--- title: "How Briefwell Works: The Full Pipeline Explained" description: "A walkthrough of Briefwell's pipeline from source setup through collection, extraction, AI triage, and delivery as an Update, Report, or activity entry." url: "https://briefwell.ai/guides/how-briefwell-works" section: "Guides" updated: "2026-07-24" --- # How Briefwell Works Briefwell is set up per source with a defined objective and goal, then runs a repeating pipeline that collects, extracts, compares, and triages changes before delivering results. Briefwell works by watching the sources you choose on a schedule, comparing each check to the previous version to build a record of what changed, then having AI judge whether the change is notable and explain it. Notable changes are delivered two ways: an immediate Update for time-sensitive changes, or a scheduled Report that composes findings into a document with inline citations and optional human approval. Everything below the notability bar is kept in a searchable activity feed rather than discarded. ## At a glance - **Setup:** Point Briefwell at a page or a topic, name the details you want captured, set a standing goal, and choose a schedule and review policy. - **Preview:** A real preview against the live source, a sample of the AI output, and a real cost estimate run before anything is saved. - **Per-run pipeline:** Read the source, capture the details you named, and compare against the prior version to build a record of what changed. - **AI triage:** Judges notability against the source's goal and writes a plain-language explanation of what changed. - **Delivery:** Immediate Updates, cadence-based Reports with citations and optional approval, or the activity feed. - **Evidence:** Every result keeps its source snapshot and previous-to-new change record, traceable to the source. ## Setup: from objective to sandbox preview Setting up a source starts with picking what to watch — a page, a topic, a bundle of sources, or a collection — and naming, in plain English, the details you want captured, which appear as columns on each item. A standing goal or context is set once and steers how AI interprets future changes, so it does not need to be re-explained on each run. Delivery and review policy are chosen next: your schedule and urgency settings, not solely AI's judgment, determine what becomes an Update versus what waits for a Report. Before anything is activated, the source can be previewed — a real preview runs against the actual source, a sample of the AI output is shown on a real item, and a real cost estimate is calculated, all before anything is saved. The preview can be adjusted and rerun as many times as needed, since nothing persists until the source is activated. ## The per-run pipeline Once active, each run follows the same sequence: read the current state of the source, capture the details you named — using a fast, exact read first and asking the AI to interpret the page only when that comes up empty — and compare the result against the previous version to produce a record of what changed. That record then goes through AI triage, which judges notability using the standing goal set during setup. ## Where results go: Update, Report, or activity Based on the triage judgment and the configured review policy, a detected change becomes one of three things: an Update, raised immediately for event-driven, time-sensitive changes; an entry included in a scheduled Report, a composed document delivered on a cadence such as daily or weekly with inline citations and optional human approval; or an archived entry in the all-activity feed for changes that do not clear the notability bar, kept searchable rather than discarded. Every result keeps its source snapshot and the previous-value-to-new-value record of what changed, so it can be traced back to the source. ## What runs on top of the pipeline Beyond core delivery, a high-priority Update can be routed into a connected tool — automatically creating an issue, task, record, or row in tools like Linear, Jira, Notion, HubSpot, Salesforce, Airtable, or Google Sheets. This routing is opt-in and additive: it acts on an update rather than replacing delivery, and it never overrides a human decision. A per-workspace AI assistant can read updates, activity, sources, and channels to help triage, comment, or run a source in a bounded way. Results support Kanban triage, assignment, comments with mentions, custom fields that AI fills and a person can edit or revert, and a searchable, versioned history. ## How it works 1. Point Briefwell at the web pages, feeds, APIs, or research topics you want to watch. 2. It checks them on a schedule, compares each result with the previous version, and keeps a record of what changed. 3. When a change matters, Briefwell creates an update — judged and explained, with the evidence attached — and delivers it to Slack, email, your CRM, or a signed webhook. ## Questions ### What is the actual pipeline Briefwell runs? Read the source, capture the details you named — a fast, exact read first, with the AI interpreting the page only when that comes up empty — compare against the previous version, then run AI triage to judge notability before routing the result to an Update, a Report, or the activity feed. ### What is the Briefwell sandbox? A setup step where you paste a URL or define a research topic and run a real preview fetch, see sample AI output on a real item, and get a real cost estimate before activating the source. You can adjust and rerun it as many times as you want; nothing is saved until you activate. ### What is a per-source goal in Briefwell? A persistent objective or context set once per source that steers how AI interprets future changes on that source, so the goal does not need to be re-entered or re-tuned on every run. ### Does Briefwell notify on every detected change? No. Every detected change is compared and triaged, but only changes judged notable become an Update or are included in a Report; the rest are kept in the searchable activity feed. ### Can Briefwell act on an update in another tool? Yes. For high-priority Updates, Briefwell can automatically create an issue, task, record, or row in a connected tool such as Linear, Jira, Notion, HubSpot, or Salesforce. This routing is opt-in, additive to normal delivery, and never overrides a human decision. ## Related - [What Is Change Monitoring](https://briefwell.ai/guides/what-is-change-monitoring) - [Website Change Detection](https://briefwell.ai/guides/website-change-detection) - [Competitive Intelligence](https://briefwell.ai/guides/competitive-intelligence) - [AI Web Monitoring](https://briefwell.ai/guides/ai-web-monitoring) - [Updates vs Reports](https://briefwell.ai/guides/updates-vs-reports) - [Security and Privacy](https://briefwell.ai/guides/security-and-privacy)