Good prompting is a skill anyone can learn. These AI prompting tips from the Mississippi Artificial Intelligence Network (MAIN) show how to turn a vague request into a clear instruction that gets useful results from tools like ChatGPT, Claude, Gemini, Copilot, and Grok.
Updated July 2026 with official guidance for GPT-5.6, Claude Fable 5, Gemini 3.5 and 3.1, Microsoft 365 Copilot, and Grok 4.5.
A strong prompt states the outcome you want, supplies the context and sources that matter, defines important constraints or boundaries, and explains what a successful answer should include. Current frontier models usually need less step-by-step scripting than older models, but they still benefit from clear goals and a specific deliverable.
Build real AI skills, free. This guide is a free resource from the Mississippi Artificial Intelligence Network (MAIN). MAIN’s self-paced AI courses are free and open, and Mississippi educators can earn CEUs.
What makes a good AI prompt?
A good AI prompt tells the tool what outcome you need, why you need it, what information it should use, what limits it must respect, and how you will judge the result. You do not need technical language or an elaborate formula. You need a clear task and enough relevant context.
A current prompt formula
Outcome: [what you need]. Context and sources: [audience, background, files, links, or text to use]. Constraints and boundaries: [requirements, limits, and actions that need approval]. Success looks like: [facts, evidence, or qualities the answer must include]. Return: [format, length, and tone]. Ask a question only if missing information would materially change the result.
For example: “Draft a friendly appointment reminder for current customers. Use the date, location, and cancellation policy in the notes below. Keep it under 150 words and do not add details that are not in the notes. Success means the customer can quickly see when and where the appointment occurs and how to reschedule. Return one subject line and one email draft. Ask one question only if a missing fact would make the reminder inaccurate.”
Seven AI prompting tips
These AI prompting tips work across almost every AI assistant. You can apply them whether you use ChatGPT, Claude, Gemini, Copilot, or Grok.
- Lead with the outcome. State what you want completed, who it is for, and why it matters. Current reasoning models can often determine the steps themselves, so focus first on the destination.
- Give relevant context and sources. Provide the facts, files, links, examples, audience details, and prior decisions the model should use. Tell it which source is authoritative when sources disagree.
- Define constraints, boundaries, and success criteria. Say what must be included, what must not happen, which actions require your approval, and how you will judge a good result.
- Use clear structure for complex prompts. Separate instructions, context, source material, and the final task with short headings, bullets, quotation marks, or tags. Put the question after a long document or dataset.
- Show examples when consistency matters. A short example can communicate the tone, format, or level of detail you want. Use a few varied examples for repeatable tasks, but remove old examples that no longer help.
- Choose the right mode and tools. Use deeper reasoning or research modes for complex work and faster modes for routine tasks. For current or obscure facts, ask the tool to search or use grounded sources and provide citations.
- Iterate, verify, and keep human judgment. Correct what missed, ask for a revised draft, and check facts, citations, calculations, names, dates, tone, bias, accessibility, and real-world fit before using the result.
What changed with the latest frontier models?
The fundamentals did not disappear, but official provider guidance now places more emphasis on concise instructions, model-selected reasoning, source grounding, and clear boundaries for tools or actions. The guidance below was reviewed in July 2026.
OpenAI: GPT-5.6
OpenAI recommends leaner prompts: state each instruction once, focus on the outcome, and provide domain context, hard constraints, approval boundaries, success criteria, and the required format. Tell the model when an important ambiguity should trigger a question. For API workflows, choose reasoning effort based on task difficulty instead of automatically using the highest setting.
Anthropic: Claude Fable 5
Anthropic says brief, direct instructions can steer Fable 5 effectively. Explain the reason behind the request, define what the model may act on, and require progress claims to be supported by actual evidence in long tasks. Ask for assumptions, evidence, or a concise explanation—not private chain-of-thought.
Google: Gemini 3.5 and 3.1
Google recommends precise, direct prompts and consistent structure with headings or tags. With long source material, put the context first and the specific question at the end. Gemini 3 models can be concise by default, so request the depth and conversational style you want. Use Search grounding for recent or obscure facts.
Microsoft 365 Copilot
Microsoft’s prompt ingredients remain goal, context, source, and expectations. Copilot can route a request to a faster or deeper model, and GPT-5.6 is rolling out as a preferred model in Microsoft 365 Copilot. Name the files, emails, meetings, or other work sources it should use and describe the finished outcome.
xAI: Grok 4.5
xAI describes Grok 4.5 as a model for coding, agentic tasks, and knowledge work. Its current user documentation emphasizes natural back-and-forth, file uploads, and connected sources. Use the same durable pattern: a clear outcome, relevant source material, constraints, and a defined deliverable.
Skip the prompt theater
Phrases such as “think step by step,” assigning an elaborate expert persona, or repeating the same rule many times are not automatic upgrades. Ask for the evidence, assumptions, checks, and final format you actually need. Add a role only when it supplies useful perspective or standards for the task.
Weak prompt vs. stronger prompt
The easiest way to improve a prompt is to add the missing goal, audience, context, and format. Here are a few quick examples.
Weak: Write an email.
Stronger: Write a friendly 120-word email to parents reminding them that the field trip permission form is due Friday. Include what to bring, where to return the form, and a polite closing.
Research
Weak: Tell me about AI in healthcare.
Stronger: Give me a plain-language overview of three current uses of AI in healthcare administration, not diagnosis. Search for recent information, cite authoritative sources with dates, and distinguish verified facts from examples or predictions. Include benefits, risks, and questions a clinic leader should ask before adopting a tool.
Planning
Weak: Make a training plan.
Stronger: Create a four-week beginner AI training plan for employees at a small manufacturing company. Use the attached acceptable-use policy as a constraint. Include weekly goals, practice activities, and a short safety reminder for each week. Success means each activity can be completed in 30 minutes or less with tools already approved in the policy.
Prompt templates you can copy
Use these as starting points. Replace the bracketed text with your situation, then revise the output before sharing it.
1. Draft something
Draft a [type of document] for [audience]. The outcome is [goal]. Use only the facts in [source or pasted notes]. Include [required points], avoid [prohibited content], and keep it to [length]. Success means [criteria]. Return [format].
2. Summarize information
Summarize the text below for [audience]. Focus on [main concern]. Return the summary as [format]. Include key decisions, open questions, and action items. Text: [paste text]
3. Improve a draft
Review this draft for clarity, accuracy, tone, and missing information. Do not rewrite it yet. First, give me the top five improvements I should make and explain why. Draft: [paste draft]
4. Compare options
Compare [option A], [option B], and [option C] for [situation]. Use [named sources or attached files] and state the date of any time-sensitive facts. Use a table with benefits, risks, cost or effort, and questions to ask before deciding. End with a recommendation tied to [decision criteria].
5. Learn a concept
Teach me [topic] as if I am a beginner. Use simple language, one example, one analogy, and five practice questions. Tell me what beginners usually misunderstand.
6. Complete a multi-step task
The outcome is [finished deliverable]. Use [files, tools, or sources]. You may [approved actions]. Ask before [external, destructive, costly, or irreversible actions]. Check your work against [requirements or rubric] before finishing. Return [deliverable] plus a short list of what you verified and any remaining uncertainty.
How to fix an answer that misses
You do not have to accept the first response. Treat AI prompting like a conversation. A short follow-up prompt often works better than rewriting everything.
- If it is too vague: “Make this more specific and include three concrete examples.”
- If it is too long: “Cut this by half and keep only the most important points.”
- If the tone is wrong: “Rewrite this in a warmer, more professional tone without sounding casual.”
- If it missed the audience: “Revise this for [audience] at a [beginner/intermediate/advanced] level.”
- If it may be inaccurate: “List the claims I should verify before using this, and explain what kind of source would confirm each one.”
Protect sensitive information
Never paste confidential, student, employee, customer, or patient-identifying information into a public AI tool. Instead, remove names and personal details, or use an approved, secured tool for sensitive work. When in doubt, leave it out.
Before you paste, check for
- Names, addresses, phone numbers, emails, student IDs, employee IDs, account numbers, or medical details
- Private business information, contracts, internal financial data, unreleased plans, or passwords
- Copyrighted materials, confidential documents, or data your organization has not approved for AI use
Prompting guides from the major AI providers
Want to go deeper? These official provider resources cover current models and durable prompting practices. Some model-specific pages are written for developers, but their advice about clarity, context, constraints, sources, and output requirements also applies to everyday use.
Current GPT-5.6 model guidance, including leaner prompts, reasoning effort, autonomy boundaries, response length, and success criteria.
Model-specific guidance for Fable 5, including instruction following, long tasks, evidence-backed progress, boundaries, and reasoning requests.
Current Gemini prompt design strategies for direct instructions, examples, structured prompts, long context, tools, and Search grounding.
Microsoft’s current four-part framework: tell Copilot the goal, context, source, and expectations, then refine the response as needed.
The current Grok user overview covers natural conversation, file uploads, connectors, and Grok 4.5 capabilities.
Practice with MAIN’s prompting guides
Ready to practice? MAIN offers free, role-specific prompting guides with copy-and-paste examples. In addition, our free courses build the foundational skills behind these AI prompting tips.
Small business prompts
Healthcare prompts
K-12 prompts
Higher education prompts
Manufacturing prompts
MAIN free AI courses
Common questions about AI prompting
What is the key principle when writing a basic AI prompt?
The key principle is to use clear, specific instructions. Tell the AI what you want, who the response is for, what context matters, and what format you need.
Do longer prompts always work better?
No. Longer prompts are not automatically better. Current frontier models often perform better with leaner instructions that state the outcome, relevant context, constraints, success criteria, and format without repetition. Add detail when the task is complex, sensitive, or audience-specific.
Should I ask an AI to “think step by step”?
Usually, no. Reasoning models decide how much internal reasoning a task needs, and some providers discourage requests to reveal private chain-of-thought. Ask instead for a concise explanation, key assumptions, supporting evidence, calculations, checks performed, or the reasons behind the recommendation.
How do I get current information from an AI tool?
Ask the tool to search or use a grounded research feature, cite the sources it used, and include publication or event dates. Name preferred authoritative sources when possible. If the tool cannot access current sources, treat its answer as background information and verify it separately.
Should I trust the first AI answer?
No. Treat the first response as a draft. Review it, ask follow-up questions, and verify important facts before using it for school, work, public communication, policy, finance, legal, health, or safety-related decisions.
Have a question about prompting or AI adoption?