After over a decade in the testing and certification industry, I've seen some absurd cases. One agency spent RMB 40,000 outsourcing an FCC guide – the final piece was a history of the Federal Communications Commission – no mention of the actual test process – completely useless for clients.
From 2025, large language models have gradually entered the industry. Test automation is still difficult to scale – RF spurious determination and safety fault analysis still rely on engineer judgement. But content creation and document organisation have been significantly impacted. Top certification bodies have long had LIMS systems – some test automation in place. But most smaller agencies still rely on offline exhibitions and client referrals – their content strategy is chronically underdeveloped.
In B2B, certification is widely acknowledged as one of the toughest content areas.
First, the technical barrier is high. To explain Bluetooth BQB properly, you need to understand SIG membership rules, the differences between QDID, DID, and the now‑deprecated DN numbering – and the underlying test logic. A technical error – peers spot it immediately – reputation damage. But if you write too shallow – search engines won't give you traffic.
Second, the coverage is fragmented. Bluetooth, Wi‑Fi, FCC, CE, CarPlay, HDMI, DAB+, Mexico NOM – dozens of certification areas – each needing content. One person simply cannot master all categories.
Third, standards update too fast. When a regulation version changes – test parameters, certification processes, and compliance deadlines all shift – old articles become invalid – requiring mass updates. Even internal engineers struggle to keep up – external writers are even less equipped.
1. Traditional content production models – no longer sufficient
Building a dedicated writing team covering all certification categories – labour costs are prohibitive. The certification industry is unique: practitioners who both understand testing compliance and write well are extremely rare. The relatively practical approach – internal engineers writing part‑time. But engineers' core work is reports and project testing – writing is an extra burden – output is highly unstable.
I've also tried external industry writers – the best one lasted only two months. His honest feedback: the knowledge density is too high – continuous output is unsustainable.
二、AI Tools – Where Can They Plug Into Content Production?
AI's most obvious value: research aggregation and first‑draft writing.
Previously, writing an FCC guide meant reading FCC Part 15 original text, extracting test procedures from TCB public materials, gathering industry cases – dozens of browser tabs open simultaneously – half a day just to organise materials. Now: upload the regulation PDF to AI – quickly extract core test items, limit values, certification processes – structured outline in 10–15 minutes – then human review to correct AI misinterpretations.
Be highly cautious – AI makes basic technical errors. It frequently confuses SDoC vs. Certification scope, old vs. new standard numbers, certificate validity rules – even the boundary between RED and LVD. I've seen an AI‑generated CE guide that incorrectly listed Module A as Module B – if published, foreign trade factories would plan certification based on it – and get it completely wrong.
1. Multi‑version content – bilingual first drafts simultaneously
The same technical knowledge needs to be presented differently for different audiences:
·R&D: test standards, compliance paths.
·Procurement: timelines, cost comparisons.
·Sales: market‑friendly explanations.
With the same factual base, AI can quickly generate multiple first‑draft versions – then human refinement.
Bilingual content can also leverage AI. After the Chinese draft is finalised, AI generates an English first draft – human verification of professional terminology. The certification industry has many globally consistent terms: DoC, TCF, Notified Body, QDID, Module A. General translation models often make errors – e.g., translating SDoC as "supplier declaration" or Notified Body as "notification body." Best practice: build a company‑specific terminology base to constrain AI output – don't rely on general models alone.
三、AI Cannot Replace Professionals – Only Reduce Repetitive Work
AI is an assistive tool – not a substitute for domain expertise. EN 62368-1 creepage distances for different pollution degrees, FCC Part 15E UNII‑2 DFS detection thresholds, EN 18031 TLS version compliance criteria – these must be verified by qualified engineers. AI knowledge bases are outdated, biased, and prone to confusing different standard versions.
Human final review is non‑negotiable.
In practice, many agencies chase output volume and compress review processes. AI first‑draft – light proofreading – publish. Engineer skims and approves. I'm aware of three agencies that have been caught by this. In certification, content tolerance is extremely low – incorrect compliance guidance misleads exporters – leading to customs detention, flawed certification planning, and financial loss.
Simple "suggestions" are not enough – companies must establish a hard process: no professional final review – no publication.
1. Where does the engineer's time go?
AI's real value is taking over 80% of repetitive work. Previously, an engineer spent about two days on an article – 1.5 days on research, formatting, terminology checking, regulation‑number verification – only half a day on professional judgement. With AI handling material organisation, formatting, translation, and rewriting – engineers focus only on technical review – output efficiency improves significantly. But: review time cannot be compressed – AI optimises prep work – it does not reduce professional judgement investment.
To put it bluntly: if your agency prioritises article volume and compresses review time – AI will only bring more risk.
四、Available Tools and Implementation Pitfalls
Current mainstream solutions fall into three categories:
1. Conversational LLMs
Domestic models like DeepSeek – suitable for material organisation, outlines, first drafts, and document translation. Advantages: stable general capability, handles professional terminology, long‑context windows for full regulation documents. Claude and GPT‑4o – not yet commercially compliant in China – cross‑border access carries network and data‑leakage risks. For domestic agencies, don't blindly chase overseas models – prioritise compliant access conditions.
2. Local deployment + private knowledge bases
Large certification bodies are deploying these. Build a private knowledge base with historical test reports, certification cases, and standard documents – paired with RAG retrieval. Engineers querying – can directly reference standard texts and past project cases. High technical barrier – small and medium agencies struggle to implement.
Long‑term maintenance costs are easily overlooked: standards are continuously updated, old vs. new document version control, expired standards clean‑up. If the knowledge base retains outdated materials – RAG will continue to output incorrect parameters. I'm aware of one lab that spent six months building a knowledge base – imported three old IEC standards – every limit value the system output was wrong – by the time they discovered it, they had already issued over a thousand reference conclusions.
3. Industry SaaS products – be realistic
Some domestic labs have developed internal tools to automatically generate simple certification proposals for standard products – but these are not commercially available as standardised SaaS. Many products marketed as "vertical AI SaaS for certification" are just general Q&A tools with industry marketing hype – they cannot automatically perform compliance gap analysis or generate complete certification plans. Don't be misled by presentation slides.
For AI in testing and certification, contact BlueAsia at 13534225140 (King) or email king.guo@cblueasia.com.
Related News
