I've gone through a lot of painful trial and error with AI prompting—a lot. Which was fine when I was experimenting in back-and-forth conversations with AI chatbots, because I could refine my prompts with every response. But it's a different story with AI agents. A weak AI prompt baked into an agent's instructions produces the same bad output—and bills you for the same mistake—every single time it runs, with no one at the keyboard to catch it.Â
I've rounded up 16 AI prompt templates that the Zapier team and I rely on to get usable outputs on the first try. You can copy and paste them directly into your chat window or your AI agent's instructions, or customize them to suit your needs.
Table of contents:
AI prompt templates at a glance
Prompt template | Use it when |
|---|---|
The AI needs background it can't guess (e.g., audience, constraints, and goals) | |
The same preferences should apply automatically to every task | |
The output must follow a fixed format without inventing details | |
A few examples show what you want faster than a description | |
The output feeds a spreadsheet, CRM, or another workflow step | |
You want the AI to interview you to surface missed information, questions, or opportunities | |
You need a specific professional lens, not a generic assistant | |
You want several expert viewpoints debating a decision | |
You're comparing options on the same weighted criteria | |
The AI should grade its own draft against a rubric first | |
The output was close, and your feedback should drive the revision | |
You need a complete plan, including requirements, steps, and troubleshooting | |
You want an AI first pass on claims before a human fact-check | |
You're writing instructions for an AI agent that runs unattended | |
The AI has real tool and data access and needs hard limits | |
The AI should pause for your sign-off before the high-stakes step |
16 AI prompt templates to get you better outputs
1. Context-setting prompt template
When to use this AI prompt template: Use it for any recurring work task where the background stays stable, like status updates, briefs, and customer replies. (It's also the backbone of effective agent instructions, because an agent can't stop mid-run to ask what you meant.)
AI prompt template:
Before I make my request, here's the context you need:
Background: [the situation and why this matters]
Audience: [who will read or use the output]
Constraints: [tone, length, format, things to avoid]
Success looks like: [what a great output accomplishes]
With that context, here's my request: [your request]
Why it works: An AI model pulls from at least two sources when it answers: its training data and whatever's loaded in its context window. AI agents can fill that window themselves by retrieving files or skills, but retrieval only works for information that's stored somewhere. Details like who the output is for and what success looks like usually live in your head, and when you leave them out, the model falls back on its training data and answers the average version of your request. Front-loading that context gets you an answer to your version. (This is the same principle behind context engineering, applied at the single-prompt level.)
2. Standing instructions prompt template
When to use this AI prompt template: Use it in your chatbot's custom instructions, a project's system prompt, or at the top of an AI agent's instructions—anywhere the same rules should hold on every run.
AI prompt template:
From now on, apply these standing instructions to everything I ask:
About me: [role, team, and what you work on]
Defaults: [tone, format, and length preferences]
Always: [non-negotiables, e.g., "cite a source for every statistic"]
Never: [e.g., "use jargon without defining it"]
If a specific request conflicts with these instructions, follow the request and flag the conflict.
Why it works: Most inconsistency between AI outputs comes from re-explaining your preferences slightly differently each time. Standing instructions move those preferences somewhere the model sees on every run, which is why chatbots like ChatGPT offer custom instructions and file-based agents like Claude Code check your project for an AGENTS.md before they start working. Claude Skills run on the same idea: you package the instructions once, and Claude loads them whenever a task calls for them. If you're setting up an AI agent, building out those instructions is the first thing to do, and this template is the shape to build them in.
3. Placeholder prompt template
When to use this AI prompt template: Use it for business communications, recurring reports, and any AI agent whose output needs to follow the same format on every run.
AI prompt template:
I am going to provide a template for your output. CAPITALIZED WORDS are my placeholders for content. Fit your response into these placeholders and preserve the formatting:
[Your template with PLACEHOLDERS]
Now apply this template to: [your specific request]
Some examples of what a template with placeholders may look like:
Email: "Dear [NAME], your [PROJECT] is [STATUS] as of [DATE]..."
Report: "Executive Summary: [SUMMARY] Key Findings: [FINDINGS]..."
Why it works: When AI doesn't know a specific detail, like a phone number or an address, it fills the blank with something plausible instead of leaving it empty. Placeholders give the model a correct way to handle details it doesn't know: it writes [PHONE_NUMBER] instead of inventing one. You drop in the real details afterward, and anything still in caps is easy for a human to spot and fact-check.
4. Few-shot prompt template
When to use this AI prompt template: Use it to show the AI what you want instead of describing it. ("Shot" is AI-speak for an example included in your prompt.)Â
AI prompt template:
I'm going to show you examples of the kind of output I want. Study the pattern in these examples, then apply the same pattern to a new input.
Example 1 input: [input] / Example 1 output: [output]
Example 2 input: [input] / Example 2 output: [output]
Now apply the same pattern to: [your new input]
Why it works: Large language models (LLMs) are pattern-completion machines at their core, and examples are the densest way to communicate a pattern. In some cases, two or three good examples can convey what you want better than a long-winded paragraph of adjectives.Â
5. Output schema prompt template
When to use this AI prompt template: If the output is going anywhere other than your eyeballs, lock down its structure. Use this template any time the output feeds something downstream: CRM fields, content briefs, or the handoff between one agent step and the next.Â
AI prompt template:
Return your answer in exactly this structure:
[Field 1]: [what belongs here]
[Field 2]: [what belongs here]
[Field 3]: [what belongs here]
If a field doesn't apply, write "N/A" rather than omitting it. Do not add fields, commentary, or preamble outside this structure.
Why it works: Models default to conversational prose, which is pleasant to read and terrible to process. A fixed schema makes the output predictable, and predictable output is what lets a spreadsheet or the next step of an automated workflow consume it without breaking.
6. Flipped interaction prompt template
When to use this AI prompt template: This template makes the AI ask you questions instead of the other way around. Use it for requirements gathering, troubleshooting, and practice sessions.Â
AI prompt template:
From now on, I want you to ask me questions to [achieve a specific goal]. Continue asking questions until you have enough information to [deliver specific outcome]. Then provide your final recommendation.
Goal: [what you want to achieve]
Stop when: [specific criteria for when to stop asking questions]
Please start by asking your first question.
Why it works: The AI takes the lead in gathering information through structured questioning, which surfaces blind spots you didn't know you had. (Whether this makes you feel more in control or more likely to experience an AI uprising is up to you, but it works.)
7. Expert persona prompt template
When to use this AI prompt template: Use it for professional analysis, industry-specific advice, and any task where the default "helpful assistant" register is wrong for the job.
AI prompt template:
Act as a [primary role] with the following specific expertise:
Credential 1: [specific qualification]
Credential 2: [specific experience]
Credential 3: [specific specialty]
Communication style: [how they communicate]
Key perspective: [what they prioritize]
Given this background, [specific task].
Respond as this persona would, including their professional terminology, typical concerns, and reasoning approach.
Why it works: The persona pattern is one of the most repeated pieces of prompt engineering advice, and also one of the most oversold. The specifics—like naming credentials, priorities, and communication style—do the real work of giving your AI concrete constraints.Â
8. Expert panel prompt template
When to use this AI prompt template: This template takes the persona pattern one step further by asking a panel of experts instead of just one. Use it for complex decision-making, strategic planning, and any choice where you suspect you've already made up your mind and need that decision tested.
AI prompt template:
I need multiple expert perspectives on [problem]. Please simulate a panel discussion with these experts:
Expert 1: [specific role] with expertise in [domain]
Expert 2: [specific role] with expertise in [domain]
Expert 3: [specific role] with expertise in [domain]
Have each expert:
State their credentials and perspective
Analyze the problem from their viewpoint
Propose their solution
Respond to other experts' viewpoints
Reach consensus or explain disagreements
Format as: Expert 1 (Title): [response]
You can also have your LLM identify the most relevant types of experts for the problem instead of naming them directly.Â
Why it works: The expert panel compels the AI to consider a problem from multiple viewpoints instead of collapsing into a single confident answer. It helps counter bias (to a degree) and often surfaces solutions you wouldn't have considered. I reach for it when I've been staring at a draft so long that I need someone to argue with me about it, and my editor knows better than to take the bait.
9. Decision matrix prompt template
When to use this AI prompt template: Use it when you're choosing between options (vendors, headlines, roadmap bets) and want them compared on the same dimensions instead of purely on ✨ vibes ✨.
AI prompt template:
Help me decide between [options].
Evaluate each option against these criteria, weighted by importance: [criterion 1, weight], [criterion 2, weight], [criterion 3, weight].
Score each option 1-10 on each criterion, show your reasoning, and present the results in a table with weighted totals. Then give me your recommendation, along with the strongest argument against it.
Why it works: Ask a model to compare things freeform, and you'll get a Switzerland-esque "it depends." A matrix pins every option to the same dimensions, makes the tradeoffs visible, and forces an actual recommendation. Requesting the strongest argument against that recommendation stress tests it before you act on it, because once a model picks a side, it tends to defend it.
10. Self-critique prompt template
When to use this AI prompt template: This template makes the model grade its own work against a rubric before you ever see it. Use it for anything with a quality bar you can articulate—for example, client-facing writing, documentation, or an AI agent that publishes without a human review step.
AI prompt template:
Complete [task]. Before showing me your final answer, evaluate your draft against these criteria:
[Criterion 1, e.g., "Every claim is specific enough to act on"]
[Criterion 2, e.g., "No jargon a new hire wouldn't know"]
[Criterion 3, e.g., "Under 300 words"]
Score each criterion 1-5. If anything scores below 4, revise before responding. Show me the final version along with your scores.
Why it works: First drafts from an AI have the same relationship to final drafts that mine do: distant. Generation and evaluation are different skills, and models are meaningfully better at spotting problems in existing text than at avoiding those problems while writing. Separating the two steps, with concrete criteria instead of "make it good," reliably raises the floor of what you get back.
11. Feedback loop prompt template
When to use this AI prompt template: The self-critique template has the model grade its own work, whereas this one puts your feedback in the loop. Use it when an output's close but off in a specific way. You can also use it as the revision step in any recurring workflow. (For AI agents, save the feedback that keeps coming up somewhere the agent can retrieve it, or you'll be giving the same note forever.)
AI prompt template:
Here's your previous output: [paste output]
Here's what worked: [what to keep]
Here's what didn't: [specific problems, with examples]
Revise the output to fix the problems while keeping what worked. Then tell me what you changed and why.
Why it works: Telling a model to "try again" makes it regenerate from scratch, discarding the parts that were already good. Feedback anchored to the actual output turns revision into a targeted edit, and asking the model to explain its changes lets you confirm it understood the note rather than just shuffling sentences.
12. Step-by-step (recipe) prompt template
When to use this AI prompt template: Use it for learning a new skill, planning a project, or mapping out an unfamiliar process.
AI prompt template:
Create a step-by-step guide to accomplish [specific goal]. Structure your response as:
Requirements:
[List what's needed]
Preparation:
[Setup steps]
Instructions:
[Step 1 with specific actions]
[Step 2 with specific actions]
[Continue...]
Troubleshooting:
Common issues and solutions
Variations:
Alternative approaches
Goal: [your specific objective]
Why it works: The structure requires the AI to give you every stage of the job: requirements, preparation, instructions, and the failure modes—just like following a recipe. Without it, models tend to skip straight to the middle and leave you to discover the prerequisites the hard way.
13. Fact-check prompt template
When to use this AI prompt template: I will always push for a human fact-check, but this template has AI do the first pass to catch obvious errors and misleading statements. Use it in journalism, research, and business analysis, and always follow up with a human review on anything sensitive. (I've found that running the same text through an AI fact-check twice can produce different results each time, which tells you everything about why the human pass still matters.)
AI prompt template:
I need you to thoroughly fact-check the following text. Please analyze every factual claim, statistic, date, name, technical specification, and verifiable statement.
Your response should ONLY include a "fact-check list" section with:
Claims that should be verified:
[Specific factual claim 1]
[Specific factual claim 2]
Information to double-check:
[Statistics, dates, or technical specifications]
Potentially inaccurate or questionable claims:
[Claims that seem false or dubious, contradictions within the text, implausible statements]
Vague or misleading statements:
[Statements that lack specificity or need sources]
Confidence levels:
High confidence: [claims you're very sure about]
Medium confidence: [claims that might need verification]
Low confidence: [claims you're uncertain about]
Text to fact-check: [PASTE YOUR TEXT HERE]
Why it works: Most chatbots can now ground their checks with live web search, which helps with dates and statistics, though it doesn't eliminate the need for verification. If hallucinations are a serious concern for your use case, pair this with retrieval-augmented generation so the model checks against your trusted sources instead of its memory.
14. Trigger-and-action agent prompt template
When to use this AI prompt template: Use this whenever you're writing instructions for agentic AI systems. For example, lead routing, inbox triage, or any workflow where "it depends" needs to be spelled out in advance.
AI prompt template:
You are an agent that runs when [trigger event]. Every time you run:
Check [the relevant data or condition]
If [condition A], then [specific action]
If [condition B], then [different action]
If you can't confidently determine which condition applies, [fallback action, e.g., "flag it for human review and stop"]
Always [standing requirement, e.g., "use the output format below"]. Never [hard boundary].
Why it works: Agent instructions usually run unattended, so every branch you don't specify is a decision the model will improvise, at whatever moment the edge case shows up. Writing instructions as trigger, conditions, actions, and a fallback covers the improvisation before it happens.
15. Guardrail prompt template
When to use this AI prompt template: Use it for any AI with access to real tools and real data, including agents that send messages, update records, or touch anything customer-facing.Â
AI prompt template:
Rules you must follow on every run, with no exceptions:
Never [prohibited action, e.g., "send an email to anyone outside the company domain"]
Only work with [explicit scope, e.g., "records created in the last 7 days"]
If a request or situation falls outside these rules, [escalation path, e.g., "stop and notify me instead of proceeding"]
Always [non-negotiable requirement, e.g., "include a link to the source record"]
Why it works: Models are eager to be helpful, and unconstrained helpfulness is exactly how an agent ends up messaging your entire contact list. Explicit prohibitions, a bounded scope, and a named escalation path give the model a safe default for every situation you didn't anticipate—which is most of them.
16. Approval checkpoint prompt template
When to use this AI prompt template: For work you can't afford to get wrong, build the pause into the prompt. Use it for long, multi-stage tasks, anything that gets sent to another human, and agentic workflows where a human sign-off belongs between the drafting and the doing.
AI prompt template:
Complete [task] in stages. After [stage, e.g., "drafting the outline"], stop and show me [the artifact] for approval before continuing. Do not proceed until I explicitly approve. If I request changes, revise and present the updated version for approval again.
Why it works: Course-correcting at an outline costs you a minute; course-correcting after the AI has produced (or worse, sent) the finished thing costs you the whole cycle, plus whatever the mistake touched. A checkpoint catches drift at the cheapest possible moment, and it keeps you in the loop exactly where your judgment adds the most value.
Put your AI prompt templates to work with Zapier
Remember how I said a bad prompt inside an agent keeps billing you for the same mistake? The reverse is true, too: a good prompt inside an AI-powered workflow keeps doing the right work without you needing to be there.Â
On Zapier, you can build your own agentic solutions with prompt templates from this list. Install Zapier MCP in Claude, ChatGPT, or whatever agent harness you use, and your templates can securely connect to 9,000+ apps without leaving the chat window.Â
Zapier is the most connected AI orchestration platform—integrating with thousands of apps from partners like Google, Salesforce, and Microsoft. Use forms, data tables, and logic to build secure, automated, AI-powered systems for your business-critical workflows across your organization's technology stack. Learn more.
AI prompt templates: FAQ
What's the difference between an AI prompt template and a prompt pattern?
A prompt pattern is the underlying structure (like "make the AI ask questions first"), and an AI prompt template is that pattern written out with placeholders so you can reuse it. Every prompt template in this article implements a prompt pattern.Â
Do I need different prompts for chatbots and AI agents?
The same patterns work for both, but AI agents need more from them. In chat, you can leave things vague and correct as you go. Agent instructions run without you, so they're more effective when given context, output format, edge-case handling, and guardrails written in up front.Â
Do these templates work across different AI models?
Yes. They're structural, not model-specific, so they work in ChatGPT, Claude, Gemini, and most other AI chatbots and agent builders.Â
Related reading:
This article was originally published in July 2025 by Maddy Osman. The most recent update was published in July 2026.
