How to Get Better Results from AI

Most people get mediocre AI output for a simple reason: vague prompts produce vague answers. When you give an AI model a one-sentence instruction with no context, no constraints, and no stated format, it has to guess at almost everything. The good news is that learning to prompt well is a learnable skill — and the gap between a weak prompt and a strong one is massive. This guide walks you through the five elements of a great prompt, how to iterate effectively, and a few model-specific quirks worth knowing.

Why AI gives generic results (and how to fix it)

AI language models are trained to be helpful to the widest possible audience. Without specifics, they default to the most common, safest, most generic answer possible. The fix is simple: front-load your prompt with context. Tell the model who you are, what you already know, what you're trying to accomplish, and what a good output looks like. The five elements that consistently produce better results are:

1. Role — Give the model a persona or expert identity relevant to your task.
2. Context — Explain your situation, what you've already tried, and any relevant background.
3. Task — State clearly what you want the model to do, using action verbs.
4. Format — Specify how you want the output structured (bullet points, numbered list, prose, JSON, etc.).
5. Constraints — Set limits: length, reading level, tone, what to avoid, or any hard rules.

Example Prompts

WEAK: "Write a blog post about remote work."

STRONG:
You are an experienced content strategist writing for a B2B SaaS blog.
Target audience: HR managers at companies with 50–500 employees.

Write a 700-word blog post titled "5 Signs Your Remote Work Policy Is Quietly Hurting Retention".
Structure: hook paragraph, 5 numbered sections (each ~100 words), closing CTA.
Tone: direct and data-aware, not preachy.
Avoid: generic advice like "communicate more" or "use Slack".
Include: at least two concrete examples or statistics.

You are a senior TypeScript engineer with expertise in React performance.

I have a component that re-renders 3–4x more than expected. Here is the code:
[PASTE CODE HERE]

Please:
1. Identify all the re-render causes in order of impact
2. Suggest specific fixes with code snippets
3. Flag any other issues you notice (not just performance)

Format your response as a numbered list. Keep explanations concise.
Do not suggest rewriting the whole component — I need targeted fixes only.

You are a strategy consultant with experience in SaaS pricing models.

Context: I run a B2B software company with $2M ARR. We currently charge a flat $99/mo.
Competitors charge between $49–$299 depending on seats and features.
Our top 20% of customers use 5x the API calls of everyone else.

Task: Analyze whether we should move to usage-based pricing.
Cover: pros and cons, transition risk, a recommended pricing structure, and what we'd need to measure.
Output: a structured memo with headers. Aim for 500 words. No fluff.

You are a research analyst summarizing academic findings for a non-technical executive audience.

Topic: The effect of sleep deprivation on workplace decision-making.

Summarize the current state of research. I need:
- A 3-sentence executive summary
- Key findings (bullet points, max 6)
- Any meaningful debate or uncertainty in the literature
- Practical implications for managers

Keep it under 400 words. Use plain English — no jargon.
If you don't know something with confidence, say so.

You are a productivity coach who works with founders and senior executives.

My situation: I run a 12-person startup, working ~60h/week. I have trouble saying no to meetings.
My goal: get back 8–10 hours of deep work per week without damaging team relationships.

Give me a specific, actionable plan I can implement starting Monday.
Format: day-by-day changes for week 1, then ongoing habits.
Be direct. Skip the motivational framing — I need tactics, not inspiration.

Iteration strategy

A single prompt is rarely your best prompt. After you get a first response, try one of these iteration moves: (1) Ask for variations — "Give me 3 different versions of this with different tones." (2) Ask the model to critique its own output — "What are the weakest parts of that response? How would you improve it?" (3) Add a constraint you forgot — "Now rewrite it in under 200 words." (4) Shift the persona — "Now respond as a skeptic who would push back on this plan." Each iteration narrows the output toward exactly what you need.

Model-specific tips

ChatGPT (GPT-4o): Responds well to explicit persona instructions and tends to give thorough, well-organized lists. For creative tasks, asking it to "be unconventional" genuinely shifts the output. Watch for over-hedging — adding "be direct, skip the caveats" helps.

Claude (Anthropic): Excellent at nuanced reasoning and following complex multi-part instructions. Does well with long context. If you want concise output, be very explicit — it tends toward thoroughness by default. Claude responds well to prompts that explain the "why" behind a request.

Gemini: Strong at tasks that benefit from Google's knowledge graph, especially current events and factual lookups. Gemini 2.5 Pro handles long, multi-step reasoning tasks well. For creative work, give it explicit stylistic direction since its defaults lean generic.

Related resources