Jira intent detection community post

[#1]

Jira search just got smarter with intent detection

TL;DR

We’re updating Jira’s search so it can better understand what you’re trying to do and route you to the right place automatically:

• Work‑style queries (e.g. “tickets”, “ABC‑123”) keep using Jira’s issue search

• Question‑style and knowledge queries (e.g. “how do I onboard new engineers?”) can open Rovo Search, with an AI answer and relevant content from across your Atlassian and connected apps

• You’ll still be in control - it’s easier to switch between Jira issue results and Rovo Search when you need to

We’re rolling this out gradually to Jira Cloud customers over the coming weeks and closely monitoring performance, relevance, and feedback to ensure the best experience for all our Jira users.

---

Why we’re changing Jira search

Jira now has two powerful, but very different, search experiences:

• The Jira search results page - built for deterministic work search: issues, filters, JQL, exports, bulk actions

• Rovo Search - built for knowledge and cross‑tool search: natural language questions, policies, docs, and content from connected apps, with AI summaries

Today, Jira’s search always sends you to Jira’s issue search, even when your query is clearly a question that Rovo could answer better. That leads to a few problems:

• You have to manually discover and choose to search with Rovo

• It’s easy to feel like you’re in the “wrong place” and bounce between views

• We under‑use Rovo’s strengths for knowledge discovery and cross‑app queries

Our goal is simple:
Make Jira search intent‑aware, so you land in the right experience on the first try!

---

What’s changing

When you start your search in Jira, we now look at the intent behind your query and choose the best starting point after you hit enter/return.

1. Work queries stay in Jira issue search

For deterministic, work‑oriented queries, Jira will continue to open the issue search results you know today.

Examples:

• ABC-123 or ABC- (issue keys / project keys)

• ticket name

For these, Jira keeps doing what it already does well:

• Fast, reliable issue lists

• Familiar filters and JQL

• Bulk actions, exports, and all the power‑user tools

2. Knowledge queries route to Rovo Search

For non‑deterministic, question‑style queries, search can now send you straight to Rovo Search after you hit enter/return.

Examples:

• “What should I work on next?”

• “How do I set up incident management?”

• “Where is our Q3 planning doc?”

• “Who owns the billing service?”

In these cases, Rovo can:

• Answer your question directly with an AI summary

• Pull in relevant content from Jira, Confluence, and connected apps

• Suggest next steps, related questions, and key links

You get less hopping between surfaces, and more “I just typed what I meant and got what I needed!”

---

You stay in control

Intent detection is designed to help, not to trap you.

When you land on Jira issue results, you’ll have a clear way to “Search all apps” with Rovo if you realise you’re actually asking a broader question

When you land on Rovo Search, it’s easy to switch back to Jira issues if you really wanted a precise list to triage or bulk‑edit

We’re watching for any sign of “ping‑pong” behaviour (bouncing back and forth) and will tune the routing rules if we see confusion.

---

How intent detection works (at a high level)

Under the hood, Jira now looks at your query and classifies it as either:

• Deterministic / work search

  • JQL, full or partial issue keys, project keys

  • Numeric‑only queries

  • Many single‑word “work” terms
    → These stay in Jira issue search

• Non‑deterministic / knowledge search

  • Multi‑word natural language questions (“who/what/where/when/how/why”)

  • Jira‑context phrases with action/time/status signals (e.g. “show my work items from last month”, “summarise blockers for project X”)
    → These can route to Rovo Search

For this first release we’re using carefully‑tuned, rule‑based heuristics that we’ve experimented internally and tested with Jira customers. We’re also running a more advanced ML classifier in observation‑only mode behind the scenes to learn from real traffic and support smarter routing in the future.

A few important notes:

• Performance matters: we’re holding ourselves to strict latency guardrails so search stays fast

• Relevance and trust matter: we only scale up the intent detection routing if we see better success in Jira search metrics

• Language support: for now, intent detection will primarily run on English‑language queries. For non‑English tenants, Jira will continue to use the existing search behaviour while we improve the model

---

Availability and rollout

We’re rolling intent‑aware search out gradually to Jira Cloud:

• Starting with a small % of Standard tenants, then expanding to more Standard, Premium, and Enterprise sites

• Eligible customers need to have AI features enabled in their Atlassian organization

• Some cohorts (for example, highly regulated environments or tenants on restrictive release tracks) will be excluded initially while we validate performance and compliance

If you don’t see the new behaviour yet, you’re not missing anything - it just means your site is still on the existing search experience. We’ll expand coverage as soon as we’re confident in the results.

---

What’s next

This is our first step toward a smarter, intent‑aware Jira search experience. Next, we’re looking at:

• Refining the routing rules based on real‑world data and feedback

• Improving Jira work queries powered by AI (e.g. “my recent tickets”, “issues assigned to Alex last week”) so you can speak naturally and still get precise issue lists

• Expanding language support once our models perform well beyond English

---

We’d love your feedback

Your feedback will shape how we evolve this:

• If you’re in the rollout, please try a mix of issue‑style queries and knowledge search questions from Jira’s search and see where you land

• Let us know when it gets it right (landed you in exactly the right place), and when it gets it wrong (you felt stuck or had to hop around)

You can leave comments here on the Community post, or use the in‑product “Give feedback” link from Jira search/Rovo Search.

[Julia Foden]

Hi @Owen Wallis 

Can you confirm that it will still be possible to go directly in one click from the search box at the top of the screen to the Issue Navigator / Work Item Search? This is where I want to get to, always. I would prefer not to have to scroll through suggestions and other clutter but at least the link is there at the moment. Please do not force us to go through some circuitous route to get to the Issue Navigator.

[Debbie Jolley]

Totally agree with Julia - I always want to get to the Work Item search for a JQL query - literally nothing else on that dropdown is ever of interest to me so please don't hide that under any extra unnecessary layers.  And I go to JQL multiple times most days

[Savannah Ehrhardt]

Really hoping the JQL detection piece improves (or even just removes the parser part). I often am looking for a specific project and enter something in the top search bar like:

project = ABC

If it hit enter, it understands it's JQL and goes to the work item search thankfully. However, it invariably interprets it as:

project = ABC and status = Project and text ~ "="

which requires a bunch more clicks to remove the nonsense and get back to my original query. It would take way longer to type out my request in natural language so I'm not looking for that, although I understand some users probably are.

[Owen Wallis]

Hi @Julia Foden @Debbie Jolley thanks for the comments - yes you can go to Work item search in one click.

We have added a link to Work items (highlighted here) so you can go directly there.

@Savannah Ehrhardt we do want to improve the JQL functionality - its a common request. I would love to chat more about what you'd like to see. You can find a time with me here https://calendly.com/owallis (I am Australia based so hopefully timezones work).

[Rune Rasmussen]

Are there any plans to let individual users set their default for deterministic/non-deterministic search behaviour?

I, for example, do deterministic search 99% of the time, so having Search try and guess what kind of search I want to do sounds like something that would just be frustrating and detrimental.

[Owen Wallis]

Hi @Run no plans as yet. What we hope you will find is the intent detection, is your scenario, would just be routing you to Jira search anyways. If not please use the in-app feedback to let us know what's not working so well, thanks!