Good AI prompts come from having a good structure, just like when creating a proposal or grant writing. A clear prompt turns a vague request into usable output – whether you’re working in Microsoft Copilot, Claude, Gemini, ChatGPT, or Perplexity. This guide covers the frameworks, structures, and techniques that will help you get better responses every time.
Key Takeaways
- A structured prompt produces a more useful AI response than an open-ended question, regardless of which tool you’re using.
- The IMPACTS framework gives advanced users a repeatable method for complex, multi-part requests.
- Multimodal and agentic prompting extend structured prompting from single-step text tasks into multi-format, multi-step business workflows.
Table of Contents
Why Does Prompt Engineering Matter?
Reid Johnston, Teal cofounder and Chief Intelligent Transformation Officer explains, “Artificial intelligence, regardless of platform, is only as good as the prompt you give it. That’s why you hear so much about prompt engineering. It’s an important part of navigating this technology.”
Most people new to AI treat them like a search engine. They input short, vague questions that return equally vague answers. But the moment you add structure, context, and intent, the output changes completely.
What is the Basic Structure for a Good AI Prompt?
A reliable beginner framework works across every major AI tool:
Act as a [Role] and [Task] a [Output] in a [Tone] way.
Each element has a specific job:
- Role: The professional perspective the AI tool should adopt, such as accountant, marketing director, financial advisor, or project manager.
- Task: The action you want taken, such as create, write, summarize, design, research, explain, suggest, or personalize.
- Output: The format of the deliverable, such as a plan, schedule, strategy, or list.
- Tone: How the response should read, such as professional, formal, informal, confident, friendly, optimistic, or apologetic.
Example prompt using this structure.
Act as a content manager and create a content plan for one month focusing on the financial services industry. Plan 6 humorous social media posts, 1 formal press release introducing a new app, 2 enthusiastic blogs, and 2 professional emails.
The great news is that the output generated from this prompt doesn’t have to stop in the chat window, either. Cowork tools (built into Claude and Copilot) can take that same output and push it straight into the project management platform your team already uses.
For example, I could tell Claude Cowork to turn every deliverable from that prompt into its own task in monday.com, our project management tool, without copying and pasting a single line myself.
5 Elements of an Effective AI Prompt
There are five elements that apply to an effective prompt – whether you’re asking an AI tool to draft an email or build a quarterly strategy:
1. Clarity and specificity.
Vague prompts return vague answers.
Instead of asking, “How can we improve sales?” try, “What strategies can we implement to increase our online sales by 20% in the next quarter?” The more specific the question, the more actionable the response will be.
2. Contextual information.
Tell the model your industry, your current challenges, and your specific goals. More context produces more tailored responses. Don’t assume the AI tool knows your situation.
3. Actionable language.
Use verbs that direct the output, such as analyze, evaluate, recommend, optimize, and summarize. These tell the model what type of thinking you want applied to your question.
4. Iterative refinement.
If the first answer isn’t right, adjust the prompt rather than accepting a mediocre result. Be more specific, reframe the question, or focus on a different aspect of the problem.
5. Feedback loops.
When team members use AI tools, have them share what works. The prompts that consistently produce good output are worth documenting and reusing across the organization.
The IMPACTS Framework
For complex requests, such as strategy documents, detailed plans, or multi-audience communications, the IMPACTS framework gives you a systematic way to build a complete prompt. It covers seven dimensions that any AI tool needs to produce professional-grade output.
- Instructions: What do you want the AI tool to produce?
- Motivation: What is the goal behind this request?
- Perspective: Who is the author? What expertise or point of view should the response reflect?
- Audience: Who will read this? What do they already know, and what do they need?
- Context: What background information is relevant? Industry, geography, constraints, prior decisions.
- Tone: How should this read? Persuasive, analytical, empathetic, authoritative?
- Structure: How long should the output be, and what format should it take?
Example IMPACTS prompt.
Create a marketing plan for a small healthcare practice in Minneapolis that serves elderly patients. The plan should be written from the perspective of a seasoned marketing director (20+ years of experience) and target the younger family members of elderly patients. The goal is to educate them on why their participation is important and how the clinic makes it easier for them. The marketing plan should be persuasive yet compassionate, include a mix of strategies (content marketing, digital ads, community outreach), provide messaging recommendations that address affordability and convenience, be structured with clear paragraphs and bullet points, and be no more than 1,500 words with a strong call to action.
What is Multimodal Prompting?
Multimodal prompt engineering combines text with images, documents, audio, or video in a single request instead of typing everything out. This means it can turn a screenshot or a PDF into a starting point instead of something you have to describe or copy-paste into the chat.
How to use this in a business context:
- Paste a screenshot of a broken invoice template and ask the model to rebuild it in a cleaner layout.
- Upload a competitor’s PDF pricing sheet and ask for a side-by-side comparison table.
- Share a photo of a whiteboard from a planning session and ask for a written action plan with owners and deadlines.
Which AI tools support multimodal prompting?
Most major AI tools now accept more than plain text, though the level of support varies by platform:
AI tool | What it can process | Where it fits in a business workflow |
|---|---|---|
Microsoft Copilot | Text, images, and documents inside Word, Excel, and Outlook | Drafting and summarizing directly inside the Microsoft 365 apps |
Anthropic Claude | Text, images, PDFs, and code | Analyzing long documents, contracts, and reports in one pass |
Google Gemini | Text, images, audio, and video | Working across Gmail, Docs, and other Google Workspace apps |
Perplexity | Text and images, grounded in live web search | Research that needs current, cited sources rather than the model's training data alone |
How Does Agentic Prompting Automate Business Workflows?
Agentic prompting instructs an AI tool to complete multiple steps on its own instead of stopping after one response. This ties back to our earlier monday.com example.
Instead of asking a model to draft one email, an agentic prompt asks it to review last week’s open support tickets, draft responses to the three with the longest wait times, and flag anything that mentions a regulated client for a person to review before it goes out.
What to check before using agentic prompts.
Agentic prompting removes the human check at each individual step, so that review has to happen somewhere. Add an explicit instruction that sends anything touching client messaging, compliance, or money to a person before it’s finalized or sent.
What Other AI Prompting Techniques Should Business Leaders Know?
These are worth knowing once your team is comfortable with the basics.
Few-shot prompting
Few-shot prompting means giving the model two or three examples of the output you want before asking for something new. It’s useful when you need consistent formatting across a lot of similar requests, such as weekly reports or client update emails.
Context engineering
Context engineering is managing what the model already knows before you ask it a question, rather than re-explaining your business every time. Uploading your style guide once, or pointing the tool at existing documents, saves the re-typing and produces more consistent answers.
Retrieval-augmented generation
Retrieval-augmented generation, or RAG, grounds an AI tool’s answer in a specific set of documents or current sources instead of relying only on what the model learned during training. Perplexity builds this in by default; other tools support it once you upload files or connect a knowledge base.
How You Can Build AI Skills & Programs in Your Business
Teal hosts monthly virtual events where Reid Johnston demonstrates his favorite AI tips and shares the latest updates.
We’re a managed IT services provider serving SMBs and nonprofits in the DMV and Twin Cities regions. Recognized on the MSP 501 list for eight consecutive years, we provide managed AI, Copilot optimization, AI projects, and more as part of our managed services.
Resource
AI Adoption Guide
Get the framework for choosing, governing, and profiting from AI as a regulated organization.
