# Tasks Tasks are PromptLoop's action tools that transform manual research into automated workflows. Each task takes specific inputs (like website URLs or search terms) and returns structured outputs that enhance your data. PromptLoop Tasks as AI agents that pre plan their steps so that you can get the benefit of autonomous AI research without the risks of unreliable actions on each row of data. **What makes tasks valuable:** - Run the same workflow on a single test case or thousands of data points - Get consistent, formatted results every time **Helpful Links** - [Create a task](/account/custom) - [Example Tasks to Copy](/account/custom/library) - [Your Account Tasks](/account/custom) <Callout title="New to PromptLoop?"> The fastest way to get started is by copying one of our proven task templates. You can easily customize it to fit your specific needs, and creating or editing tasks is completely free with all plans. Don't worry about making mistakes—tasks can be deleted or modified at any time. </Callout> ## Tasks Overview ### What is a Task? Tasks are tools to find and generate new data based on a set of input(s). These are AI Agents with guardrails that allow you to fill out a dataset with information in the same way over thousands of input rows. They are simple to create, edit and improve. Examples: <Table data={[ ["Find qualified leads for HVAC services", "Company website URL", "HVAC service offering (True/False)\nEmergency services (True/False)\nService area coverage (Text)\nCompany size (Number)"], ["Identify software compatibility", "Company website URL", "Uses ServiceTitan (True/False)\nCRM system (Single Category)\nIntegration capabilities (Text)\nTech stack (Multiple Categories)"], ["Research investment targets", "Company name (Search Term)", "Annual revenue range (Single Category)\nEmployee count (Number)\nRecent funding (Text)\nOwnership structure (Text)"], ["Monitor competitor pricing", "Competitor website URL", "Premium plan price (Number)\nFree trial length (Number)\nEnterprise offerings (True/False)\nFeature comparison (Text)"], ["Generate sales outreach content", "Prospect company name (Search Term)", "Personalized intro paragraph (Text)\nPain points addressed (Multiple Categories)\nCase study recommendation (Text)\nCall-to-action (Text)"] ]} columns={["User Intent", "Input Type", "Task Data Points (Output)"]} caption="Example tasks showing business needs, inputs, and resulting outputs" /> ### Structure **Input**: Each task has one or more inputs. Often this is a website that will be the basis of a search, but it could also be a company name. **Output**: Each task will return one or more datapoints. These will be new columns in your dataset. When you are testing a task one row of new information (for one input) will show up. Let's take an example where we want to do something simple. Find the main product or service a company sells and whether or not it serves other businesses or consumers. Our inputs would be company websites, usually a list of them in an uploaded dataset, and our outputs would be two new datapoints, that would return as new columns. **Example Structure** Find Product and B2B/B2C **Input Columns:** Company Website **Output Columns:** Product, Customer Type <InputOutputBadges inputs={[{name: "Input Name", description: "Company Website"}]} outputs={[{name: "Output Name", description: "Product"},{name: "Output Name 2", description: "Customer Type"}]} /> You can add, edit and customize each of the returned items. These will be new columns added to your dataset. When a task is a **List Task**, multiple rows are added. Each new row will have its own datapoints in the same format. ### Search Items and Task Editing <Image src={'https://img.promptloop.com/editor.jpg'} /> Each task is editable. This means you can add and adjust what the models are looking for. Search items are the results of your task, returned as new columns in your dataset. Each search item includes: - **Search Item Name** - The column header for this result - **Query** - The specific question or action to perform - **Action Instructions** - Additional guidance on formatting or handling special cases <Callout title="Multiple Search Items Limit"> While you can add additional search items to a task at no cost, beyond six items on a single task can in some cases degrade performance. For hard to find info or specific navigations we recommend splitting cases with many items up into multiple tasks. </Callout> ### Output Format Options <Table data={[ ["Text", "Free-form text response for descriptions, summaries, and explanations"], ["True/False", "Binary response for simple verification tasks and qualifying criteria"], ["Number", "Numeric output for counts, measurements, statistics, or values"], ["Link", "URL output for web links and resources"], ["Image Link", "Image URL output for tasks requiring image source extraction"], ["JSON", "Structured data containing multiple data points in standardized format"], ["Script", "Code or script output for generating executable commands"], ["Single Category", "One selection from predefined options for classification tasks"], ["Multiple Categories", "Multiple selections for tagging and multi-label classification"] ]} columns={["Format Type", "Best Used For"]} caption="Available output formats for task results" /> ### Task Types There are three main types of tasks: - **Website Tasks** - These take a website as an input, and return datapoints as part of the task. The model will navigated the site automatically depending on what you are looking for. This includes loading all available material, finding linked pages, and more. - **Search Task** - If you do not have a list of websites as a starting point, a search task allows you to use a search term to kick off a task. This will return similar datapoints but can start with something like an entity name or other search. - **List Task** - A list task is like a website task but returns multiple rows of data. You define each datapoint and describe the overall entity you are looking for, such as a person on a listing, and mutliple rows of data will be returned. This varies by site and happens automatically so its difficult to estimate how long it will run, or how many rows (credits) will be returned. <Image src={'https://img.promptloop.com/task-options-95c7336b.jpg'} /> You can select each when making tasks and templates fall into these categories. ## Creating a Task ### Step 1: Create a new task There are two simple ways to create a task that meets your needs: #### Option 1: Instant Generation If you prefer to describe what you need in plain language, use our simple prompt form. Think of it like sending an email request - just describe what data you're looking for and let PromptLoop build the task. **Important: Task Input Types** Tasks generally start with one of two input types: 1. **Website URL** - When you have specific websites to research. The task will analyze the content of each URL you provide. 2. **Search Term** - When you have entity names (companies, people, products) but need to find information about them. The task will first find relevant websites, then extract the data you need. #### Option 2: Copy a Template Starting with a pre-built template is the fastest way to get results. Our templates are proven to work and can be quickly customized for your specific needs. **Available Templates:** - **Find Information from a Website** - Extract specific data points from any web page. Perfect for researching company details, product information, or service offerings. - **Extract List from Website** - Build structured lists from pages that contain multiple items. Ideal for product catalogs, contact directories, or profile listings. ### Step 2: Test Your Task After creating your task, test it with a sample input to verify it works as expected. Enter a website URL or search term and click **Run** to see the results in real-time. <Image alt={'Custom Tasks AI job completion'} src={'https://web-public-photos.s3.amazonaws.com/testing-new.jpg'} width={800} height={420} /> Testing shows you exactly what data the task will return and identifies any potential issues before running it on your full dataset. You can immediately make adjustments if the results aren't what you expected. ### Step 3: Edit Your Task PromptLoop makes it easy to refine your tasks as your needs evolve. Toggle between **Edit** and **Test** using the buttons at the top of the task page. <Image alt={'Edit Task'} src={'https://img.promptloop.com/editor.jpg'} /> The editor allows you to modify output items and instructions to better achieve your goals. For major changes like modifying inputs or changing the fundamental operation, we recommend creating a new task. <Image alt={'Editing Tasks'} src={'https://web-public-photos.s3.amazonaws.com/CleanShot%202024-05-08%20at%2010.37.24%402x.jpg'} width={800} height={420} /> <Callout title="Not all Tasks are Editable"> Some complex tasks with multiple steps cannot be edited through the standard interface. If your task falls into this category, you can either reach out to our team for assistance or create a new task. Remember, task creation is free with all plans. </Callout> ### Step 4: Run Your Task on a Dataset Once your task is working as expected, run it on your full dataset. Upload your data containing the appropriate inputs (website URLs or search terms), and run your task to create a new version with all the outputs. <Image alt={'Datasets'} src={'https://web-public-photos.s3.amazonaws.com/CleanShot%202024-05-08%20at%2010.25.10%402x.jpg'} width={800} height={420} /> After processing, you can search, filter, and analyze your enriched data in the Datasets tab. All columns are preserved, with new columns added for each output from your task. ## Advanced Options and More ## Chained Tasks Transform complex workflows into simple automated processes by connecting multiple tasks together. Chained tasks are perfect for when you need to: - Enrich data through multiple steps - Process information sequentially - Combine different data sources ### How It Works <Callout title="Website Data Enrichment Chain Example"> 1. Task 1: Extract company details from a website URL 2. Task 2: Find LinkedIn company profile using company name from Task 1 **Input**: website URL ↓ Task 1 Output: company_name, company_email ↓ Task 2 Output: social_profiles **Final Output**: Complete company profile with website details and LinkedIn information </Callout> Chains streamline running multiple PromptLoop tasks that you use regularly. They're particularly valuable for comprehensive research workflows that would otherwise require multiple manual steps. ### Building Your First Chain 1. **Start with Inputs**: Add your initial data input (e.g., Website URL) 2. **Add Tasks**: Click "Add Step" to include each task in your sequence 3. **Map Data**: Connect outputs from previous steps to inputs of the next task 4. **Create Chain**: Save your workflow as a single automated process ### Current Limitations ⚠️ Note: Once created, chains cannot currently be edited. We recommend testing your workflow thoroughly before saving. ## Real-World Applications PromptLoop tasks handle an enormous variety of specialized research needs. Our customers use tasks for: - **Sales Intelligence**: Identify decision-makers, determine tech stack, and find qualification signals - **Market Analysis**: Determine if companies offer specific products or operate in particular markets - **Lead Enrichment**: Find contact information, company size, and services offered - **Competitor Research**: Track pricing changes, feature updates, and content publishing - **Investment Research**: Conduct due diligence, ownership research, and market sizing ### Example ROI Calculation "By automating research on 1,000 prospects weekly at 5 minutes saved per prospect, PromptLoop saves your team 83 hours per week - that's more than two full-time employees." ## Need More Help? PromptLoop is constantly evolving with new capabilities. If you have specific requirements or questions, please [request a demo](/demo) and our team will help you achieve your business goals. <Callout title="Ready to Transform Your Research?"> Get started today by creating your first task or schedule a [demo](/demo) to see how PromptLoop can work for your specific business needs. </Callout>

    Tasks

    Tasks are PromptLoop's action tools that transform manual research into automated workflows. Each task takes specific inputs (like website URLs or search terms) and returns structured outputs that enhance your data. PromptLoop Tasks as AI agents that pre plan their steps so that you can get the benefit of autonomous AI research without the risks of unreliable actions on each row of data.

    What makes tasks valuable:

    • Run the same workflow on a single test case or thousands of data points
    • Get consistent, formatted results every time

    Helpful Links

    New to PromptLoop?

    The fastest way to get started is by copying one of our proven task templates. You can easily customize it to fit your specific needs, and creating or editing tasks is completely free with all plans. Don't worry about making mistakes—tasks can be deleted or modified at any time.

    Tasks Overview

    What is a Task?

    Tasks are tools to find and generate new data based on a set of input(s). These are AI Agents with guardrails that allow you to fill out a dataset with information in the same way over thousands of input rows. They are simple to create, edit and improve.

    Examples:

    Example tasks showing business needs, inputs, and resulting outputs
    User IntentInput TypeTask Data Points (Output)
    Find qualified leads for HVAC servicesCompany website URLHVAC service offering (True/False) Emergency services (True/False) Service area coverage (Text) Company size (Number)
    Identify software compatibilityCompany website URLUses ServiceTitan (True/False) CRM system (Single Category) Integration capabilities (Text) Tech stack (Multiple Categories)
    Research investment targetsCompany name (Search Term)Annual revenue range (Single Category) Employee count (Number) Recent funding (Text) Ownership structure (Text)
    Monitor competitor pricingCompetitor website URLPremium plan price (Number) Free trial length (Number) Enterprise offerings (True/False) Feature comparison (Text)
    Generate sales outreach contentProspect company name (Search Term)Personalized intro paragraph (Text) Pain points addressed (Multiple Categories) Case study recommendation (Text) Call-to-action (Text)

    Structure

    Input: Each task has one or more inputs. Often this is a website that will be the basis of a search, but it could also be a company name.

    Output: Each task will return one or more datapoints. These will be new columns in your dataset. When you are testing a task one row of new information (for one input) will show up.

    Let's take an example where we want to do something simple. Find the main product or service a company sells and whether or not it serves other businesses or consumers. Our inputs would be company websites, usually a list of them in an uploaded dataset, and our outputs would be two new datapoints, that would return as new columns.

    Example Structure Find Product and B2B/B2C

    Input Columns: Company Website

    Output Columns: Product, Customer Type

    Company Website

    Product

    Customer Type

    You can add, edit and customize each of the returned items. These will be new columns added to your dataset. When a task is a List Task, multiple rows are added. Each new row will have its own datapoints in the same format.

    Search Items and Task Editing

    Each task is editable. This means you can add and adjust what the models are looking for. Search items are the results of your task, returned as new columns in your dataset. Each search item includes:

    • Search Item Name - The column header for this result
    • Query - The specific question or action to perform
    • Action Instructions - Additional guidance on formatting or handling special cases
    Multiple Search Items Limit

    While you can add additional search items to a task at no cost, beyond six items on a single task can in some cases degrade performance. For hard to find info or specific navigations we recommend splitting cases with many items up into multiple tasks.

    Output Format Options

    Available output formats for task results
    Format TypeBest Used For
    TextFree-form text response for descriptions, summaries, and explanations
    True/FalseBinary response for simple verification tasks and qualifying criteria
    NumberNumeric output for counts, measurements, statistics, or values
    LinkURL output for web links and resources
    Image LinkImage URL output for tasks requiring image source extraction
    JSONStructured data containing multiple data points in standardized format
    ScriptCode or script output for generating executable commands
    Single CategoryOne selection from predefined options for classification tasks
    Multiple CategoriesMultiple selections for tagging and multi-label classification

    Task Types

    There are three main types of tasks:

    • Website Tasks - These take a website as an input, and return datapoints as part of the task. The model will navigated the site automatically depending on what you are looking for. This includes loading all available material, finding linked pages, and more.
    • Search Task - If you do not have a list of websites as a starting point, a search task allows you to use a search term to kick off a task. This will return similar datapoints but can start with something like an entity name or other search.
    • List Task - A list task is like a website task but returns multiple rows of data. You define each datapoint and describe the overall entity you are looking for, such as a person on a listing, and mutliple rows of data will be returned. This varies by site and happens automatically so its difficult to estimate how long it will run, or how many rows (credits) will be returned.

    You can select each when making tasks and templates fall into these categories.

    Creating a Task

    Step 1: Create a new task

    There are two simple ways to create a task that meets your needs:

    Option 1: Instant Generation

    If you prefer to describe what you need in plain language, use our simple prompt form. Think of it like sending an email request - just describe what data you're looking for and let PromptLoop build the task.

    Important: Task Input Types

    Tasks generally start with one of two input types:

    1. Website URL - When you have specific websites to research. The task will analyze the content of each URL you provide.

    2. Search Term - When you have entity names (companies, people, products) but need to find information about them. The task will first find relevant websites, then extract the data you need.

    Option 2: Copy a Template

    Starting with a pre-built template is the fastest way to get results. Our templates are proven to work and can be quickly customized for your specific needs.

    Available Templates:

    • Find Information from a Website - Extract specific data points from any web page. Perfect for researching company details, product information, or service offerings.

    • Extract List from Website - Build structured lists from pages that contain multiple items. Ideal for product catalogs, contact directories, or profile listings.

    Step 2: Test Your Task

    After creating your task, test it with a sample input to verify it works as expected. Enter a website URL or search term and click Run to see the results in real-time.

    Custom Tasks AI job completion

    Testing shows you exactly what data the task will return and identifies any potential issues before running it on your full dataset. You can immediately make adjustments if the results aren't what you expected.

    Step 3: Edit Your Task

    PromptLoop makes it easy to refine your tasks as your needs evolve. Toggle between Edit and Test using the buttons at the top of the task page.

    Edit Task

    The editor allows you to modify output items and instructions to better achieve your goals. For major changes like modifying inputs or changing the fundamental operation, we recommend creating a new task.

    Editing Tasks
    Not all Tasks are Editable

    Some complex tasks with multiple steps cannot be edited through the standard interface. If your task falls into this category, you can either reach out to our team for assistance or create a new task. Remember, task creation is free with all plans.

    Step 4: Run Your Task on a Dataset

    Once your task is working as expected, run it on your full dataset. Upload your data containing the appropriate inputs (website URLs or search terms), and run your task to create a new version with all the outputs.

    Datasets

    After processing, you can search, filter, and analyze your enriched data in the Datasets tab. All columns are preserved, with new columns added for each output from your task.

    Advanced Options and More

    Chained Tasks

    Transform complex workflows into simple automated processes by connecting multiple tasks together. Chained tasks are perfect for when you need to:

    • Enrich data through multiple steps
    • Process information sequentially
    • Combine different data sources

    How It Works

    Website Data Enrichment Chain Example
    1. Task 1: Extract company details from a website URL
    2. Task 2: Find LinkedIn company profile using company name from Task 1

    Input: website URL

    ↓ Task 1 Output: company_name, company_email ↓ Task 2 Output: social_profiles

    Final Output: Complete company profile with website details and LinkedIn information

    Chains streamline running multiple PromptLoop tasks that you use regularly. They're particularly valuable for comprehensive research workflows that would otherwise require multiple manual steps.

    Building Your First Chain

    1. Start with Inputs: Add your initial data input (e.g., Website URL)
    2. Add Tasks: Click "Add Step" to include each task in your sequence
    3. Map Data: Connect outputs from previous steps to inputs of the next task
    4. Create Chain: Save your workflow as a single automated process

    Current Limitations

    ⚠️ Note: Once created, chains cannot currently be edited. We recommend testing your workflow thoroughly before saving.

    Real-World Applications

    PromptLoop tasks handle an enormous variety of specialized research needs. Our customers use tasks for:

    • Sales Intelligence: Identify decision-makers, determine tech stack, and find qualification signals
    • Market Analysis: Determine if companies offer specific products or operate in particular markets
    • Lead Enrichment: Find contact information, company size, and services offered
    • Competitor Research: Track pricing changes, feature updates, and content publishing
    • Investment Research: Conduct due diligence, ownership research, and market sizing

    Example ROI Calculation

    "By automating research on 1,000 prospects weekly at 5 minutes saved per prospect, PromptLoop saves your team 83 hours per week - that's more than two full-time employees."

    Need More Help?

    PromptLoop is constantly evolving with new capabilities. If you have specific requirements or questions, please request a demo and our team will help you achieve your business goals.

    Ready to Transform Your Research?

    Get started today by creating your first task or schedule a demo to see how PromptLoop can work for your specific business needs.