────┼──────────┼──────────┼─────────────┼─────────────┤
│ Layer 3 UI Injection │ None │ None │ Medium │ Medium │ Medium │ Future Option │
└───────────────────────┴──────────┴──────────┴──────────┴──────────┴─────────────┴─────────────┘
Why Layer 1 is Worth Doing
BYOK Plugin Users (AI Engine, AI Power, GPT3 AI Content Generator, etc.):
Current: User → Enters OpenAI Key → Request sent to api.openai.com → Unusable in China
WPMind: User → Enables Gateway → Request routed to DeepSeek/Tongyi Qianwen → Works normally
- User’s active choice, does not involve bypassing any paid services.
- OpenAI API format is an open standard, not a private protocol.
- For Chinese users, it’s a qualitative change from “completely unusable” to “perfectly usable”.
Why Layer 2 Shouldn’t Be Done
- Ethical Issue — Essentially bypassing commercial plugins’ paid AI services. Even if GPL allows code modification, this is not good business practice.
- Maintenance Nightmare — Every time Yoast/Rank Math updates API formats, adapters may break.
- ROI Extremely Low — Invest significant development and maintenance resources, but covers limited user scenarios.
- Brand Risk — Viewed as “parasitic” on other plugins, affects WPMind’s reputation in the community.
Better Alternative Strategies
Instead of doing Layer 2 to “replace” commercial plugins’ AI services, it’s better to:
Strategy 1: Develop our own AI features well.
WPMind Content Creation Assistant → Directly replace Rank Math Content AI functions.
WPMind GEO Suite → Stronger SEO optimization than Yoast AI.
Strategy 2: Expand through official ecosystem.
WordPress AI Building Blocks → WPMind injects as a Provider.
Official Filter Hooks → Seamless integration of domestic models into all plugins using the official API.
Strategy 3: SEO Plugin Extension Packs (Legitimate Ecosystem Cooperation).
WPMind GEO for Rank Math → Extends Rank Math, doesn’t replace it.
WPMind GEO for Yoast → Same as above.
Summary: AI Gateway only does Layer 1 (OpenAI Compatible Layer). This is a legally clear, ethically uncontroversial, technically simple, and immensely valuable solution for Chinese users. Layer 2 should be abandoned. WPMind’s long-term value lies in its own AI capabilities and deep integration with the WordPress official ecosystem, not in being parasitic on other commercial plugins.