When to run a workflow (vs. open a Focus)

A workflow runs the same task across every source in a dataset and returns a structured result: one row per document, one PDF per file, one audit per contract. A chat is a conversation, dataset-grounded but open-ended. Picking between them is mostly a question of "do I want the same thing from every document, or do I want to talk about them?"

The split

Workflow	Chat (Focus)
Non-interactive batch: kick off and check back	Conversational: back-and-forth
Same operation across every source	One question (or thread of questions) over the dataset
Structured output (spreadsheet, per-doc reports)	Markdown answers with citations and inline outputs
Good for N similar documents → N similar results	Good for exploring, comparing, building up understanding

Pick a workflow when

• You want the same shape of answer from each document: summaries, extracted fields, audit reports.
• You have many documents (tens to hundreds) and would otherwise repeat the same chat that many times.
• The output needs to be structured: a spreadsheet of extracted data, a folder of per-document summaries.
• You want to kick it off and come back later. Workflows persist; you can leave the page.

Pick a Focus when

• You want a conversation. Each turn is shaped by the last.
• You don't yet know the right question; you're exploring.
• The answer involves comparing or synthesizing across sources, not "do X to each".
• You want a single composed artifact at the end (one report, one chart) rather than per-source output.

The six workflows

Workflow	What it does	Output
Summarize a collection	One summary per document	Markdown / PDF per source
Extract structured data	Pull the same fields from every document	Single XLSX, one row per source
Audit against a reference	Compare each document to a template or checklist	Compliance report per source
Reduce AI signatures	Rewrite AI-generated text to sound more human	Revised text per source
Generate SEO metadata	Title, description, keywords, OG tags	Metadata block per source
Ask the same question of every document	One question, one answer per source	Per-document answers

When Ask docAnalyzer suggests one

If you're chatting with Ask docAnalyzer and your request looks batch-shaped ("summarize each of these 30 contracts", "find dates in all of them"), Ask will suggest a workflow rather than answer inline. Confirm and the workflow runs against the matching dataset. This is usually the right thing: Ask is making the same call you'd make manually.

How workflows run

All six workflows use the same loop: pick a dataset (a label, a multi-selection, or the result of a Smart Search), set a few inputs, kick it off. The run produces one result per source, with a downloadable artifact and a saved record you can come back to.

Workflows are credit-metered; see How credits work.

Reshape a result into a download

Workflow runs produce a result in one shape (a per-source list or a single answer). When you want that result as a downloadable file in a different shape, open the workflow chip in your chat. The chip opens the result in the side panel, and the chat box below stays available, so you can ask the chat to reshape it: filter rows, derive new columns, and produce a spreadsheet, PDF, HTML, chart, or any of the standard downloads. See Save and export your work for the full toolbox.

What's next

Pick a workflow above, or:

• Combine sources into a dataset: what feeds a workflow.
• How credits work: workflow runs deduct credits.