Marketplace Discovery Is Fragmenting, and Small Vendors Feel It Most
In a previous piece I looked at how Atlassian Marketplace search has been rebuilt: a new ranking engine, intent matching, and trust signals feeding relevance. That analysis had a quiet assumption baked in, the same one most listing advice carries, that discovery means search. A customer types a phrase, a ranked list appears, and the game is to place well on it.
That assumption is aging fast. Search is now one of four doors into an app, and the other three are growing. More importantly, a vendor's ability to see into each door varies enormously, and the pattern of who can see what is not neutral. It favors the established.
The four channels
The first door is Marketplace search, the familiar one. A customer searches, ranking decides the result. It is also the channel a vendor can observe best, though that observation is partial rather than complete: Atlassian's vendor reporting covers your own search-keyword and zero-result data, you can connect a Google Analytics property to your listing for traffic and on-page behavior, and you can read your own rank for a phrase by searching. No impressions, no click-through rates, no systematic rank tracking, but a real measurable surface.
The second is in-product recommendation. Atlassian surfaces app suggestions inside its own products, in Jira, the app switcher, and parts of Confluence, where a customer can act on a suggestion without ever opening the Marketplace. For the vendor there is no direct signal here at all. An install that originates from an in-product suggestion is indistinguishable in your reporting from any other install. You cannot see how often you are suggested, to whom, or why.
The third is AI assistance, and it is the newest. Rovo rolled out across paid Cloud plans through 2025, reaching Premium and Enterprise in the spring and Standard plans in the autumn, with no free tier, and Atlassian shipped a Remote MCP Server that connects Jira and Confluence data to external AI assistants. The likely direction, more prediction than measured fact today, is that when a customer asks an assistant how to do something, the answer increasingly arrives as a native feature or a recommendation, and the trip to the search box never happens. The appetite is already visible: in Rustem Shiriiazdanov's zero-result analysis, the largest cluster of searches returning nothing in mid-2025 was AI and MCP related, customers hunting for capabilities the platform itself was racing to ship. This channel is structurally dark for vendors: Atlassian publishes nothing on Rovo queries or recommendations. There is no report to pull and no analytics to connect.
The fourth is the oldest and the easiest to forget: recommendation from outside the Marketplace entirely. Solution Partners and consultancies advise customers on which apps to adopt, and customers routinely act on that advice through relationships, private offers, and direct deals that never touch a search result. Visibility here is not fixed, it is relational. A vendor with strong partner relationships gets told what was recommended, to whom, and why; a vendor without them gets nothing. What stays constant is that Atlassian's own reporting leaves no native trace of this channel at all.
The asymmetry is the story
Lay the four doors side by side and a pattern appears. The one channel a vendor can genuinely observe and influence through craft, search, is the channel whose share of discovery is most plausibly shrinking. The channels growing around it are either structurally dark (AI), invisible by design (in-product), or visible only through relationships (partner).
Now ask who that pattern favors.
An established vendor has installs and reviews feeding the in-product and AI recommendation engines, whatever their exact mechanics, because recommendation systems generally lean on adoption and engagement signals. It has partner relationships built over years, which means both a presence in the fourth channel and a line of sight into it. For that vendor, discovery fragmentation is a measurement nuisance: attribution gets fuzzier, but the channels still work in its favor.
A small or new vendor holds the opposite hand. No install base feeding recommenders. No partner network recommending it or reporting back. Its entire discoverable surface is, in practice, the search box, the one channel where a well-written listing and a genuinely useful app compete on merit rather than accumulated position. Search is the small vendor's home turf precisely because it is the channel where craft can beat incumbency.
That reframes the engine change I covered in the previous piece in a more hopeful light than it might first appear. Intent-based ranking, whatever its incumbency risks, is still ranking: a surface where the right words and a real solution can win. The harder problem for a small vendor is everything outside the box, the growing share of discovery that happens in places it cannot see, cannot measure, and cannot enter without the relationships it has not yet built.
What a vendor can actually do
The honest version of the advice splits by channel.
For search, the playbook is real and within reach: write the searchable fields so they match genuine intent, earn reviews and respond to them, complete support details, qualify for the trust badges you can. This is the channel where effort converts most directly.
For in-product and AI recommendation, there is no direct playbook, and anyone selling one is ahead of the facts. The plausible indirect lever is the same engagement and quality signals that feed search ranking, on the reasonable but unverified assumption that recommendation systems read similar inputs. Treat that as a working hypothesis, not a tactic with a dashboard.
For the partner channel, the lever is the relationship itself. Partner programs, co-marketing, solution-partner outreach, and being genuinely easy to implement are the entry fee. For a small vendor this is slow and front-loaded, which is exactly why the channel tilts toward the established. But unlike the AI channel, it is at least a door that effort can open.
The thread across all four: a vendor's listing quality and product reality now radiate into channels beyond the one where they are written.
The search box is where you write; it is no longer the only place you are read.
Further reading, all linked above: Rovo availability and Standard-plan rollout; Atlassian's Remote MCP Server announcement; and the 2025 Marketplace search and zero-result analysis (Actonic). The previous piece in this analysis covered the Algolia to OpenSearch ranking change, the concentration collapse in search demand, and the rise of trust signals.