Writing better prompts for AI tools is the single skill that most improves the quality of output you get — more than switching to a more expensive model, more than trying a different tool entirely, more than any other change you could make. I learned this gradually and somewhat painfully: months of mediocre AI output that I blamed on the tools, before I understood that the tools were producing exactly what my prompts deserved, which was generic responses to generic instructions. When I started writing specific, structured, constrained prompts, the output quality improved dramatically on the same tools with the same models. You’ll find the complete rundown in our Complete Guide to AI Tools.
This guide covers how to write better prompts for AI tools in 2026 — the techniques that consistently produce better results across different tools and different use cases, and the common mistakes that produce the kind of mediocre output that makes people conclude AI tools aren’t useful. The honest truth is that most people who conclude AI tools are disappointing haven’t yet invested in their prompting — and prompting is a learnable skill, not a natural talent.
A useful framing before getting into the techniques: think of prompting as briefing a capable but literal collaborator who will do exactly what you ask and nothing more. If you give vague instructions, they produce a vague result. If you give specific, structured instructions with clear success criteria, they produce something much closer to what you actually want. The collaborator has excellent general capability — they know a great deal, write well, and follow instructions carefully. What they don’t have is your context, your judgment about what matters, or any ability to read your mind about what “good” looks like for your specific use case. Every prompting technique below is about giving that collaborator more of what they need.
Technique 1: Specify the role and context first
The single fastest improvement to most prompts is adding a role and context at the beginning. Without this, the AI produces a generic response calibrated to an imaginary average user. With it, the response is calibrated to your specific situation and audience.
Role specification: “You are an experienced marketing copywriter specialising in B2B SaaS” produces different output from “You are a journalist writing for a general audience” — even on the same topic. The role sets the knowledge base, the vocabulary, the assumptions about the reader, and the appropriate style. It works because the AI adjusts everything that follows based on the role you’ve specified.
Context specification: “I am writing for compliance officers at mid-size banks who are evaluating AI governance policies for the first time” produces more targeted output than “I am writing for a business audience.” Context tells the AI who the eventual reader is and what they specifically care about.
Combined example that works: “You are an experienced HR consultant. I am writing an internal policy document for a 50-person technology company establishing an AI use policy for the first time. The audience is employees who are not technical specialists. Write in a clear, approachable tone that does not assume legal or technical background.”
Compare that to the alternative — “Write an AI use policy” — and the quality difference in output is immediately obvious.
Technique 2: Be specific about format and length
AI tools default to a format and length based on patterns in their training data when you don’t specify one. These defaults are often wrong for your specific need. The default for a blog post prompt might be 600 words when you need 1,500. The default might be headers when you want flowing prose. The default might be bullet points when you want a narrative.
Specifying format and length takes ten seconds and prevents the most common output mismatch. Specific format elements worth including explicitly:
- Length: word count, not vague terms like “short” or “detailed.” “Around 200 words” is unambiguous. “Brief” is not.
- Structure: whether to use headers, whether to use bullet points or prose, how many sections, whether to include a summary or introduction.
- Tone: specific descriptors rather than generic ones. “Conversational but professional, like a knowledgeable friend explaining something they know well” is more useful than “professional.”
- Opening and closing constraints: if you don’t want the response to start with a rhetorical question, a definition, or a statement about how important the topic is, say so explicitly. These are the most common AI defaults and they’re often not what you want.
Technique 3: Provide an example of what good looks like
Providing an example of what good output looks like is the most powerful single technique for improving prompt results — and the most consistently underused one. Words like “professional,” “engaging,” “clear,” and “warm” mean different things to different people, and to AI tools these words trigger a probability distribution of outputs based on how they appear in training data, not based on your specific interpretation of them.
An example anchors the output to something specific. “Write a product description in this style: [paste example]” produces a much more precisely calibrated result than “write a product description in a warm, professional tone.” The example doesn’t have to be your own writing — it can be any published example that captures the style you want.
For recurring tasks, building a prompt template that includes an example is worth the upfront investment. I have templates for the types of writing I do regularly that include a short example paragraph. The output consistently requires less editing than prompts without examples — the difference is significant enough that I won’t start a recurring content task without one.
Technique 4: Tell the AI what not to do
Negative constraints — specifying what not to do — are as useful as positive specifications and often more efficiently targeted at the specific failure modes you want to prevent. AI tools have characteristic patterns that appear repeatedly in their output: certain opening phrases, certain transitional expressions, certain structural habits. If you know what you don’t want, specifying it directly prevents those patterns from appearing.
Common negative constraints worth including for writing prompts:
- “Do not start with a rhetorical question” — a very common AI opening that rarely serves the content well
- “Do not use the phrases ‘In today’s X world’ or ‘It’s no secret that’” — generic transitions that signal AI-generated content to careful readers
- “Do not use bullet points” — if you want prose, specifying this prevents the AI from defaulting to lists
- “Do not include a formal conclusion summarising what was already said” — a common AI habit that adds length without value
- “Do not use jargon” — if the audience is non-specialist, this prevents domain-specific vocabulary that would confuse them
- “Do not hedge every statement with ‘it’s worth noting that’” — AI tools over-hedge, and removing the hedge instruction reduces the tentative quality of the output
Technique 5: Chain of thought for complex reasoning tasks
For complex tasks that require reasoning — analysing a problem, evaluating options, making a recommendation — asking the AI to show its reasoning produces better final answers. The technique is simple: add “Think through this step by step before giving your answer” or “Explain your reasoning as you go.”
This works because it forces the AI to construct an explicit reasoning chain rather than jumping to a pattern-matched conclusion, which reduces errors in the reasoning and makes the output more transparent and easier to verify. It’s particularly valuable for analytical tasks where you want to understand the reasoning, not just the conclusion — evaluating options, identifying risks, diagnosing problems, or making recommendations where the process matters as much as the answer.
For tasks involving calculations or formal logic, chain of thought is especially important. AI tools make arithmetic errors and logical errors in the middle of otherwise fluent output. Making the reasoning explicit makes these errors visible, which is the first step to catching them.
Technique 6: Use multi-turn iteration rather than single prompts
One of the most common prompting mistakes is abandoning a conversation when the first response isn’t what you wanted and starting a completely new prompt from scratch. Iterating within the same conversation — using the AI’s response as a starting point and giving specific feedback — often produces better results than any single prompt can achieve.
The iteration prompts that work best:
- “This is good but [specific aspect] is not right. Can you revise it to [specific improvement]?” — far more effective than “try again”
- “The tone is too [formal/casual/technical]. Can you adjust it to [specific target]?”
- “This is missing [specific element]. Can you add it in a way that fits naturally with the existing content?”
- “I like [specific part]. Can you make the rest more like that section?”
- “The opening is too abstract — can you start with a concrete example instead?”
Specific feedback produces specific improvements. “This isn’t quite right” gives the AI no useful information to work with. “The second paragraph is too long and loses the main point — can you cut it to two sentences?” gives it exactly what it needs to make a targeted improvement. The more specific your feedback, the more useful the revision.
Technique 7: Build and maintain a prompt library
The prompts that produce good results for recurring tasks are worth saving and reusing. A prompt that took 15 minutes to develop and produces reliably good output on a type of writing you do every week is a durable asset. A folder of tested, working prompts — organised by task type — is one of the most underrated productivity investments available to regular AI tool users.
What to include in a prompt library entry:
- The full prompt including role, context, format specifications, negative constraints, and any example
- The specific task type it’s designed for
- Notes on what works and what to adjust for variations of the task
- A note on which AI tool(s) it works best with — prompts often perform differently across tools
For teams, a shared prompt library creates a form of institutional knowledge about how to get good output from AI tools for the team’s specific use cases — reducing the individual learning curve for new team members and preventing the same prompting mistakes from being made repeatedly.
A complete example: weak prompt vs strong prompt
Pulling all these techniques together — here’s the difference between a weak and a strong prompt for the same task.
Weak prompt: “Write a LinkedIn post about working from home productivity.”
Strong prompt: “You are writing a LinkedIn post for a productivity consultant who works with remote teams. The audience is mid-level managers who are struggling with their team’s focus and output while working from home. Write 150 words in a direct, conversational tone — the style of someone sharing a hard-won insight rather than giving generic advice. Do not start with a rhetorical question. Do not use bullet points. End with a specific observation rather than a call to action. Here is an example of the tone I want: [paste a short example paragraph].”
The second prompt takes 45 seconds more to write. The output quality difference is significant — not because the AI is more capable, but because it has been given enough to work with. This is the core insight behind prompting: AI tools are not limited by their capability on most everyday tasks. They’re limited by the specificity of what they’re asked to do.
Prompting technique reference
| Technique | What it does | Example |
| Role and context | Calibrates expertise level and audience assumptions | “You are a compliance officer writing for non-technical employees” |
| Format specification | Prevents default format mismatches | “200 words, no bullet points, no headers, conversational tone” |
| Examples | Anchors output to a specific style more precisely than adjectives can | “Write in this style: [paste example paragraph]” |
| Negative constraints | Prevents characteristic AI patterns and generic defaults | “Do not start with a rhetorical question or use ‘In today’s world’” |
| Chain of thought | Improves reasoning quality and makes logic verifiable | “Think through this step by step before answering” |
| Specific iteration | Targets improvements precisely rather than restarting | “The opening is too abstract — start with a concrete example” |
| Prompt library | Preserves working prompts for recurring tasks | Saved prompt templates by task type with notes on variations |
Our guide on using AI tools for writing applies these prompting techniques specifically to written content creation tasks. For the broader workflow of using AI tools for productivity — including prompt templates for recurring professional tasks — our guide on AI tools for productivity covers how to build prompting habits that compound into real time savings over weeks and months. If this sounds familiar, How to Use AI Tools for Email Marketing is worth a look.
