> ## Documentation Index
> Fetch the complete documentation index at: https://docs.strawberrybrowser.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Research workflows

> Run higher-quality Strawberry research with source rules, sampling, structured outputs, confidence, and review.

Research tasks are where Strawberry can save large blocks of time, but quality depends on clear instructions and review.

## Start with the decision

Tell Strawberry what decision the research supports.

```text theme={null}
I am deciding which 20 companies to prioritize for outbound this month.
Research fit, urgency signals, and likely buyer title.
```

The decision helps the companion choose sources and avoid irrelevant detail.

## Define source rules

Good source rules:

* Use official websites first.
* Use LinkedIn for company and people context.
* Use primary filings or databases when relevant.
* Avoid old blog posts unless they are the only available source.
* Mark uncertain data instead of guessing.

## Use samples before scale

For large tasks, ask for a representative sample first:

```text theme={null}
Do the first 10 companies, show me the output, and wait before scaling to 100.
```

This prevents wasted credits and catches formatting issues early.

## Request structured output

Use a table or spreadsheet when you need to compare results.

Include columns such as:

* Name
* URL
* Source
* Evidence
* Confidence
* Notes
* Next action

## Ask for gaps

Require a "missing or uncertain" section. This is better than forcing the companion to fill every cell.

## Review before acting

Before sending messages, updating CRM, or making business decisions, review:

* Source quality
* Data freshness
* Duplicates
* Uncertain rows
* Whether the output matches your constraints

## Best-practice prompt

```text theme={null}
Research 20 competitors in this market.
Use official websites, pricing pages, product docs, and recent primary sources.
Return a table with positioning, target customer, pricing model, strongest claim,
weakness, source URL, and confidence.
Do not guess. If data is missing, write "not found" and explain where you looked.
```