How To Prompt AI Properly In 2025

The Five Prompting Principles That Work

1. Be Specific, Not Clever

Good prompting is about clarity, not creativity. Tell the model exactly what you want — output length, tone of voice, format, audience.

Bad:
“Write me a caption.”
Better:
“Write a caption in a bold, irreverent tone to promote a food pop-up aimed at London creatives. Limit to 280 characters. Must include a call to action.”

Generative models like GPT-4 and Claude 3 respond best when parameters are set clearly. Vagueness equals guesswork (Anthropic Developer Docs).

2. Give the AI a Role

Framing your prompt with a role makes a huge difference in tone and authority. “You are a travel journalist” or “You are a cynical Gen Z TikToker” creates more aligned, believable content.

This technique is called “role prompting” and is a core part of how platforms like Jasper, Copy.ai and Notion AI help marketers generate usable copy faster (Lakera Blog).

3. Use Examples

The fastest way to get the tone and structure right? Show it. “Write me a tagline like Nike” is okay. “Here are three examples of Nike-style copy: Now write one for a premium matcha brand” is better. This is known as “few-shot prompting” — and it helps language models mirror the style you want (Wikipedia: Prompt Engineering).

ChatGPT and Claude are particularly effective at patterning behaviour off examples — even just one or two lines.

4. Add Reasoning When Needed

If you want a more structured or analytical answer — like a media plan, brand audit, or messaging hierarchy — you’ll get better results by prompting the model to think step-by-step.

This is called “chain-of-thought prompting” and has been widely adopted in content, operations and research use cases (Arxiv, Google Brain).

Prompt:
“Explain how you’d audit a luxury hospitality Instagram account, then summarise your top 3 recommendations.”

You’ll likely get a more logical, usable answer than if you’d just asked for tips cold.

5. Specify Format

AI tools don’t assume anything — not even formatting. If you want bullets, ask. If you want a press release structure, say so. If it’s for Instagram, tell it the character limit. This is especially useful when working across tools like Surfer SEO, Canva Magic Studio, or Notion AI, which now offer generative content natively but rely on clean prompts to avoid filler.

As Harvard Professional & Executive Development put it: “The more structure and constraints you give your AI assistant, the more efficiently it will return high-quality output” (Harvard DCE).

What Tools Need Better Prompting?

You’re probably already using some of these. Each one works better when prompted well:

• ChatGPT (OpenAI) – Best for copywriting, strategy, and campaign drafting. Start with a role, add clear instructions, and define your output format.

• Claude (Anthropic) – Stronger on structured, step-by-step responses. Ideal for reports, rewrites, tone matching, or editorial strategy.

• Midjourney / DALL·E – Image-generation tools. Prompting needs to include subject, composition, style references, even lighting. “90s snack packaging, flat lay, harsh flash, white backdrop” yields better results than “make something fun”.

• Jasper.ai – Templates help with structure, but strong prompts still perform better than defaults. Works well with tone, length and format constraints.

• Canva Magic Studio – Good for social captions, headlines and blog intros. Specify tone, length, and brand voice directly in the prompt field.

• Gupshup / Smartly.io – Used for ad automation and chatbot flows. Prompting is part of how you train response logic — from WhatsApp campaigns to dynamic ad creatives.

More brands are now building internal prompt libraries — documented, repeatable prompts by task — to keep tone consistent and save time across teams (Futurepedia).

What Not to Do

• Don’t assume the AI knows your business. If it’s not public info, spell it out.

• Don’t ask for perfection first time. Prompting is iterative.

• Don’t ignore hallucinations. If it makes something up, flag it, rerun it, and test another phrasing.

• Don’t over-rely. AI is powerful, but it’s not brand-safe out of the box. Always review and refine.

According to The Verge, even the best models still generate false data and speculative copy if not constrained by good prompts.