Using PromptLoop Tasks

PromptLoop Tasks are no‑code AI agents. Give them a list of websites or search terms and they return clean, structured data—at any scale.

1. Why Tasks?

Automate research that normally takes hours per prospect or company.
Scale instantly from a single test row to thousands of data points.
Keep results consistent with strict formats, confidence scores, and guardrails.

New to PromptLoop?

Copy one of our proven task templates and tweak it. Task creation and edits are free on every plan, so experiment freely.

Helpful Links

2. Task Basics

Inputs vs. Outputs

	What it is	Example
Input	The starting value PromptLoop analyzes	https://acme.com · "Acme Corp"
Output	One or more data points the AI returns	description · phone_number · is_b2b

A single input row → one (or many) output columns. When using List format the task can add extra rows—one per list item.

Task Types

Task Type	Best When You Have	What It Does	Example
Website	A list of URLs	Crawls the site and extracts answers	Find pricing on competitor pages
Search	Only a name/keyword	Searches first, then crawls	Get LinkedIn URL for “Acme Corp”

Crawl Depth

Choose in the Browsing Type dropdown when creating/editing a Website task.

Depth	How Deep It Goes	When to Use
Single Page	Only the exact URL provided	You want data only from that page (e.g., marketplace listing)
Smart Crawl (default)	Follows relevant links a few levels deep	Most website tasks—fastest & cheapest
Deep (more credits)	Explores far more pages	Hard‑to‑find info buried deep in sites

3. Quick‑Start (Most Users)

Step 1 – Create a Task

Option A – AI‑Generated Task (fastest)

Go to Tasks → New Task.
Describe what you need and the desired columns.
Review the auto‑generated columns, tweak if required, and Save.

PromptLoop uses your prompt to pre‑fill each column’s search query and format, saving tons of setup time.

Option B – Copy a Template

Browse the template library and clone a task that’s close to your goal (e.g., Find Information from a Website, Extract List from Website).

Step 2 – Test & Iterate

Open the task and paste one sample input.
Click Run Test.
Review what information was found and hover over the info icon to see reasoning.
Tweak queries or formats, then retest.

Step 3 – Run on a Dataset

Upload or select a dataset with your inputs.
Click Launch Job and pick your task.
PromptLoop adds new columns or rows with the outputs.

Explore, filter, and export results from the Datasets tab.

4. Editing a Task

Toggle Edit ↔ Test at the top of the task page.

Search Items (Output Columns)

Field	What to Enter
Name	Column header—keep it unique and short
Query	Natural‑language question (e.g., “What is the main phone number?”)
Instructions	Extra rules (e.g., “≤ 10 words”)
Source Toggle	Add a _source column with the URL found
Format	Pick Text, Number, Category, List, etc.

Multiple Search Items Limit

Tasks with > 6 items can run slower or lose accuracy. For complex jobs split into multiple tasks.

Answer & Category Formats

Format	Best For
Text	Descriptions & summaries
True/False	Yes/no checks
Number	Counts, prices
Link	URLs
Image Link	Image URLs
JSON	Multi‑field objects
Script	CLI commands or code
Single Category	One label from options
Multiple Categories	Multi‑label tagging
List	One row per list item (people, addresses)

Categories ensure uniform outputs. Define the allowed options; anything else returns Not Found.

Lists extract repeatable items (people, products, etc.) from a single site page.

Generating Lists

Need a list gathered across the internet? Use Dataset Generation instead. List output here covers items inside one website.

Lists Recap

Adds extra rows—one per item.
Each list column you define becomes a standard output column.

5. Advanced Topics

Managed Tasks

Built In Solutions: Managed Tasks are PromptLoop-maintained, no-code AI agents you can run directly on your datasets to return clean, structured columns—at any scale.

Where you will find them

Datasets → Add Column → Managed Tasks Choose a task, map inputs, set parameters, preview, and Launch Job. New columns are appended to your dataset. Credits for the run are visible in the launch step.

What you can run today

AI Column Generate or transform text into clean, typed columns (summaries, categories, normalizations) with strict formats.
Email Enrich Given a person + company/domain, return a verified/probable work email and related metadata (confidence, source, etc.).
Phone Enrich Given a person + company/domain, return a verified/probable work or mobile phone

Credits

Credits vary by task and options and are shown clearly in the launch step before you confirm the job.

Chained Tasks

Run multiple tasks back‑to‑back in one workflow (e.g., crawl website → enrich via LinkedIn). Build a chain under Chains in the sidebar.

Immutable: once saved, a chain cannot be edited. Duplicate to make changes.

Building a Chain (step-by-step)

Open Chains in the sidebar and click New Chain.
Add Inputs – choose the dataset column that kicks off the workflow (e.g., website_url).
Add Step → Select Task – pick your first task (e.g., “Company Details”).
Map Outputs to Next Step – click the ➔ icon next to an output and choose which task input it feeds.
Add Additional Steps as needed (max 5).
Save Chain – give it a name and description. ⚠️ Chains are immutable; duplicate to edit later.
Run on Dataset – launch just like any other task; results from every step are appended to the dataset.

Best Practices

Start with Smart Crawl, switch to Deep only if data is missing.
Keep queries generic (not company‑specific).
Use Categories for anything you’ll filter or group on later.

6. Troubleshooting & FAQs

Why is my task returning blanks? Test the URL in a single‑page crawl to debug.
How many rows can a task handle? Tens of thousands; performance scales automatically.
Can I change the input column later? No; create a new task to switch input type.

Next Steps

Datasets – upload data, launch jobs, and analyze outputs.
Accuracy & Confidence – interpret scores and improve tasks.
Signals – automate change monitoring over time.

Ready to Transform Your Research?

Get started now by creating your first task, or schedule a demo to explore advanced use‑cases with our team.

Using PromptLoop Tasks

PromptLoop Tasks are no‑code AI agents. Give them a list of websites or search terms and they return clean, structured data—at any scale.

1. Why Tasks?

Automate research that normally takes hours per prospect or company.
Scale instantly from a single test row to thousands of data points.
Keep results consistent with strict formats, confidence scores, and guardrails.

New to PromptLoop?

Copy one of our proven task templates and tweak it. Task creation and edits are free on every plan, so experiment freely.

Helpful Links

2. Task Basics

Inputs vs. Outputs

	What it is	Example
Input	The starting value PromptLoop analyzes	https://acme.com · "Acme Corp"
Output	One or more data points the AI returns	description · phone_number · is_b2b

A single input row → one (or many) output columns. When using List format the task can add extra rows—one per list item.

Task Types

Task Type	Best When You Have	What It Does	Example
Website	A list of URLs	Crawls the site and extracts answers	Find pricing on competitor pages
Search	Only a name/keyword	Searches first, then crawls	Get LinkedIn URL for “Acme Corp”

Crawl Depth

Choose in the Browsing Type dropdown when creating/editing a Website task.

Depth	How Deep It Goes	When to Use
Single Page	Only the exact URL provided	You want data only from that page (e.g., marketplace listing)
Smart Crawl (default)	Follows relevant links a few levels deep	Most website tasks—fastest & cheapest
Deep (more credits)	Explores far more pages	Hard‑to‑find info buried deep in sites

3. Quick‑Start (Most Users)

Step 1 – Create a Task

Option A – AI‑Generated Task (fastest)

Go to Tasks → New Task.
Describe what you need and the desired columns.
Review the auto‑generated columns, tweak if required, and Save.

PromptLoop uses your prompt to pre‑fill each column’s search query and format, saving tons of setup time.

Option B – Copy a Template

Browse the template library and clone a task that’s close to your goal (e.g., Find Information from a Website, Extract List from Website).

Step 2 – Test & Iterate

Open the task and paste one sample input.
Click Run Test.
Review what information was found and hover over the info icon to see reasoning.
Tweak queries or formats, then retest.

Step 3 – Run on a Dataset

Upload or select a dataset with your inputs.
Click Launch Job and pick your task.
PromptLoop adds new columns or rows with the outputs.

Explore, filter, and export results from the Datasets tab.

4. Editing a Task

Toggle Edit ↔ Test at the top of the task page.

Search Items (Output Columns)

Field	What to Enter
Name	Column header—keep it unique and short
Query	Natural‑language question (e.g., “What is the main phone number?”)
Instructions	Extra rules (e.g., “≤ 10 words”)
Source Toggle	Add a _source column with the URL found
Format	Pick Text, Number, Category, List, etc.

Multiple Search Items Limit

Tasks with > 6 items can run slower or lose accuracy. For complex jobs split into multiple tasks.

Answer & Category Formats

Format	Best For
Text	Descriptions & summaries
True/False	Yes/no checks
Number	Counts, prices
Link	URLs
Image Link	Image URLs
JSON	Multi‑field objects
Script	CLI commands or code
Single Category	One label from options
Multiple Categories	Multi‑label tagging
List	One row per list item (people, addresses)

Categories ensure uniform outputs. Define the allowed options; anything else returns Not Found.

Lists extract repeatable items (people, products, etc.) from a single site page.

Generating Lists

Need a list gathered across the internet? Use Dataset Generation instead. List output here covers items inside one website.

Lists Recap

Adds extra rows—one per item.
Each list column you define becomes a standard output column.

5. Advanced Topics

Managed Tasks

Built In Solutions: Managed Tasks are PromptLoop-maintained, no-code AI agents you can run directly on your datasets to return clean, structured columns—at any scale.

Where you will find them

What you can run today

AI Column Generate or transform text into clean, typed columns (summaries, categories, normalizations) with strict formats.
Email Enrich Given a person + company/domain, return a verified/probable work email and related metadata (confidence, source, etc.).
Phone Enrich Given a person + company/domain, return a verified/probable work or mobile phone

Credits

Credits vary by task and options and are shown clearly in the launch step before you confirm the job.

Chained Tasks

Run multiple tasks back‑to‑back in one workflow (e.g., crawl website → enrich via LinkedIn). Build a chain under Chains in the sidebar.

Immutable: once saved, a chain cannot be edited. Duplicate to make changes.

Building a Chain (step-by-step)

Open Chains in the sidebar and click New Chain.
Add Inputs – choose the dataset column that kicks off the workflow (e.g., website_url).
Add Step → Select Task – pick your first task (e.g., “Company Details”).
Map Outputs to Next Step – click the ➔ icon next to an output and choose which task input it feeds.
Add Additional Steps as needed (max 5).
Save Chain – give it a name and description. ⚠️ Chains are immutable; duplicate to edit later.
Run on Dataset – launch just like any other task; results from every step are appended to the dataset.

Best Practices

Start with Smart Crawl, switch to Deep only if data is missing.
Keep queries generic (not company‑specific).
Use Categories for anything you’ll filter or group on later.

6. Troubleshooting & FAQs

Why is my task returning blanks? Test the URL in a single‑page crawl to debug.
How many rows can a task handle? Tens of thousands; performance scales automatically.
Can I change the input column later? No; create a new task to switch input type.

Next Steps

Datasets – upload data, launch jobs, and analyze outputs.
Accuracy & Confidence – interpret scores and improve tasks.
Signals – automate change monitoring over time.

Ready to Transform Your Research?

Get started now by creating your first task, or schedule a demo to explore advanced use‑cases with our team.