{"id":22926,"date":"2026-07-30T21:46:02","date_gmt":"2026-07-30T21:46:02","guid":{"rendered":"https:\/\/scannn.com\/16-ai-prompt-templates-for-better-ai-agent-outputs\/"},"modified":"2026-07-30T21:46:02","modified_gmt":"2026-07-30T21:46:02","slug":"16-ai-prompt-templates-for-better-ai-agent-outputs","status":"publish","type":"post","link":"https:\/\/scannn.com\/lv\/16-ai-prompt-templates-for-better-ai-agent-outputs\/","title":{"rendered":"16 AI Prompt Templates for Better AI Agent Outputs"},"content":{"rendered":"\n<div id=\"\">\n<p class=\"css-zgi8za e17ungnj5\">I&#8217;ve gone through a lot of painful trial and error with AI prompting\u2014<i class=\"css-0 e17ungnj13\">a lot<\/i>. 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&#8217;s a different story with AI agents. A weak AI prompt baked into an agent&#8217;s instructions produces the same bad output\u2014and bills you for the same mistake\u2014every single time it runs, with no one at the keyboard to catch it.\u00a0<\/p>\n<div>\n<aside data-testid=\"half-width-cta\" class=\"css-i2l7n9 e1b4wb1z0\">\n<p>Your AI can talk. Zapier MCP makes it act.<\/p>\n<\/aside>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\">I&#8217;ve rounded up 16 AI prompt templates that the Zapier team and I rely on to get usable outputs on the first try.\u00a0You can copy and paste them directly into your chat window or your AI agent&#8217;s instructions, or customize them to suit your needs.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Table of contents:<\/strong><\/p>\n<h2 class=\"css-1ilpmu8 e17ungnj0\">AI prompt templates at a glance <\/h2>\n<div>\n<table class=\"css-1estjpa e17ungnj15\">\n<thead>\n<tr>\n<th>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Prompt template<\/strong><\/p>\n<\/th>\n<th>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Use it when<\/strong><\/p>\n<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#context-setting\" target=\"_self\">Context-setting<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">The AI needs background it can&#8217;t guess (e.g., audience, constraints, and goals) <\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#standing-instructions\" target=\"_self\">Standing instructions<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">The same preferences should apply automatically to every task<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#placeholder\" target=\"_self\">Placeholder<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">The output must follow a fixed format without inventing details<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#few-shot\" target=\"_self\">Few-shot<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">A few examples show what you want faster than a description<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#output-schema\" target=\"_self\">Output schema<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">The output feeds a spreadsheet, CRM, or another workflow step<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#flipped-interaction\" target=\"_self\">Flipped interaction<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">You want the AI to interview you to surface missed information, questions, or opportunities <\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#expert-persona\" target=\"_self\">Expert persona<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">You need a specific professional lens, not a generic assistant<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#expert-panel\" target=\"_self\">Expert panel<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">You want several expert viewpoints debating a decision<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#decision-matrix\" target=\"_self\">Decision matrix<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">You&#8217;re comparing options on the same weighted criteria<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#self-critique\" target=\"_self\">Self-critique<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">The AI should grade its own draft against a rubric first<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#feedback-loop\" target=\"_self\">Feedback loop<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">The output was close, and your feedback should drive the revision<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#step-by-step\" target=\"_self\">Step-by-step (recipe)<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">You need a complete plan, including requirements, steps, and troubleshooting<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#fact-check\" target=\"_self\">Fact-check<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">You want an AI first pass on claims before a human fact-check<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#trigger-and-action\" target=\"_self\">Trigger-and-action<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">You&#8217;re writing instructions for an AI agent that runs unattended<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#guardrail\" target=\"_self\">Guardrail<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">The AI has real tool and data access and needs hard limits<\/p>\n<\/td>\n<\/tr>\n<tr>\n<td>\n<p class=\"css-zgi8za e17ungnj5\"><a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"#approval\" target=\"_self\">Approval checkpoint<\/a><\/p>\n<\/td>\n<td>\n<p class=\"css-zgi8za e17ungnj5\">The AI should pause for your sign-off before the high-stakes step<\/p>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<h2 class=\"css-1ilpmu8 e17ungnj0\">16 AI prompt templates to get you better outputs<\/h2>\n<h3 class=\"css-bmqgnp e17ungnj1\">1. Context-setting prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: Use it for any recurring work task where the background stays stable, like status updates, briefs, and customer replies. (It&#8217;s also the backbone of effective agent instructions, because an agent can&#8217;t stop mid-run to ask what you meant.)<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Before I make my request, here&#8217;s the context you need:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Background: [the situation and why this matters]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Audience: [who will read or use the output]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Constraints: [tone, length, format, things to avoid]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Success looks like: [what a great output accomplishes]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">With that context, here&#8217;s my request: [your request]<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: An <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/types-of-ai-models\/\" target=\"_self\">AI model<\/a> pulls from at least two sources when it answers: its training data and whatever&#8217;s loaded in its <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/context-window\/\" target=\"_self\">context window<\/a>. <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/ai-agent\/\" target=\"_self\">AI agents<\/a> can fill that window themselves by retrieving files or skills, but retrieval only works for information that&#8217;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 <i class=\"css-0 e17ungnj13\">your<\/i> version. (This is the same principle behind <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/context-engineering\/\" target=\"_self\">context engineering<\/a>, applied at the single-prompt level.)<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">2. Standing instructions prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template:<\/strong> Use it in your chatbot&#8217;s custom instructions, a project&#8217;s system prompt, or at the top of an AI agent&#8217;s instructions\u2014anywhere the same rules should hold on every run.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">From now on, apply these standing instructions to everything I ask:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">About me: [role, team, and what you work on]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Defaults: [tone, format, and length preferences]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Always: [non-negotiables, e.g., &#8220;cite a source for every statistic&#8221;]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Never: [e.g., &#8220;use jargon without defining it&#8221;]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">If a specific request conflicts with these instructions, follow the request and flag the conflict.<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: 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 <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"http:\/\/zapier.com\/blog\/how-to-use-chatgpt\/#custom-chatgpt\" target=\"_self\">ChatGPT offer custom instructions<\/a> and file-based agents like Claude Code check your project for an AGENTS.md before they start working. <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/claude-skills\/\" target=\"_self\">Claude Skills<\/a> run on the same idea: you package the instructions once, and Claude loads them whenever a task calls for them. If you&#8217;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.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">3. Placeholder prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: Use it for business communications, recurring reports, and any AI agent whose output needs to follow the same format on every run.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">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:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Your template with PLACEHOLDERS]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Now apply this template to: [your specific request]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Some examples of what a template with placeholders may look like:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Email: &#8220;Dear [NAME], your [PROJECT] is [STATUS] as of [DATE]&#8230;&#8221;<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Report: &#8220;Executive Summary: [SUMMARY] Key Findings: [FINDINGS]&#8230;&#8221;<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: When AI doesn&#8217;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&#8217;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.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">4. Few-shot prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: Use it to <i class=\"css-0 e17ungnj13\">show<\/i> the AI what you want instead of describing it. (&#8220;Shot&#8221; is AI-speak for an example included in your prompt.)\u00a0<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">I&#8217;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.<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Example 1 input: [input] \/ Example 1 output: [output]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Example 2 input: [input] \/ Example 2 output: [output]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Now apply the same pattern to: [your new input]<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/best-llm\/\" target=\"_self\">Large language models (LLMs)<\/a> 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.\u00a0<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">5. Output schema prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: 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.\u00a0<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Return your answer in exactly this structure:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Field 1]: [what belongs here]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Field 2]: [what belongs here]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Field 3]: [what belongs here]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">If a field doesn&#8217;t apply, write &#8220;N\/A&#8221; rather than omitting it. Do not add fields, commentary, or preamble outside this structure.<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works:<\/strong> 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.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">6. Flipped interaction prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: This template makes the AI ask <i class=\"css-0 e17ungnj13\">you<\/i> questions instead of the other way around. Use it for requirements gathering, troubleshooting, and practice sessions.\u00a0<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">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.<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Goal: [what you want to achieve]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Stop when: [specific criteria for when to stop asking questions]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Please start by asking your first question.<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: The AI takes the lead in gathering information through structured questioning, which surfaces blind spots you didn&#8217;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.)<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">7. Expert persona prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: Use it for professional analysis, industry-specific advice, and any task where the default &#8220;helpful assistant&#8221; register is wrong for the job.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Act as a [primary role] with the following specific expertise:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Credential 1: [specific qualification]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Credential 2: [specific experience]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Credential 3: [specific specialty]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Communication style: [how they communicate]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Key perspective: [what they prioritize]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Given this background, [specific task].<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Respond as this persona would, including their professional terminology, typical concerns, and reasoning approach.<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: The persona pattern is one of the most repeated pieces of <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/prompt-engineering\/\" target=\"_self\">prompt engineering<\/a> advice, and also one of the most oversold. The specifics\u2014like naming credentials, priorities, and communication style\u2014do the real work of giving your AI concrete constraints.\u00a0<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">8. Expert panel prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: 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&#8217;ve already made up your mind and need that decision tested.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">I need multiple expert perspectives on [problem]. Please simulate a panel discussion with these experts:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Expert 1: [specific role] with expertise in [domain]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Expert 2: [specific role] with expertise in [domain]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Expert 3: [specific role] with expertise in [domain]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Have each expert:<\/i><\/p>\n<ol class=\"css-1buo3wp e17ungnj8\">\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">State their credentials and perspective<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Analyze the problem from their viewpoint<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Propose their solution<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Respond to other experts&#8217; viewpoints<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Reach consensus or explain disagreements<\/i><\/p>\n<\/li>\n<\/ol>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Format as: Expert 1 (Title): [response]<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\">You can also have your LLM identify the most relevant types of experts for the problem instead of naming them directly.\u00a0<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: The expert panel compels the AI to consider a problem from multiple viewpoints instead of collapsing into a single confident answer. It helps counter <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/ai-ethics\/\" target=\"_self\">bias<\/a> (to a degree) and often surfaces solutions you wouldn&#8217;t have considered. I reach for it when I&#8217;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.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">9. Decision matrix prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: Use it when you&#8217;re choosing between options (vendors, headlines, roadmap bets) and want them compared on the same dimensions instead of purely on  vibes .<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Help me decide between [options].<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Evaluate each option against these criteria, weighted by importance: [criterion 1, weight], [criterion 2, weight], [criterion 3, weight].<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">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.<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: Ask a model to compare things freeform, and you&#8217;ll get a Switzerland-esque &#8220;it depends.&#8221; 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.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">10. Self-critique prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: 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\u2014for example, client-facing writing, documentation, or an AI agent that publishes without a human review step.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Complete [task]. Before showing me your final answer, evaluate your draft against these criteria:<\/i><\/p>\n<ul class=\"css-6hywhk e17ungnj9\">\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Criterion 1, e.g., &#8220;Every claim is specific enough to act on&#8221;]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Criterion 2, e.g., &#8220;No jargon a new hire wouldn&#8217;t know&#8221;]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Criterion 3, e.g., &#8220;Under 300 words&#8221;]<\/i><\/p>\n<\/li>\n<\/ul>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Score each criterion 1-5. If anything scores below 4, revise before responding. Show me the final version along with your scores.<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: 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 &#8220;make it good,&#8221; reliably raises the floor of what you get back.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">11. Feedback loop prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: 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&#8217;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&#8217;ll be giving the same note forever.)<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Here&#8217;s your previous output: [paste output]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Here&#8217;s what worked: [what to keep]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Here&#8217;s what didn&#8217;t: [specific problems, with examples]<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Revise the output to fix the problems while keeping what worked. Then tell me what you changed and why.<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: Telling a model to &#8220;try again&#8221; 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.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">12. Step-by-step (recipe) prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: Use it for <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/learning-new-skills\/\" target=\"_self\">learning a new skill<\/a>, planning a project, or mapping out an unfamiliar process.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Create a step-by-step guide to accomplish [specific goal]. Structure your response as:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Requirements:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Preparation:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Instructions:<\/i><\/p>\n<ol class=\"css-1buo3wp e17ungnj8\">\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Step 1 with specific actions]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Step 2 with specific actions]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Continue&#8230;]<\/i><\/p>\n<\/li>\n<\/ol>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Troubleshooting:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Variations:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Goal: [your specific objective]<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: The structure requires the AI to give you every stage of the job: requirements, preparation, instructions, and the failure modes\u2014just like following a recipe. Without it, models tend to skip straight to the middle and leave you to discover the prerequisites the hard way.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">13. Fact-check prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: 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&#8217;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.)<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">I need you to thoroughly fact-check the following text. Please analyze every factual claim, statistic, date, name, technical specification, and verifiable statement.<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Your response should ONLY include a &#8220;fact-check list&#8221; section with:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Claims that should be verified:<\/i><\/p>\n<ol class=\"css-1buo3wp e17ungnj8\">\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Specific factual claim 1]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">[Specific factual claim 2]<\/i><\/p>\n<\/li>\n<\/ol>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Information to double-check:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Potentially inaccurate or questionable claims:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Vague or misleading statements:<\/i><\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Confidence levels:<\/i><\/p>\n<ul class=\"css-6hywhk e17ungnj9\">\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">High confidence: [claims you&#8217;re very sure about]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Medium confidence: [claims that might need verification]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Low confidence: [claims you&#8217;re uncertain about]<\/i><\/p>\n<\/li>\n<\/ul>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Text to fact-check: [PASTE YOUR TEXT HERE]<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: Most chatbots can now ground their checks with live web search, which helps with dates and statistics, though it doesn&#8217;t eliminate the need for verification. If <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/ai-hallucinations\/\" target=\"_self\">hallucinations<\/a> are a serious concern for your use case, pair this with <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/retrieval-augmented-generation\/\" target=\"_self\">retrieval-augmented generation<\/a> so the model checks against your trusted sources instead of its memory.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">14. Trigger-and-action agent prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: Use this whenever you&#8217;re writing instructions for agentic AI systems. For example, lead routing, inbox triage, or any workflow where &#8220;it depends&#8221; needs to be spelled out in advance.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">You are an agent that runs when [trigger event]. Every time you run:<\/i><\/p>\n<ol class=\"css-1buo3wp e17ungnj8\">\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Check [the relevant data or condition]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">If [condition A], then [specific action]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">If [condition B], then [different action]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">If you can&#8217;t confidently determine which condition applies, [fallback action, e.g., &#8220;flag it for human review and stop&#8221;]<\/i><\/p>\n<\/li>\n<\/ol>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Always [standing requirement, e.g., &#8220;use the output format below&#8221;]. Never [hard boundary].<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: Agent instructions usually run unattended, so every branch you don&#8217;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.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">15. Guardrail prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: 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.\u00a0<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Rules you must follow on every run, with no exceptions:<\/i><\/p>\n<ul class=\"css-6hywhk e17ungnj9\">\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Never [prohibited action, e.g., &#8220;send an email to anyone outside the company domain&#8221;]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Only work with [explicit scope, e.g., &#8220;records created in the last 7 days&#8221;]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">If a request or situation falls outside these rules, [escalation path, e.g., &#8220;stop and notify me instead of proceeding&#8221;]<\/i><\/p>\n<\/li>\n<li class=\"css-oiqntf e17ungnj7\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Always [non-negotiable requirement, e.g., &#8220;include a link to the source record&#8221;]<\/i><\/p>\n<\/li>\n<\/ul>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: 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&#8217;t anticipate\u2014which is most of them.<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">16. Approval checkpoint prompt template<\/h3>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">When to use this AI prompt template<\/strong>: For work you can&#8217;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 <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/agentic-ai\/\" target=\"_self\">agentic workflows<\/a> where a human sign-off belongs between the drafting and the doing.<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">AI prompt template: <\/strong><\/p>\n<div class=\"css-bp306l ex3dp150\">\n<details open=\"\" data-zds=\"true\" class=\"css-17ootyu-Accordion\">\n<summary data-zds=\"true\" tabindex=\"0\" role=\"button\" aria-expanded=\"true\" class=\"css-1vrdju1-Accordion__summary\"><span data-zds=\"true\" class=\"css-yjx27x-Accordion__arrow\"><span aria-hidden=\"true\" data-testid=\"iconContainer\" data-zds=\"true\" class=\"css-mqmm1m-Icon--arrowSmallUp--animate--disable-pointer-events--24x24--neutral600\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" fill=\"none\" viewbox=\"0 0 24 24\" height=\"24\" width=\"24\" size=\"24\" color=\"neutral600\" name=\"arrowSmallUp\"><path fill=\"#2D2E2E\" d=\"M19 15.18v-2.62L12 6.7l-7 5.86v2.62l7-5.88 7 5.88Z\"\/><\/svg><\/span><\/span><\/summary>\n<div data-zds=\"true\" class=\"css-15b1gn0-Accordion__content\">\n<article class=\"css-1fdcukk\">\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">Complete [task] in stages. After [stage, e.g., &#8220;drafting the outline&#8221;], 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.<\/i><\/p>\n<\/article>\n<\/div>\n<\/details>\n<\/div>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Why it works<\/strong>: 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.<\/p>\n<h2 class=\"css-1ilpmu8 e17ungnj0\">Put your AI prompt templates to work with Zapier<\/h2>\n<p class=\"css-zgi8za e17ungnj5\">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 <i class=\"css-0 e17ungnj13\">right<\/i> work without you needing to be there.\u00a0<\/p>\n<p class=\"css-zgi8za e17ungnj5\">On <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/\" target=\"_blank\">Zapier<\/a>, you can build your own agentic solutions with prompt templates from this list. Install <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/mcp\" target=\"_blank\">Zapier MCP<\/a> in Claude, ChatGPT, or whatever agent harness you use, and your templates can securely connect to <a class=\"css-1wqcgat-Nav__link css-0\" data-zds=\"true\" href=\"https:\/\/zapier.com\/apps\" target=\"_blank\"><span class=\"css-1kza99e erlp8tt0\">9,000+ apps<\/span><\/a>\u00a0without leaving the chat window.\u00a0<\/p>\n<div class=\"css-edievn ewagcly0\">\n<p class=\"css-1jr6g2k e17ungnj6\">Zapier is the most connected AI orchestration platform\u2014integrating 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&#8217;s technology stack. <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/l\/contact-sales?demo_source=cs_blog_link_callout_contact_sales_shortzapierexplainer\" target=\"_blank\"><u class=\"css-0 e17ungnj14\">Learn more<\/u><\/a>. <\/p>\n<\/div>\n<h2 class=\"css-1ilpmu8 e17ungnj0\">AI prompt templates: FAQ<\/h2>\n<h3 class=\"css-bmqgnp e17ungnj1\">What&#8217;s the difference between an AI prompt template and a prompt pattern?<\/h3>\n<p class=\"css-zgi8za e17ungnj5\">A <strong class=\"css-0 e17ungnj11\">prompt pattern<\/strong> is the underlying structure (like &#8220;make the AI ask questions first&#8221;), and an <strong class=\"css-0 e17ungnj11\">AI prompt template<\/strong> is that pattern written out with placeholders so you can reuse it. Every prompt template in this article implements a prompt pattern.\u00a0<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">Do I need different prompts for chatbots and AI agents?<\/h3>\n<p class=\"css-zgi8za e17ungnj5\">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&#8217;re more effective when given context, output format, edge-case handling, and guardrails written in up front.\u00a0<\/p>\n<h3 class=\"css-bmqgnp e17ungnj1\">Do these templates work across different AI models?<\/h3>\n<p class=\"css-zgi8za e17ungnj5\">Yes. They&#8217;re structural, not model-specific, so they work in ChatGPT, Claude, Gemini, and most other <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/best-ai-chatbot\/\" target=\"_self\">AI chatbots<\/a> and <a class=\"css-19a5n3-Link\" data-color=\"primary\" data-weight=\"inherit\" data-zds=\"true\" href=\"https:\/\/zapier.com\/blog\/best-ai-agent-builder\/\" target=\"_self\">agent builders<\/a>.\u00a0<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><strong class=\"css-0 e17ungnj11\">Related reading<\/strong>:<\/p>\n<p class=\"css-zgi8za e17ungnj5\"><i class=\"css-0 e17ungnj13\">This article was originally published in July 2025 by Maddy Osman. The most recent update was published in July 2026. <\/i><\/p>\n<\/div>\n<p><a href=\"https:\/\/zapier.com\/blog\/ai-prompt-templates\/?utm_source=tldrmarketing\">Source link <\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>I&#8217;ve gone through a lot of painful trial and error with AI prompting\u2014a 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&#8217;s a different story with AI agents. A weak AI prompt baked into an agent&#8217;s instructions produces the [&hellip;]<\/p>\n","protected":false},"author":16,"featured_media":22927,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[143],"tags":[],"class_list":["post-22926","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai"],"_links":{"self":[{"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/posts\/22926","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/users\/16"}],"replies":[{"embeddable":true,"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/comments?post=22926"}],"version-history":[{"count":0,"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/posts\/22926\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/media\/22927"}],"wp:attachment":[{"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/media?parent=22926"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/categories?post=22926"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/scannn.com\/lv\/wp-json\/wp\/v2\/tags?post=22926"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}