Why AI Detection Matters to Your Business
AI tools are quickly becoming part of everyday business writing, from marketing captions and product descriptions to customer replies, proposals and internal documents. At the same time, a new concern is emerging: how can you tell whether content was written by a person, generated by software, or heavily edited by both?
A recent WIRED report examined Pangram, an AI-detection startup that has attracted attention after its software flagged books, articles and prize-winning stories as potentially AI-generated. The company reportedly has 24 employees and has raised US$13 million, according to the report. Its rise reflects a wider business issue that you may soon face: AI detection scores can influence trust, hiring, publishing, education and business relationships.
For a Malaysian SME, the lesson is not simply “use an AI detector”. The more important lesson is to create a sensible policy for AI-assisted work without treating an automated score as final proof.
What Happened
Pangram analyses text and produces an estimated percentage indicating how much AI may have been involved. According to WIRED, the company’s technology uses “synthetic mirroring”, where human writing is compared with text generated by language models. It also uses “hard negative mining”, meaning that examples of false positives are added to training so the system can improve.
The company became widely known after its software gave a high AI-generated score to Shy Girl, a novel by Mia Ballard. The author denied using AI, but the book’s planned release was later canceled by Hachette, according to the article. Other disputed cases followed, including a New York Times Modern Love column, a Commonwealth Short Story Prize winner, and several novels.
WIRED also reported that Substack integrated Pangram so readers could check possible AI use. However, writers and publishing professionals have raised concerns about false accusations and the consequences of relying on a single automated result. Jane Friedman, an author and publishing expert quoted in the article, described strong anger toward AI-detection software.
The report makes clear that Pangram itself does not claim to provide an absolute answer. Its result is a best-guess percentage. That distinction matters because a percentage may look precise while still requiring human review and context.
Why This Matters for Malaysian SMEs
You may already be using AI to draft social media posts, translate product information into Bahasa Malaysia or English, prepare sales emails, summarise meetings, or respond to common customer questions. Your staff may also be using AI independently. Without a clear process, you may not know which documents contain AI assistance, whether confidential information was entered into an external tool, or whether a supplier’s content is original.
Consider a local online retailer that asks an AI tool to create 100 product descriptions. The descriptions may be grammatically correct but contain incorrect specifications, exaggerated claims or unsuitable Bahasa Malaysia. An AI detector may identify the text as likely machine-generated, but that does not tell you whether the copy is accurate, legally safe or suitable for Malaysian customers. Human checking remains necessary.
The same applies to recruitment. If you use an AI detector to assess a candidate’s cover letter, a high score should not automatically disqualify that person. A candidate may have used software for grammar correction, translation or restructuring. English may not be their first language. A detector can support a review, but it should not replace an interview, work sample or reference check.
For agencies and professional-service firms, the issue can affect client trust. A client may ask whether a proposal, report or campaign was created with AI. You should be able to explain what tools were used, what information was supplied, and what review took place. Transparency is more useful than making an unsupported claim that work is “100 percent human”.
| Business area | Possible risk | Practical response |
|---|---|---|
| Marketing | Incorrect or generic AI-written claims | Check facts, tone, brand rules and local context |
| Recruitment | Rejecting applicants based on an uncertain score | Use interviews and work samples as primary evidence |
| Client work | Disputes about originality or disclosure | Record AI use and agree on expectations in writing |
| Operations | Confidential data entering public AI tools | Set rules for personal data, contracts and customer records |
Do Not Treat a Detector as a Judge
An AI-detection score is an indicator for further review, not a verdict about a person, employee or supplier.
This principle is especially important for smaller companies, where one hiring decision or client dispute can have a major operational impact. AI detectors can produce false positives, particularly when writing is formal, highly edited, translated, short, or produced by someone who does not normally write in English. They can also miss AI-generated text that has been substantially rewritten.
If a document receives a high score, ask practical questions. Who created it? What instructions were given? Were company facts checked? Are there previous drafts, notes or source materials? Can the writer explain the reasoning behind the recommendations? These questions provide stronger evidence than a percentage alone.
A Sensible AI Policy for Your Team
You do not need a complicated policy. A one-page internal guide can establish a consistent approach. State which tasks may use AI, which tasks require approval, and which information must never be entered into a public tool. Customer identification details, passwords, unpublished financial information, private contracts and sensitive employee records should be handled carefully.
Ask employees to keep a simple record for important work: the tool used, the purpose, the main prompt or instruction, and the human checks completed. This creates an audit trail without slowing down routine work. For marketing and sales content, require a fact check. For legal, financial, medical or safety-related content, require review by an appropriately qualified person.
You should also decide when disclosure is appropriate. A customer generally needs to know if an automated system is handling their enquiry or making a decision about them. A client may require disclosure if AI is used to produce deliverables. The correct approach depends on the relationship, the task and your agreement.
The Bigger Picture
The Pangram story shows that AI detection is becoming part of the trust infrastructure around digital content. WIRED reported that Pangram serves industries including education, legal work and recruitment, while creative writing remains a major part of its training text. The demand exists because people want to distinguish human judgment from automated output.
But detection alone will not solve the underlying problem. Businesses need provenance, review and accountability. Instead of asking only whether a document was written by AI, ask whether the information is correct, whether the recommendation is responsible, whether confidential data was protected, and whether someone in your organisation is willing to stand behind the result.
For Malaysian SMEs, this is an opportunity to compete through reliable processes. You can use AI to speed up drafting while keeping human responsibility at the centre. Train your team to edit for Malaysian language, culture and customer expectations. Keep records for important work. Use detection tools cautiously, and never let an automated percentage make a serious decision on its own.
AI-assisted work is likely to become normal across business. The companies that earn trust will not necessarily be those that avoid AI. They will be the ones that use it openly, check its output carefully and make sure you remain accountable for what reaches your customer.
Ready to Streamline Your Operations?
Technology moves fast. Your operations should keep up. AutoRunBiz builds AI systems that run your daily workflows — from WhatsApp order capture to accounting. Book a free 15-min ops audit →
