Research: projects, sources, and findings

Use Research to collect inputs in one place—interview notes, links, documents, and structured observations—then get discrete findings you can act on. Findings are grouped by status (New, Reviewed, Converted, Dismissed) and every card carries a source badge plus optional Audience truth and Data point badges. You can save a finding straight to the Insight Library as a candidate (no AI call, no credits) or run batch synthesize insights across multiple findings at once—batch synthesis uses AI credits and needs at least three findings. On /insights you can search, sort, load more pages, filter by Insight type and status (including Actioned and Archived), use Bulk select so every visible insight is checkable regardless of filters (bulk approve / reject runs on candidates in the selection only; bulk archive works on selected non-archived insights), archive or restore from each insight card, edit candidate text before you approve, then review how insights feed Intelligence (sidebar: between Research and Strategy).

Projects

Each research project is a container with its own sources and findings. On the Research page, projects appear as a horizontal row of chips: each chip shows the name, an amber “X new” badge when there are unreviewed findings, and a source count. Hover a chip to see last updated (relative time). Use Search projects to filter by name, turn on New findings only to hide projects with nothing new, and use Show archived when you need to find a project you previously archived.

Click New project to type a name, or use a Quick start chip (for example Competitive scan) to create a project in one click. Open the ⋯ menu on any chip to Rename, Duplicate shell (creates an empty project named “Copy of …” — sources are not copied), Export findings (CSV) (downloads all findings for that project with source titles), Archive project (hides it from the default list; data is kept), or Restore project when Show archived is on. Archiving asks for confirmation.

Adding sources

Choose a tab in the Add source panel. Every tab requires a project first; the primary button enables when that tab's required fields are filled (while a save request is sending, buttons stay disabled on every tab until it finishes—after that you can switch tabs normally even while findings extract in the background). Required-field issues show inline where possible (no toast-only errors). The form shows a live status line as the source saves and extracts.

• Paste text — free-form text, like workshop notes or a transcript paste. Title is required. Long text is trimmed to 50,000 characters on save.
• URL clip — enter a public page URL and click Fetch preview to load the page title, hostname, and a short content preview. Click Save URL clip to extract findings. If the site blocks automated fetching (paywalls, login-only pages), the form will reveal a manual paste fallback so you can still save the article text.
• File — upload a PDF, DOCX, TXT, MD, or CSV (max 10 MB). The form shows the file name, type, and size before you submit; unsupported types and oversized files are rejected client-side without a network round-trip. Long documents are split into chunks (max 8) for extraction.
• Competitor — name, positioning, and at least one strength + one weakness are required. Optional Analyse button (uses credits) takes a competitor URL and auto-fills the form; you can edit before saving. Strengths/weaknesses textareas show X / 5 and warn if you go past 5 lines (only the first 5 are saved). The save call uses one Synthesize call (credits).
• Observation — quick capture for one specific market signal (max 500 characters). Optional Where did you see this?, Market segment, and Confidence auto-expand once you start typing.
• Stakeholder — name, role, and notes are required. The conversation date sits above the fold for traceability; sentiment stays in the optional details panel. Use > in your notes to mark direct quotes.
• Audience — segment + note required. Quick-fill chips (SMB founders, Enterprise buyers, Consumers) populate the segment field. Optional Source of insight auto-expands once you describe what you learned.

After you save a source, the app extracts findings in the background. Extraction aligns implications with your current Brand Intelligence view (same merged workshop + approved strategy fields Halo uses), rather than stale workshop snapshots alone. The status line moves through Saving source… → Extracting findings… → N findings extracted — View findings. If extraction errors out, the form keeps a Retry extraction button visible; the source is not lost.

Drafts on the structured tabs (Paste text, URL clip, Observation, Stakeholder, Audience) auto-save to your browser after 3 seconds of idle typing and reload the next time you open the tab for the same project.

Reviewing findings

The findings panel groups every finding by status into four collapsible sections: New (expanded by default), Reviewed, Converted, and Dismissed. Each finding shows its finding text, a labelled strategic implication, an amber “HMW” pill for the suggested How might we question, and a source badge (e.g. Competitor · Acme, Document · pitch-deck.pdf, Market observation). An extra blue Audience truth badge marks findings that came from an audience note, and a gray Data point badge marks statistics pulled out of document uploads.

Per-finding actions

• Mark reviewed — moves the finding to the Reviewed group so the New list stays focused.
• Convert to insight — opens an inline panel on the card (no popup). The fields are pre-populated from the finding, and you can edit the insight text, the one-sentence summary, the insight type, and the strategy recommendation. Save as insight sends the finding straight to the Insight Library as a candidate—no AI call and no credits used.
• Edit — inline edit of the finding text and strategic implication, for when the AI draft needs a small fix.
• Dismiss — hides the finding with a 5-second Undo affordance in case you change your mind. Dismissed findings also show a Restore button in their group.

Bulk actions

Check the boxes on two or more findings (or use the per-group Select all) to reveal the bulk bar. From there you can Mark all reviewed, Dismiss selected, or Send to synthesis—the last option uses AI credits and requires at least three findings to run.

Keyboard shortcuts

While the findings panel is visible and you are not typing in a field, use J and K to move between findings, R to mark reviewed, C to open the convert panel, and D to dismiss.

Insight Library

When you approve an insight in the Library and tie it to a strategy field, the app links the recommendation back to the underlying research findings for traceability.

The Library renders each candidate with a layout tuned to its insight type: Tension insights show a side-by-side “A vs B” view with a lightning bolt divider and a How might we… prompt; Whitespace insights show a competitor gap next to an audience need plus a full-width gap statement; Audience truth insights lead with a large headline in the Segment needs X to accomplish Y format and list the findings that back it up. Pattern and Competitive implication insights use a generic card layout. Candidates share Approve, Approve + update strategy, Reject, and Archive. Approved and actioned insights also include Add to Brand Knowledge so you can save the insight as a user-selectable snippet in Brand Knowledge (nothing is injected until you pick that snippet in Chat or a tool). Approved, rejected, actioned, or other non-candidate rows still have Archive; the Archived status tab shows hidden items and each has Restore.