How to Write Better Prompts for Any AI Chatbot
Learn practical prompting techniques that get better results from any AI chatbot , ChatGPT, Claude, Jasper, Writesonic, and more. Covers structure, context, role-setting, and common mistakes to avoid.
Disclosure: This article contains affiliate links to Jasper and Writesonic. If you sign up through a link on this page, we may earn a commission at no extra cost to you.
Most people’s first instinct when using an AI chatbot is to type a question the way they’d type it into Google, short, keyword-heavy, a bit vague. That works fine for a search engine, but it’s not how chatbots think.
AI chatbots are designed to complete tasks, not just retrieve facts. The more clearly you describe what you want, who it’s for, and what format you need it in, the more useful the output. A weak prompt gets you a generic response. A well-structured prompt gets you something you can actually use.
This guide covers the techniques that make a real difference, and they apply whether you’re using ChatGPT, Claude, or a writing-specific tool like Jasper or Writesonic.
Prerequisites
- An account with at least one AI chatbot (ChatGPT, Claude, Jasper, Writesonic, or similar)
- A specific task in mind, prompting technique is easiest to learn when you’re solving a real problem, not practicing in the abstract
No technical background is needed. These techniques work in plain conversational English.
Step 1: Assign a Role
The single fastest improvement most people make is giving the AI a role before stating the task. This isn’t magic, it works because role context narrows the model’s frame of reference and pushes the output toward a particular style, vocabulary, and level of expertise.
Weak prompt:
Write a subject line for a marketing email.
Better prompt:
You are an experienced email copywriter who specializes in high-converting subject lines for B2B SaaS. Write five subject lines for a re-engagement email targeting trial users who haven’t logged in for 14 days.
The second version specifies domain (email copywriting), industry (B2B SaaS), task type (re-engagement), audience (inactive trial users), and deliverable count (five). The AI doesn’t have to guess at any of those, so it doesn’t.
When using a tool like Jasper’s Brand Voice feature, role-setting happens partly at the account level. But even then, adding a quick role framing in your prompt helps direct the output for specific tasks.
Step 2: Give It Context, Not Just a Topic
A topic tells the AI what you’re writing about. Context tells it why, for whom, and under what constraints. Context is the part most prompts are missing.
Think of it like briefing a colleague who just joined the project. You wouldn’t just say “write the email”, you’d explain the situation, the goal, and any constraints they need to know about.
Useful context to include:
| Context element | Example |
|---|---|
| Purpose / goal | ”This is for a newsletter going out Friday” |
| Audience | ”Readers are freelance designers, not developers” |
| Tone | ”Conversational, not corporate, like a smart friend, not a press release” |
| Constraints | ”Keep it under 200 words” |
| What to avoid | ”Don’t mention the pricing change, that’s next week” |
| Existing material | Paste a product description, brief, or previous draft directly into the prompt |
You don’t need all of these every time. One or two well-chosen context cues often make a bigger difference than a longer, unfocused prompt.
Step 3: Specify the Output Format
If you don’t say what format you want, the AI picks one, and it often picks wrong. Ask for a bullet list and get a paragraph. Ask for a paragraph and get a listicle. The fix is simple: say what you want.
Examples:
- “Give me this as a numbered list, not paragraph form.”
- “Format the output as a table with columns: Tool, Price, Best For.”
- “Write this as three separate tweets, each under 280 characters.”
- “Use plain text, no markdown headers or bold formatting.”
If you’re working in a tool like Writesonic’s Article Writer, the UI handles a lot of this structurally (you pick headings, sections, and outline before generating). But in free-form chat interfaces, format instruction is entirely on you. Be explicit.
Step 4: Use Examples When the Format Matters
If the format or style you’re going for is unusual or very specific, showing an example is faster than describing it. This is called few-shot prompting in technical circles, but the underlying idea is intuitive: you’re showing the AI what “right” looks like before asking it to produce something.
Example:
Write three product description bullets in the same style as these:
- “Handles 100 tabs without breaking a sweat, because you work that way.”
- “Zero-lag transitions. Seamless switching between every workspace you’ve got.”
Now write three for a wireless keyboard aimed at remote workers.
You don’t need to label this technique or explain what you’re doing. Just paste your examples and make the request. The AI will pick up the pattern.
This works especially well for tone matching, if you want output that sounds like your existing content, paste a paragraph of your actual writing as the example rather than trying to describe your voice in the abstract.
Step 5: Break Complex Tasks into Steps
When a task has multiple moving parts, asking for everything in one prompt often produces output that’s shallow across the board. A better approach is to chain prompts, handle one part first, review it, then move on.
Instead of:
Research the topic of content repurposing for freelancers, create an outline, and write a 1,500-word article with an intro, five sections, and a conclusion.
Try:
Step 1: Give me five possible angles for an article about content repurposing aimed at freelance writers. Just the angles, one sentence each, no full outline yet.
Once you pick an angle:
Good. Let’s use angle 3. Now create an outline with five H2 sections and two or three bullet points under each.
Then:
Now write the intro paragraph for this article. Aim for three to four sentences, conversational tone, and lead with a specific problem the reader is likely facing.
This staged approach gives you checkpoints where you can course-correct before a lot of text has been generated in the wrong direction. When using Jasper’s Canvas editor, this maps well to the section-by-section workflow the tool is built around, you write or prompt one section at a time, review, then continue.
Step 6: Tell It What Not to Do (With a Positive Reframe)
Negative instructions, “don’t use jargon,” “don’t make it too long”, are less reliable than positive ones. The AI has to interpret what “too long” means and what counts as jargon. Positive framing removes the ambiguity.
Less reliable:
Don’t make it sound corporate. Don’t use buzzwords. Not too long.
More reliable:
Write in a direct, informal tone, like you’re explaining this to a friend over coffee. Aim for 150 words. Use plain everyday language; replace any marketing or industry jargon with simpler alternatives.
The same goes for format: instead of “don’t use bullet points,” say “write this as a single flowing paragraph.”
Step 7: Iterate Instead of Restarting
One of the most useful things about chatbots, especially in tools like ChatGPT and Claude, is that the conversation retains context. You don’t have to start over every time the output isn’t quite right.
When an output misses the mark, diagnose what’s wrong and give a targeted correction:
- “The tone is too formal. Rewrite the second paragraph to sound more casual.”
- “The hook is generic. Try three different openers, one question, one statistic, one bold statement.”
- “This is 400 words. Cut it to 200 without losing the three main points.”
- “Add a specific example in the third section, something concrete a freelance writer could relate to.”
Targeted follow-ups are faster than re-prompting from scratch, and they let you refine iteratively rather than gambling on one big prompt getting everything right on the first pass.
If you’re using Writesonic’s Article Writer and the output drifts off-topic in the middle sections (a known weakness of long-form AI outputs), the correction approach works the same way, select the weak section and reprompt it specifically rather than regenerating the entire article.
Common Mistakes to Avoid
| Mistake | Why it hurts | Fix |
|---|---|---|
| Vague topic with no context | AI fills in the blanks with generic assumptions | Add purpose, audience, and constraints |
| No format instruction | Output format is a coin flip | State format explicitly in the prompt |
| Treating it like a search engine | AI tries to respond conversationally to a keyword | Write in full sentences with a clear task |
| Accepting the first output without iteration | First draft is rarely best draft | Give targeted feedback and refine |
| One giant prompt for a complex task | Output is shallow across all parts | Chain prompts; handle one part at a time |
| Pasting raw context with no instruction | AI may summarize instead of acting on it | Lead with the task, then paste the context |
Expected Outcomes
When you apply these techniques consistently, you should notice:
- First drafts that require significantly less editing
- Output that matches your tone and format without repeated corrections
- Fewer generic, one-size-fits-all responses
- Faster iteration when something isn’t working
The difference isn’t dramatic on simple tasks, if you ask a chatbot to translate a sentence, prompt structure barely matters. The gap widens on complex or stylistically specific work: long-form content, copy that needs to match a brand voice, multi-step research tasks, or any output where format and tone are part of the deliverable.
Prompting Across Different Tools
These techniques apply across chatbots, but the interface shapes how you apply them:
ChatGPT / Claude (free-form chat): You control everything through the prompt. Role-setting, context, format, all of it needs to be in your text. The iterative conversation format is your friend.
Jasper: Handles some structure at the tool level (Brand Voice, document templates), but prompt quality still determines output quality within those structures. Particularly useful: the role + context approach for generating on-brand copy, and the step-by-step approach within Canvas. If you want to see Jasper in a full content workflow, the guide to building a weekly content calendar with Jasper and Writesonic shows how this plays out in practice.
Writesonic: The Article Writer’s structured brief input handles some context elements (keyword, audience, tone) through the UI rather than freeform text. For other templates and free-form chat, apply the same techniques as above. For a full walkthrough of the Article Writer specifically, see how to use Writesonic’s Article Writer to draft and fact-check a listicle in one sitting.
Try Jasper AI Free Try Writesonic FreeA Quick Prompt Template to Start With
If you want a starting structure you can adapt for almost any content task:
You are a [role] with experience in [domain].
Task: [What you want produced]
Audience: [Who it's for]
Tone: [How it should sound]
Format: [What the output should look like]
Constraints: [Word count, things to include or avoid]
[Optional: paste any relevant context, examples, or existing material here]You won’t use all six fields every time. But running through them mentally takes about ten seconds and usually surfaces the context that was missing from your first instinct.