Complete FREE Prompt Engineering Guide – 2025

Text prompting — Level 1

Rule 1 — Remove Distractions

Avoid fillers that do not change the task, such as “please”, “can you”, “maybe you could”, “what do you think”. They add length and can introduce confuse the LLM.

Rule 2 — Give Essential Context

Tell the model the facts it needs: purpose, audience, constraints, and any key background. The less guessing it has to make, the better the output.

Rule 3 — Assign a Role

Give the model a role so it filters its research and gives better and clearer responses. Example: “You are an experienced accountant, your task is to help me with my store’s revenue calculations.”

Rule 4 — Limit Guessing

Use clear words: “focus on”, “only”, “do not include”, “avoid”. Narrow the topic to what you actually need.

Text prompting — Level 2 (for large tasks)

Rule 5 — Let the model ask you questions

Before asking for something long or details (for example, writing an essay), tell the model to list any questions it needs answered. Then answer those questions yourself in details and run the prompt again for a much better result.

Example instruction: “Before writing, list up to five clarifying questions about required tone, length, and audience.”

Rule 6 — Specify the output format and show an example

Tell the model the exact format you want: JSON, CSV, table, bullet list, markdown, etc. Provide a short example so the model can match structure and field names.

Example: Output: JSON array of objects like [{"title":"", "summary":"", "time_minutes":0}]

Rule 7 — Break big tasks into subtasks

Split complex tasks into independent steps and prompt the model for each step. For example: research → outline → first draft → editing → final formatting.

Rule 8 — Control sampling and other parameters

Add specific LLM parameters, use them in the end of the prompt, like:

• Temperature, it controls randomness. Typical range 0 to 1.0.
  • Lower values (0 to 0.3) make output more deterministic and factual.
  • Medium (0.4 to 0.7) balances reliability and creativity.
  • Higher (0.8 to 1.0) increases variety; use for creative drafts.
• Top-p, it is a probability cutoff between 0 and 1.
  • Lower top-p (for example 0.6) reduces low-probability tokens and makes output safer.
  • Top-p near 1.0 lets the model use the full distribution. Often used together with temperature.
• Max tokens / length: set a word limit to control output lenght.

Use examples:

• Factual summary: temperature: 0, top_p: 0.9, max_tokens: 300
• Creative brief: temperature: 0.9, top_p: 0.95, max_tokens: 500

You can play with these more in the LLM playground. Here is one i built myself.

Image prompting

Rule 1 — Subject

Name the main subject precisely. Instead of “cat”, say “short-haired black cat with bright blue eyes and a white chin”.

Rule 2 — Description

Describe what the subject is doing and the environment. Include poses, interactions, props, and any relevant background details.

Rule 3 — Style and aesthetic

Specify art style, camera framing, lens, or era. Examples: “cinematic close-up”, “Ghibli style”, “photorealistic studio portrait”, “isometric illustration”.

Rule 4 — Negative constraints

List what to exclude. Use words like “avoid” or “do not include” and be explicit about unwanted elements such as unwanted colors, extra people, logos, or specific objects.

Rule 5 — Resolution and composition

State desired quality and dimensions: portrait or landscape, exact pixel size if needed, aspect ratio, and any composition rules (rule of thirds, central composition).

Extras that improve results

• Lighting and mood: e.g., “soft golden hour light”, “high contrast studio lighting”.
• Color palette: e.g., “muted earth tones” or “vibrant cyan and magenta”.
• Camera details for photorealism: lens focal length, aperture, distance.
• Post-processing or film look: “35mm grain”, “matte finish”, “high dynamic range”.

Examples

Text prompt example (short product summary)

You are a product copywriter for a B2B SaaS aimed at small HR teams.
Task: Write a one-paragraph product summary (max 60 words) that highlights time savings and two key features.
Audience: HR managers with limited technical background.
Tone: Professional, concise.
Constraints: Do not mention pricing or competitor names.
Output: Plain text, single paragraph.

Advanced workflow example

1. Prompt: “List up to five clarifying questions about audience and required length.”
2. Answer questions.
3. Prompt: “Produce the final output in JSON as [ { title, summary, word_count } ]”.

Image prompt example

Subject: young woman with curly black hair, wearing a navy blazer, holding a leather notebook.
Action and setting: standing on a rainy city street at dusk, light reflecting on wet pavement.
Style: photorealistic, cinematic close-up, shallow depth of field.
Lighting: soft backlight and warm street lamps.
Negatives: avoid logos, no other people in frame, no overexposed highlights.
Resolution: landscape, 3840x2160, high quality.

Quick checklist you can copy-paste

• Remove polite fillers.
• Give context: purpose, audience, constraints.
• Assign a role.
• Narrow the scope.
• Ask the model to ask clarifying questions for big tasks.
• Specify the exact output format and give an example.
• Break large tasks into subtasks.
• Set sampling parameters and max length when available.
• For images, list subject, action, style, negatives, and resolution.

OpenAI’s full guide: https://cookbook.openai.com/examples/gpt-5/gpt-5_prompting_guide

Cool Note

I used this exact prompting technique to write this guide you’re reading, but I gave it the data I got from my research, and it just structured it 🙂