The Business Prompt Playbook  ·  Article 4 of 8

The Hidden Variables That Change Your AI Results

By Jane Chew  ·  AI Strategy Coach, DigitalAI Business Club  ·  May 2026

You write the same prompt twice and get two noticeably different answers. Or you follow someone else’s prompt example and your output looks nothing like theirs. This is not inconsistency — it is the hidden layer of variables that sit underneath prompt structure. Once you understand them, you have real control.

Variable 1: Temperature — creativity vs precision

Temperature controls how creative or predictable the AI’s output is. You typically cannot set this directly in standard chat tools, but your prompt influences it powerfully.

Strict, specific prompts — clear format, tight word counts, explicit rules — push the AI toward low-temperature behaviour: factual, logical, predictable. Open-ended, flexible prompts invite high-temperature behaviour: creative, varied, exploratory.

For business tasks that need consistent, accurate output — reports, client communications, proposals — use structured prompts with precise constraints. For creative tasks — brainstorming, naming, campaign ideas — keep the prompt open and minimal.

Variable 2: Randomness — why outputs vary

AI models include randomness by design. This is what prevents them from producing the exact same answer every time — and what makes them useful for creative tasks. But for business outputs that need consistency, it can feel frustrating.

You reduce randomness by adding fixed rules to your prompt: “Follow this format exactly,” “Do not add ideas I have not mentioned,” “Use only the information I have provided.” Each constraint tightens the possible range of outputs.

Variable 3: Model type — not all AI is the same

ChatGPT, Claude, Gemini, and Perplexity are all AI tools — but they are trained differently, on different data, with different strengths. The same prompt on different models will produce noticeably different outputs.

Claude tends to produce longer, more nuanced written content. ChatGPT is strong on structured analysis and step-by-step reasoning. Gemini integrates naturally with Google Workspace. Perplexity sources current web information with citations.

For most business writing and strategy tasks, test your specific use case on two models before committing to one. The time investment is small; the quality difference can be significant.

Variable 4: Input length and structure

Longer prompts are not always better. When a prompt is long but disorganised, the AI loses the thread. The quality of output drops not because the prompt is long — but because the signal-to-noise ratio is poor.

For long prompts, use clear labels (ROLE, TASK, CONTEXT, OUTPUT) and break information into short sections. The AI processes structure better than paragraphs of mixed information. Place the most important instructions at the end — the final instruction carries the most weight.

Variable 5: Instruction strength

Not all instructions inside your prompt carry equal weight. Instructions at the end of your prompt have the highest priority. Specific, measurable instructions (“under 80 words”) outperform vague ones (“keep it short”). Direct language (“Do not use bullet points”) is stronger than suggestions (“Try to avoid bullet points”).

When you have a critical constraint — a word limit, a tone rule, a format specification — place it in the output section at the end of your prompt. This is where the AI is paying the most attention.

Variable 6: How vagueness gets punished

When a prompt is ambiguous, the AI does not ask for clarification — it makes a choice. And the choice it makes is the one that feels safest and most broadly acceptable. This produces the generic, flat outputs that frustrate most business users.

Clarity does not just improve quality — it unlocks depth. A well-specified prompt signals to the AI that it is dealing with a specific, complex request. The AI responds with greater detail, more precise reasoning, and better structure.

Clarity equals quality. Vagueness equals safe but shallow.

In Article 5, we move to multi-step prompting — how professionals build prompt workflows that produce far better results than any single instruction can achieve alone.

The Business Prompt Playbook  ·  Article 5 of 8

Multi-Step Prompting: How to Stop Getting One-Shot Answers

By Jane Chew  ·  AI Strategy Coach, DigitalAI Business Club  ·  May 2026

One prompt trying to do everything usually does everything poorly. When a task is complex — a proposal, a strategy document, a detailed content plan — asking for the full output in a single instruction pushes the AI to skip steps, make assumptions, and produce something that needs heavy editing. The fix is a workflow, not a longer prompt.

Why multi-step works when one-shot fails

AI performs best when each instruction is clean, focused, and contains one job. A complex task — “write me a full business proposal including background, scope, pricing, and a timeline” — is not one job. It is five jobs stacked into one instruction.

The AI will attempt all five simultaneously, which means it makes trade-offs. It may nail the scope but produce a generic background. It may format the pricing clearly but rush the timeline. Multi-step prompting removes these trade-offs by separating each job into its own instruction — and letting the output of each step inform the next.

This is not about using more prompts. It is about using smarter ones in a logical sequence.

Step 1: Clarifying questions

Before the AI starts any complex task, ask it to clarify. This one instruction prevents the majority of wasted drafts.

The AI will surface the gaps in your brief — and your answers will provide exactly the context it needs to produce relevant, specific work. The AI stops guessing and starts delivering.

Step 2: Outline before execution

After the clarifying step (or after you have written a strong brief), ask for an outline only — not the full draft.

Reviewing an outline takes 30 seconds. Catching a structural problem at this stage saves you from editing a full draft later. Approve the outline, suggest changes, or redirect — then proceed to the draft with confidence.

Step 3: First draft

With the outline approved, you instruct the AI to write the full draft. Now it has a confirmed structure to follow — and the output is significantly cleaner than anything produced without that foundation.

Step 4: Self-review and improvement

This is the step most people skip — and it is the one that makes the biggest difference. After the draft is produced, you ask the AI to evaluate its own work.

The AI will identify its own gaps — repetition, weak transitions, missing specifics — and produce a second draft that addresses them. This single step eliminates the need for most manual editing.

The full workflow — a real business example

A property agent in Johor used this workflow to produce a client presentation on a new residential development. The task was complex: market context, property highlights, investment rationale, and a Q&A section.

One-shot attempt produced a generic 400-word overview. Using the 4-step workflow — clarify, outline, draft, self-review — the output was a structured 800-word presentation ready for client use, requiring only minor factual updates. Time saved: approximately 90 minutes of writing and editing.

The workflow applies to any complex business output. The steps do not change — only the brief you provide at each stage.

The Business Prompt Playbook  ·  Article 6 of 8

10 Ways to Fix a Bad AI Answer in Under 60 Seconds

By Jane Chew  ·  AI Strategy Coach, DigitalAI Business Club  ·  May 2026

A bad AI answer is not the end of the conversation — it is diagnostic information. It tells you exactly what the prompt was missing. This checklist gives you 10 targeted fixes for the most common prompt problems, each one applicable in under a minute.

The right mindset: diagnosis, not frustration

Most people react to a bad AI answer with frustration — they assume the tool is inconsistent or unreliable. The more useful response is to treat the bad answer as a diagnostic signal. Every weak output has a cause, and every cause has a fix.

This checklist maps the most common AI output problems to the specific prompt element that produced them. Run through it when an answer misses the mark.

The 10 prompt fixes

These 10 fixes cover the vast majority of prompt problems. Most issues resolve with one or two of them. The skill is identifying which one applies — and with practice, this becomes instinctive.

In Article 7, we move into the expert prompt patterns that professionals use for more complex and strategic work — including persona stacking, reverse engineering, and simulation.