Clear instructions decide whether artificial intelligence earns its keep
Most businesses using artificial intelligence see no financial return, and unclear instructions are often why. This article explains how to write instructions and examples an AI model can actually follow.
Artificial intelligence (AI) does not know what you want. It only knows what you tell it, and what examples you show it. Writing clear instructions, and good examples of the output you expect, is one of the most practical skills in using an AI model well. It is not technical. It is closer to briefing a new member of staff than to programming.
Think of instructions as a brief, and examples as a sample of the work you expect back. If you ask an AI model to write a customer reply but do not say how long it should be, what tone to use, or what information must appear, you leave it to guess. It will guess. The result may look plausible and still be wrong for your business.
This matters more than it sounds. A survey of chief executives found only 12% reported both revenue growth and cost reductions from AI, while 56% reported no significant financial benefit from it at all (SiliconANGLE). Some of that gap comes from unclear instructions: an AI model given a vague brief produces a vague result, and a business then concludes the technology does not work.
Where it goes wrong
The most common mistake is to assume the AI model already knows your business. It does not, unless you tell it every time, or unless it is connected to a system that reliably supplies that context. Even among businesses that already run a shared system for that purpose, only 23% say it is the main source their AI agents rely on for business context (Venturebeat). For most businesses, the AI model is working from whatever instructions and documents it was given on the day, and nothing more.
A second mistake is writing instructions that only work once. A business tests an idea on one task, it works, and the instructions are never written down properly or tried against a second case. Across organisations piloting or deploying AI agents, 81% are doing so, yet only 19% have scaled beyond a single use case (SiliconANGLE). Instructions that are not written clearly, and not checked against a few different examples, rarely survive contact with a second use.
An AI model does exactly what its instructions say, nothing more and nothing less.
What to ask a supplier
Ask to see the written instructions behind any assistant or tool you are buying, not just a demonstration of it working once. Ask what examples were used to test it, and whether those examples cover the awkward, real cases from your business, not only the easy ones. Ask what happens when your business changes, a new product, a new policy or a new form of words. Ask whether the instructions get updated, or quietly go stale.
Malaysia already has an advantage here. A 2026 survey of 2,000 Malaysian knowledge workers found 24% qualify as advanced "frontier" users of AI, against 16% of knowledge workers worldwide (The Vibes). That edge is built on exactly this skill: writing what you want clearly enough that the AI model can actually deliver it.
At Yunaris, every assistant we build starts with written instructions and real examples from your business, tested against the awkward cases before the easy ones. Good instructions are a sentence you can read back and agree with, not a secret.
Sources
- Human judgment reshapes AI services, SiliconANGLE, 11 October 2026
- 64% of enterprises find AI errors in data, Venturebeat, 6 October 2026
- AI in customer experience has an orchestration problem, not an adoption problem, SiliconANGLE, 3 October 2026
- Malaysian workers race ahead of employers in AI adoption, The Vibes, 20 September 2026
A question this article does not answer? Email hello@yunaristech.com.