Deck

Confluence Guide10 min read

How to Analyze Customer Research Stored in Confluence

Last updated: · By Matt Teixeira, Founder and CEO

On this page
Short answer

To analyze customer research stored in Confluence, separate raw pages (interview notes, call recaps, advisory board minutes) from the findings pages that summarize them. Put every raw page on one labeled template that keeps the customer's exact words apart from your conclusions, and index them with a content properties report. Then use Rovo or an AI assistant to group the problems, and count how many separate accounts raised each one. To combine Confluence with support tickets and sales calls, connect Claude or ChatGPT through Atlassian's Rovo MCP server, or use a feedback tool such as Deck.

Key Takeaways

  • A findings page is one person's conclusion, not evidence. Count raw pages only, and trace every findings claim back to them.
  • The content properties macro and report (formerly page properties) turn labeled pages into a sortable research index with no add-on.
  • Rovo is included on Confluence Standard, Premium, and Enterprise, with 25, 70, and 150 credits per user a month (Atlassian). Claude and ChatGPT can read Confluence through Atlassian's Rovo MCP server (Atlassian). Check every AI answer against your index.
  • In Jira Product Discovery, clip insights from raw pages and track unique accounts per idea, not the insight count.

Why Is Customer Research in Confluence Hard to Synthesize?

Confluence scatters evidence across spaces and rewards conclusions over research. Interview notes live in Product, call recaps in Sales, escalations in Support, and search finds pages by keyword, not by account or problem.

Meanwhile a "Q3 Research Findings" page gets linked from every PRD while the 20 interviews behind it go unread. A quarter later, the team cites one person's summary as if it were 20 customers.

Step 1: Separate Raw Evidence from Derived Pages

Decide what counts before you analyze anything. Search the spaces where customer-facing teams write for the last 90 days of "interview," "customer call," "advisory board," and "escalation," then sort the results:

Page typeCounts as evidence?How to use it
Interview and discovery notesYesTag and count
Sales and CS call recapsYes, weighted lowerKeep as a separate type; they lean toward what closes the deal
Advisory board minutesYesAttribute each comment to its account
Incident post-mortemsFor impact onlyNote which accounts were hit, not what they want
Findings pages, PRDs, strategy docsNoTrace each claim to raw pages, then leave them out

As a rule of thumb, sorting 90 days of one team's space takes about an hour. Skip anything older; it describes a product you've since changed.

Treat a Findings Page as One Vote
Until each claim links to the raw pages behind it, a findings page shows what its author concluded, not what customers said.

Step 2: Give Every Research Page the Same Template and Labels

A shared template makes pages countable. Create a space template for customer conversations with a content properties macro at the top. Confluence reads the two-column table inside it as fields a report can collect (Atlassian Support).

Markdown template
Title: [Account], [YYYY-MM-DD], [Type]
Labels: customer-evidence, interview

[Content properties macro]
| Account   | Acme Logistics      |
| Date      | 2026-10-14          |
| Segment   | Mid-market          |
| Type      | Interview           |
| Recording | [link]              |
| Problems  | [filled in Step 4]  |

## What They Said (Verbatim)
Participant: "Every Monday I export the audit log and rebuild it in Excel."
Interviewer: "What do you change?"
Participant: "I add account names. Without them, security won't read it."

## What They Do Today
- The workaround, and how long it takes

## Our Interpretation
- Hypotheses, not evidence
  • Put three or four fixed labels in the template. Labels are free text, so interview and interviews quietly split into two reports.
  • Paste quotes from the transcript, one Speaker: text line per turn, and mark anything from memory as a paraphrase.
  • Keep the recording link so anyone can check a quote in context.

Step 3: Build a Research Index with the Content Properties Report

The index is your answer key for Step 4. On a page called Research Index, type /contentpropertiesreport and enter your evidence label. Confluence lists every matching page, with a column per property.

Several labels in one field match pages with any of them; labels in separate fields match pages with all of them (Atlassian Support).

  • Make one report per type (customer-evidence and interview in separate fields) so sales recaps don't mix with interviews.
  • Sort by Account to spot three pages from one customer at a glance.

Step 4: Use Rovo or an AI Assistant to Find Patterns Across Pages

Let AI do the grouping and let your index keep it honest. Rovo is on by default for Confluence Standard, Premium, and Enterprise, and admins can turn it off (Atlassian Support).

  • One page: Summarize drafts the Problems property for a long interview. Check the draft against the quotes.
  • Across pages: build a Rovo Studio agent limited to your research space, so it can't quote PRDs as the customer's voice. Under custom knowledge, an agent can be limited to chosen spaces or a branch of the page tree (Atlassian Support).
  • From Claude or ChatGPT: the Rovo MCP server lets them search Confluence, pull pages into a chat, and write a page back, within the signed-in user's permissions (setup guide).

Give it a fixed list of 8 to 15 customer problems to name themes from (how to build a feedback taxonomy), and make it show its evidence:

Markdown template
Use only pages labeled customer-evidence and interview, dated in the last 90 days.

1. List every page you read, by title. Say so if you couldn't open one.
2. Group the problems participants describe into 4 to 8 themes from the list below, named after the customer's problem.
3. For each theme, list the accounts and one verbatim quote per account from "What They Said", with the page link.
4. Never use "Our Interpretation" or unlabeled pages as evidence. Do not rank the themes.

Then compare the page list with your index. Assistants read only what their search returns, so a short list means skipped interviews. Page restrictions are the other trap: an assistant sees only what you can open, so two PMs can get different answers.

Step 5: Count Accounts, Then Send Findings to Jira Product Discovery

Count separate accounts per theme, not pages: five pages from one design partner are one opinion. A workable bar for "validated" is raw evidence from three unrelated accounts in your target market.

Then move findings to where you prioritize. With Atlassian's Chrome extension, select a quote on a Confluence page and add it to a Jira Product Discovery idea as an insight, or paste the page link into the idea (Atlassian Support). Jira Product Discovery is free for up to 3 creators; Standard is $10.50 per creator a month (Atlassian).

  • Clip from the raw page, not the findings page, or the next reader lands on an opinion.
  • Add a number field for unique accounts. The insight count grows with every clip, even five from one customer.
  • Embed the idea view in your Research Index so researchers see what their work fed.

Worked Example: 34 Confluence Pages Behind One Roadmap Debate

This example is illustrative. A 60-person B2B SaaS company plans in Jira and writes in Confluence. Its "H2 Research Findings" page ranks custom dashboards first, backed by an idea with 12 insights.

  • Sort: the PM finds 34 customer pages from the last 90 days across three spaces, plus 3 post-mortems.
  • Index: the 12 dashboard insights trace back to 4 pages, 3 of them from one account.
  • Pattern: a Claude prompt over the indexed pages finds "audit log exports need rebuilding" in 8 accounts; two export-timeout post-mortems add 5 more.
  • Decide: export reliability moves to the top with 13 accounts. Dashboards stay in discovery until two more segments are interviewed.

The findings page wasn't wrong. It was a careful read of four pages from two accounts.

Common Mistakes When Analyzing Confluence Research

  • Pointing AI at the whole site. Assistants search everything you can open, PRDs included, and will quote your own team as the customer.
  • Counting meeting notes as interviews. Meeting pages mix customer comments with internal decisions. Tag only the customer's words.
  • Editing last quarter's findings instead of rerunning them. Resynthesize from raw pages each quarter, or old conclusions harden into facts.

Tools That Help

Pick by volume. As a rule of thumb, labels and a report hold up to about 20 conversations a quarter. Past that, or once Confluence needs to sit next to tickets and calls, you want a tool that does the grouping.

OptionCostGood forMain limit
Labels and content properties reportIncluded in ConfluenceAn index of every interview, sortable by accountYou read and tag by hand
RovoIncluded on Standard and up; 25 to 150 credits per user a monthPage summaries and a research-space agentSees only what search returns
Claude or ChatGPT via Rovo MCPYour AI planOn-demand synthesis of a batch of pagesManual reruns; no memory of last quarter's themes
Jira Product DiscoveryFree up to 3 creators; Standard $10.50 per creator a monthLinking evidence to ideas next to Jira deliveryYou clip insights by hand, and themes come from ideas you've already logged, not from raw pages
DeckFree ($0, 200 AI credits a month, 1 integration); Starter $99 a monthResearch pages analyzed with tickets, calls, and Slack in one Opportunity BacklogWorks with Jira epics, not Jira Product Discovery ideas

Where Deck Fits

Deck automates Step 4 and the counting in Step 5 across all your sources, so a problem raised in three interview pages and 30 Zendesk tickets lands in one theme, each insight linked to its quote.

  • Connect: an admin can connect Confluence in Settings → Integrations through Atlassian's authorization (integrations page), but Deck doesn't yet pull pages in on its own: there's no page picker and no page sync, so bring research in as interviews.
  • Bring in raw pages, not findings. Deck treats everything you give it as customer feedback, findings pages included.
  • Paste full transcripts as interviews: in Manage → User Interviews → Add Interview, choose Interview Transcript and paste the text (import interviews playbook).
  • Jira handoff: Deck imports Jira epics as Projects and refreshes their title, status, and owner daily (import Jira epics). Its docs don't cover Jira Product Discovery ideas, so link the epic once an idea reaches delivery.
  • Limitations: no write-back to Confluence, no public voting board, no highlight reels. As a planning baseline, a 30-minute interview uses about 17 AI credits.

On the Free plan, pasted interviews don't use the one integration slot, and Jira doesn't count either (Deck billing docs). To test it, run one research round through the Free plan and compare Deck's themes with your index.

Frequently Asked Questions

Partly. Rovo, included on Confluence Standard, Premium, and Enterprise, summarizes single pages well, and Rovo Chat or a custom agent can answer questions across a space. Both read only what their search returns, so ask for the list of pages used and compare it with your content properties report before you trust a pattern.

Separate raw pages such as interview notes from findings pages and PRDs. Give raw pages one labeled template with a content properties table, index them with a content properties report, group problems with Rovo or an AI assistant, and count unique accounts per problem before you call anything a finding.

Yes. Atlassian's Rovo MCP server, generally available since February 2026, lets Claude, ChatGPT, and other MCP clients search Confluence, pull pages into a conversation, and create or update pages. It respects the signed-in user's permissions, and admins choose which AI clients may connect.

No. You highlight text on a Confluence page with Atlassian's Chrome extension and attach it to an idea as an insight, or paste the page link into the idea. Clip from raw interview pages, and track unique accounts per idea in a number field.

Deck lists Confluence as an integration, but it doesn't yet import pages on its own. For now, paste interview transcripts and notes into Deck's User Interviews, where they're analyzed alongside tickets and calls.

Sources

Sources checked November 2, 2026.