Mihai Perdum

Posted on Nov 5 Originally published at leanzero.net

Rovo Search in Jira and Confluence: How to Use It

#jira #confluence #rovo #atlassian

Atlassian's "Introducing Rovo Search for Jira" post has 144,633 views. Under it, one commenter wrote that it was "not returning anything for partial text matches", and that's the doubt we wanted to test. So on 4 November we ran 14 searches in Jira and Confluence both ways on our test site. One way was Rovo search with plain sentences. The other was the JQL and CQL most of us would write instead. On three of them text ~ found 0 work items and 0 pages while Rovo search put the right work item or page first (for two of them, on the Jira tab). On a fourth, JQL found nothing, CQL found one unrelated file and Rovo put the right page second. It missed on others, and I'll show you those too.

If you keep one thing from this, keep the habit. Ask in plain words and click the Jira or Confluence tab before you decide it didn't work. Go back to JQL when you need an exact field.

One thing to know before you read the numbers. Our test site is full of fixture data. For this piece we added six Jira work items and three Confluence pages. They're written the way real tickets get written, in different words from the ones you'd search with later (the six work items carry the label rovo-search-article). Eight of the nine showed up in Rovo search results 2 to 6 minutes after we created them; the ninth, a go/no-go sign-off, never matched any of our searches. The runs used Mihai's admin account, and he's the assignee on a lot of our fixture work items, so his name shows up in some of the result lists.

[[steps]]

Open Rovo search from the Search box in Jira or Confluence.
Run a plain-language search and narrow it with the app tabs and filters.
Compare it with JQL and CQL on searches where the words differ.
Add connectors so other tools show up in the same results.
Check what each person can see, because results follow the source app's permissions.

Note

Prerequisites

A Jira or Confluence Cloud site with Rovo turned on. Atlassian's 2025 launch post rolled Rovo search in Jira out gradually to Standard, Premium and Enterprise customers with AI enabled. It excluded Free plans, customers on a release track and highly regulated industries "for now".

For the JQL and CQL comparison: an Atlassian API token and curl. Any user can run the searches; you only see what you're allowed to see.

For connectors: organisation admin in admin.atlassian.com, plus admin rights in the other tool (a Google Workspace admin for domain-wide delegation, a SharePoint global admin for SharePoint).

Tested on wolfaenpak.atlassian.net, our test site, on 4 November 2026. The menus are Atlassian's current Cloud UI.

What Rovo search covers in Jira and Confluence

Rovo search is the full-page search that looks across your Atlassian apps at once. Atlassian's support page puts it like this: "you'll see results from across your Atlassian apps such as Confluence, Jira, and Atlas. If you have connected third-party apps such as Google Drive or Slack, you'll also see relevant results from those apps." On our site, the app chips across the top read All apps, Jira, Confluence, Projects, Goals, Bitbucket, Jira Service Management and Assets.

It matches differently from JQL. Atlassian's engineering team wrote that the query is rewritten first ("fixes spellings/typos as needed, and handles acronyms"). Then it's matched with keyword scoring and with KNN, which they describe as getting "the closest documents to a query based on semantic similarity". So a page about "your first week in the warehouse team" can match a search for onboarding a new joiner, even though it never uses either word. Ranking is personal too: content you or your collaborators created ranks higher, so two people can see a different order for the same words.

Don't mix it up with Ask AI in Jira's List view. That one turns your sentence into JQL ("Rovo in Jira can translate what you type into a Jira Query Language (JQL) code") and runs it. So it inherits everything JQL can't do.

Step 1: Open Rovo search in Jira or Confluence

Click the Search box in the top bar. A quick dropdown of recommended and recent items opens first, and it isn't Rovo search yet. Type your question and press Enter.

What opens depends on the app. In Jira, Enter took us to https://<your-site>.atlassian.net/jira/rovo-search?...&text=<your words>, a page headed "Search results". It has an AI Answer table on top and the apps listed on the right with their counts. Cmd+Enter and Ctrl+Enter opened the same page. Atlassian's 2025 launch post said Enter opens the issue navigator and Cmd+Enter opens Rovo search, so if your site still behaves that way, use Cmd+Enter.

In Confluence, Enter opens https://<your-site>.atlassian.net/wiki/search?text=<your words> with All apps, Jira, Confluence and the other app chips across the top, and Cmd+Enter opens the same page on the Confluence tab. We ran all 14 searches below on Atlassian's cross-app search page, opened by address at home.atlassian.com/o/<your-org>/search?...&text=. It has the same app tabs as Confluence's page in a different order, and for "drivers keep getting logged out" both showed the same first four results.

You'll know you're in Rovo search when the address contains /jira/rovo-search? (Jira) or /wiki/search? (Confluence). For our first search in Jira, the address bar should show this:

https://wolfaenpak.atlassian.net/jira/rovo-search?tab=recommendedActions&text=drivers+keep+getting+logged+out

The Jira page is headed Search results, and the Confluence one shows the app chips. We tried https://<your-site>.atlassian.net/search?text= too and got "Oops, you've found a dead link".

Step 2: Run a plain-language search and use the tabs and filters

Search for the problem the way a colleague would say it out loud. On the cross-app page you'll often get an Overview above the results, an AI-written answer built from them with "Links in this answer open in a new tab". The top result gets a Summarize button, and some results get Related pages.

Then click the app tab. On the cross-app All apps list, Jira work items often sat below 15 to 20 Confluence pages in our runs. For "drivers keep getting logged out", the bug we wanted was number 18 on All apps (17 Confluence results above it) and number 1 on the Jira tab. I didn't expect that. Jira's own results page from Step 1 put it first on All, so where you start matters too. For the exact word "Teradata", All apps showed 20 Confluence pages before the first work item, and the Jira tab put the three Teradata issues at 1 to 3.

The filters sit behind the icon at the top right. On All apps there are three: Last updated, Type and Contributor. Click the Jira tab and you also get Assignee, Reporter, Label, Status, Work I'm watching and Work I've commented on. The Confluence tab adds Owner, Content under, Label, Status, Verified, Search title only and Show archived content.

Clicking a result opens it in a new tab. There's no hover preview.

Step 3: Compare Rovo search with JQL and CQL

To see the difference on your own site, run the same sentence through JQL. This is the count call I used (swap in your site and token):

curl -s -u "$EMAIL:$API_TOKEN" -H 'Content-Type: application/json' \
  -X POST "https://<your-site>.atlassian.net/rest/api/3/search/approximate-count" \
  -d '{"jql":"text ~ \"drivers keep getting logged out\""}'

{"count":0}

You should see a count of 0 when no work item contains all of those words, which is what happened on ours. And the CQL version for Confluence pages:

curl -s -u "$EMAIL:$API_TOKEN" -G "https://<your-site>.atlassian.net/wiki/rest/api/search" \
  --data-urlencode 'cql=text ~ "what to do when someone leaves the company"' --data-urlencode limit=3 \
  | python3 -c 'import json,sys; print(json.load(sys.stdin)["totalSize"])'

Here are all 14 from our run, measured on the cross-app results page. Rank is the position in the Rovo list as rendered; JQL is text ~ "<the search>" and CQL is text ~ with the same words.

Search	Rovo search	JQL text ~	CQL text ~
drivers keep getting logged out	HFMOB-201 #1 on the Jira tab (#18 on All apps)	0	0
what to do when someone leaves the company	handover page #1, HSD-11 #2; HFADM-161 #1 on the Jira tab	0	0
wrong invoice amounts for Norwegian customers	HFBIL-401 "off by one øre" #1 on the Jira tab	0	0
pages about onboarding a new joiner	"Your first week in the warehouse team" #2	0	1 (unrelated)
people can't sign in	HSD-1, HSD-9, HSD-8 in the top 4; SAML certificate bug not found	1 (unrelated)	0
customer personal data exposed	HFSEC-121 #6 on the Jira tab; generic Overview	0	0
app takes too long to open	ORB-7 "Reduce cold start time" #24 on the Jira tab	0	2
proof of delivery	no POD item found	0	8
low emission zone routing	correct Overview; LEZ items from #18	0	3
who owns the release checklist	Overview from unrelated docs; the page naming the owner at #30	0	0
Lena Marlow	the runbook that mentions her #1; none of her assigned work	0	1
VPN	CGL1-79 #1	4	3
Teradata	DWHM-18, -13, -4 at #1-3 on the Jira tab	3	0
SAML	HFPLT-458 at #34	22	0

Why does JQL come back empty? Atlassian's own JQL page explains a lot of it. Fuzzy and proximity operators "no longer have an affect on searches". Jira "cannot search for work items containing parts of words" and leans on stemming instead. Small words like "will" and "not" are dropped. And the text field reads summary, description, environment, comments and free-text custom fields, but no attachments. Searching inside attachments is still a "Gathering Interest" request, JRACLOUD-1210. So if a ticket says "session token" and you type "logged out", JQL has nothing to grab.

And yep, Rovo missed some. "people can't sign in" found the Okta and SSO tickets but never our "SAML assertion rejected after IdP certificate rotation" bug. That's probably the one an admin would care about most. "proof of delivery" didn't connect to items that say POD, which JQL found 51 of with text ~ "POD". And the Overview for "who owns the release checklist" answered from a Word document and Atlassian's SDK docs while the runbook that names the owner sat at number 30. We ran seven of the searches again about ten minutes later. The top 11 results came back in the same order every time, but the Overview wording changed.

Keep JQL for exact fields, counts and automation

Rovo search for "Lena Marlow" put the runbook that mentions her by name first, and none of the work assigned to her. This found her work:

curl -s -u "$EMAIL:$API_TOKEN" -H 'Content-Type: application/json' \
  -X POST "https://<your-site>.atlassian.net/rest/api/3/search/approximate-count" \
  -d '{"jql":"assignee = \"Lena Marlow\""}'

{"count":491}

In our run, a name matched text that mentions the person and none of their assignments. Anything with a field, an exact count, a bulk change or an automation rule behind it is JQL's job. Atlassian's own Ask AI page says Rovo is "less useful" when you want to "analyze the work items to create charts, summaries" or when the search needs functions "not currently available in JQL". Rovo's result counts aren't something to report from either; the search doc warns that with Data Center connectors "Result counts may look higher than expected".

Step 4: Add connectors so Rovo search covers Slack, Google Drive and SharePoint

Connectors are set up by an organisation admin at admin.atlassian.com, under Connectors, then All connectors, then Add connector. Atlassian has renamed them Teamwork Graph connectors in the docs, so you'll see both names. The Connectors page in admin.atlassian.com sums them up in one sentence: "Connectors bring third-party app content into Atlassian for search and AI, either by indexing it in the Atlassian Teamwork Graph or querying the app in real time without storing content."

Our test org has none of these connected, so this step comes from Atlassian's setup guides and we haven't run it ourselves. The page lists guides for 31 connectors. Synced connectors copy and index the content, which Atlassian says gives "the highest quality for your search and chat". Google Drive and SharePoint work this way. Direct connectors ask the other app live: "Content is not stored or indexed by Atlassian." Smart Link connectors need no admin at all. They use the links people already paste, based on "an individual user's permissions and history".

Slack. For connections made after 2 October 2025, "this connector directly integrates through Slack's search. Your Slack messages are not indexed by Atlassian." Messages in channels and DMs are searchable. "Files and canvases are not searchable." If your Slack connection is older than that, it keeps indexing for now. To move it to live search you disconnect it, wait up to 48 hours with Slack missing from search, and reconnect. You can't go back ("This switch is not reversible"), and Atlassian says older connections will eventually have to switch.

Google Drive. The admin sets up domain-wide delegation in Google, then each person connects their own Google account when Rovo asks. Docs, Sheets, Slides, PDFs up to 10 MB and plain text get indexed; videos are metadata only. Give it time: "Expect a few hours before files start appearing in Search; large drives can take longer to finish indexing."

SharePoint. It covers SharePoint and OneDrive, needs a SharePoint global admin, skips encrypted documents and caps PDFs at 10 MB. We covered the Microsoft permission prompts and site limits in what the Rovo SharePoint connector reads and how to limit it, so I won't repeat it here.

You can scope Drive and SharePoint: Atlassian's 30 July 2026 post lists allowlists and blocklists for shared drives and SharePoint sites. It also has "Time-based ingestion" so only content after a date gets indexed. Sensitivity label blocking is "Currently not supported. Coming soon". When a connector is working, it shows up as its own row in the Connectors table; ours still says "Showing 1 item".

Step 5: Check what each person can see in Rovo search

Results follow the permissions in the source app. Atlassian's sync table spells out the cases. Adding restrictions to a shared document means it "Only shows up in search results for those with permissions", removing them makes it "Visible in Search for everyone using Rovo on your site", and a deleted document is "Removed from Rovo, including Search, Chat, and Agents". The page gives no figure for how fast a change syncs, so time it on your own site before you tell anyone it's instant.

The part that catches people is content nobody restricted: "Content that isn't restricted may appear for all users in search results". Check your sharing first. Look at what's open before you connect a Drive full of "anyone with the link" files.

To check, sign in as a colleague with less access (or ask one to help), search for a document you know is restricted, and confirm it doesn't appear. Do the same after you change a share in Drive or SharePoint. Atlassian also notes that "in some rare circumstances", Recently viewed can still show something you opened before you lost access.

Rovo search vs Rovo chat: which one to open

Use search when you want to find and open the thing: a ticket, a page, a file. Use Rovo chat when you want it read and summarised or acted on. Atlassian's search page says the answers "provide follow-up questions to dive deeper with Rovo Chat". The Chat link is in the same left menu too.

A practical note on credits: Atlassian's credits page lists "Rovo Search in Atlassian apps" as Free. Chat is billed differently; our extra usage tutorial covers that side.

Use Rovo search from code: the Rovo search API question

I couldn't find a public REST endpoint called the Rovo search API in Atlassian's docs. Their engineering blog says the UI calls an internal GraphQL search service, and it isn't documented for developers.

What you can use is the Atlassian MCP server. Its supported tools page lists search, marked beta, as "Semantic search across Jira, Confluence, and connected apps using natural language, through Rovo". Next to it are searchJiraIssuesUsingJql and searchConfluence for CQL. If your prompt mentions JQL or CQL, Atlassian says the server "may use the advanced search tools instead of Rovo search". On cost, the two Atlassian pages disagree. The MCP tools page says beta tools "will not consume Rovo Credits at this time", while the credits page lists search among the calls priced at variable credits. I'd check your usage report after a week of use.

If it didn't work

"Rovo couldn't generate an answer for "…". Try rephrasing your query." We got this on 4 of 14 searches in at least one run. The results list underneath was still there. In our tests the right item was often in it. Scroll past the message, or try the Jira tab.
"Oops, you've found a dead link" on <your-site>.atlassian.net/search. Press Enter from the Search box in Confluence, or use https://<your-site>.atlassian.net/wiki/search?text=<your words>.
A new work item or page doesn't show up. Ours took a few minutes. Connectors take longer. Google Drive warns of "a few hours". On the Community, one admin wrote that "for over 24 hours, nothing has been indexed" from SharePoint, where the reply suggested starting "with smaller set of pages (not whole directory)".
Drive results from outside your allowlist. A Community thread reported exactly that. The poster's own answer was that the allowlist controls what's indexed for everyone. A person who connected their own Google account can still get their own files at runtime, which matches Atlassian's line that "your own Google Drive access is used".
Partial words in Jira. JQL matches whole words and their stems, so "auth" won't find "authentication". Type the whole word, or add a * at the end in JQL.

What you have now

[[takeaways]]

A Rovo search habit. Enter from the Search box, plain words, then the Jira or Confluence tab before you judge the results.
When to switch to JQL or CQL. Exact fields and people's assignments, plus anything you count, bulk edit or automate. In our run, assignee = found 491 work items that a name search didn't show at all.
What connectors add. An org admin can bring Slack (searched live), Google Drive and SharePoint (indexed) into the same results, and each person only sees what the source app lets them open.
What this doesn't cover. Our org has no third-party connector. The connector parts are what Atlassian documents; we didn't test them ourselves.

If you run the same test on your site, tell me which searches came back weird, especially if All apps buried your work items the way it buried ours.

Originally published on leanzero.net. More Atlassian, Forge and local-AI write-ups at leanzero.net/blog, and if you're planning a migration or a Forge app, that's what we do: leanzero.net/services.

DEV Community