OpenStation · Marketing strategy · September 16, 2026

Turning the installs curve around

Active installs peaked near 2,700 in July and are losing about 75 sites a week. This is the plan to stop the bleed, switch on the distribution Automattic already owns, and put the AI story where prospects can find it.

North Star active installs on wordpress.org

Framework Generate, Compose, Customize (WordPress VIP guide)

Horizon 90 days

In short

It is a churn problem, not a downloads problem. Weekly downloads have been flat for two months, but with 30 releases in four months nearly all of them are updates. New-install acquisition is close to zero, and the May launch cohort is draining faster than anything replaces it.

We are stuck in Generate. High output (two releases a week, a blog post every three days, a brand system) with no pipeline from release to audience to install, no measurement of what converts, and no customization where our differentiation actually lives.

Proposed targets: hold 2,000+ through October, 3,000 by end of December, 5,000 by end of Q1 2027. These are commitments to the levers below, not forecasts.

Bet 1

Stop the bleed

A retention loop we can measure: first-run activation, a deactivation feedback dialog, support hygiene, a compatibility push. Net flat by end of October.

Bet 2

Turn on Automattic's own distribution

WordPress.com, VIP, Jetpack, WooCommerce, Studio and Playground are channels we own and are not using. The one lever with the scale to move the number by thousands.

Bet 3

Reposition for what the market is buying

Keep the workspace as the visual hook. Lead the enterprise and press story with the governed AI workbench, which today is hidden from the listing.

1. Where we are

2,048

Estimated active installs

−16% in 30 days

≈75

Sites lost per week

peak ≈2,700 mid July

173

Median daily downloads, last 30 days

186 the month before

Releases since May 7

one every 4 days

5.0

Rating on 23 reviews

22 five-star

Support threads resolved

1 open, visible on the listing

268

GitHub stars

41 forks, 50 open issues

Third-party reviews found

happas.jp; nothing from Tavern, WPBeginner, SEJ, YouTube

Downloads are updates

Weekly downloads (wp.org)Releases that week

Weekly downloads, May 20 to September 15, 2026. Bars scale to 5,000. Dots mark releases shipped that week. The one week with no release for a full seven days before it (June 24 to 30) did 404 downloads, about 58 a day, and that is the ceiling of organic acquisition.

The May cohort is the whole install base. The week of May 20 did 4,615 downloads, with 1,238 on May 21 alone. That was the hackathon and launch buzz. Nothing since has come close.
Every release generates about 2,000 update downloads over the following week. Weekly totals of 1,700 to 2,900 are an install base updating, plus a trickle of new installs.
The two spikes on the wp-rankings chart (Aug 27, Sep 3) are artifacts, not growth. They line up with the 1.1.4/1.1.5 and 1.1.6 releases. wp-rankings estimates installs partly from download velocity.
Net loss of 75 a week with gross adds under 60 a day means deactivations run somewhere between 300 and 500 a week. That is the number to attack first, and nothing in the product records why a site leaves.
Caveat. Playground launches from the openstation.me demo may count as wp.org downloads. If they do, organic acquisition is lower still. Measuring demo launches separately resolves this.

Why the cohort churns (hypotheses to test)

Curiosity installs. "Automattic built a desktop for wp-admin in a hackathon" is a try-it story, not a keep-it story.
Per-user opt-in hides the value. The plugin does nothing until a user clicks the admin-bar toggle. Site owners who installed for a team may never have seen it turn on anywhere.
Compatibility friction. Windows over plugin admin pages break in ways the compat layer has not seen. At 0% support resolution, a break is a reason to leave and not come back.
No re-engagement. Two releases a week ship real improvements, but the only channel to an installed site is a changelog written for engineers.
The AI story is invisible. WPVibe pulled 6,000 installs in ten weeks on "AI manages your site". Our agents are screenshot 4 and absent from the tags.

2. What the guide says, and what it means for us

The guide is written for enterprise marketing leaders assembling AI tooling. Its framework transfers directly to a small team marketing an open-source product, and its audience is also one of our best-fit customers.

The three modes, applied to our marketing

Mode	Guide's definition	OpenStation today	What good looks like
Generate	Out-of-the-box AI creating content fast. Easy early wins. Fragments at the workflow level.	High output: 30 releases, 10 blog posts in a month, a brand system. Inconsistent framing: readme says "wp-admin forgets", site says "One tab", changelog says "fix dock overlap". No voice file, no reuse.	One voice file and one prompt library. Every release produces one story, told once, reused everywhere. Measured on downstream installs, not volume.
Compose	Connecting systems so AI flows across workflows. Where most strategies break. Pick a few integrations and own them.	Nothing is connected. Release, blog, listing, social and metrics are separate manual steps. No funnel dashboard. Support forum and GitHub issues are disconnected.	Three integrations, owned: release to content, metrics to a weekly review, support to triage. Nothing else this quarter.
Customize	Tailored capability on proprietary data. Where differentiation happens. The mistake is avoiding it.	Our proprietary assets are the demo, the install base, and Automattic's channels. None is instrumented or worked.	The Playground demo as a guided tour. First-run and deactivation loops feeding the roadmap. A WordPress.com and VIP go-to-market only we can run.

The guide's five questions, answered for our stack

Where are we using AI across content and marketing workflows? Ad hoc, per person, for drafting. No shared prompts, no guardrails, no measurement.
Which workflows create the most friction? Release to audience. Two releases a week and no repeatable way to turn them into a story, a listing update, a social post and a re-engagement touch. Second: knowing whether anything worked.
Where is out-of-the-box enough, and where does customization differentiate? Drafting, translation, captions and changelog rewriting are Generate work and fine with a voice file. The demo, onboarding and the enterprise narrative are Customize work nobody else can do for us.
Do we have governance and ownership? No named owner for the North Star, no weekly review, no definition of "new install" versus "update".
What manual work still exists? Every release note is rewritten by hand at least twice. Metrics are checked by visiting wp-rankings.

The guide is also a channel map

WordPress VIP publishes this guide to sell governed, composable, AI-ready platforms to enterprise marketing teams. Read its "AI-ready platform layer" criteria against OpenStation:

Guide's criterion	OpenStation already has
Governance and control: role-based access, audit trails, approvals	Copilot is read-only by design. Agents are real WordPress users with a capability ceiling and per-agent ability allowlists. Destructive actions go through a confirm dialog.
Workflow consistency	Spaces (one desktop per job), workspaces, shared presence, saved layouts.
Composable, API-first integration	Abilities API for every AI tool. Documented `openstation_register_*` PHP and typed JS APIs. No SaaS relay, no API keys in the plugin.
Operational resilience	Windows open in 3 ms (measured). "The desktop pays for itself by the third screen."
Content readiness	Corkboard content graph. Copilot search over the site's own content.

We are the admin layer of the exact platform story VIP is selling, and VIP is an Automattic company. Section 5 turns this into a play.

3. The model: installs = acquisition minus churn

Active installs is a stock. The flows are new installs, deactivations and deletions. Every play below is tagged with the flow it moves. Three rules:

Retention first. At the current rate, every 100 new installs a week only buys 25 net. Fixing churn multiplies every acquisition play.
Owned distribution before earned. Automattic's channels reach millions of WordPress sites. A single WordPress.com placement outweighs a year of blog posts.
One message, two registers. The workspace is the visual hook for creators and agencies. Governed AI is why enterprise teams and the press pay attention. Same product, same demo, different first sentence.

4. Positioning

Current: "wp-admin forgets. OpenStation doesn't." on the listing, and "One tab. Your whole WordPress site." on the site. Both true, both productivity claims, and neither says why a site should install it this quarter.

Proposed umbrella: The workspace where humans and AI run a WordPress site together. Built by Automattic, open source, nothing leaves your server.

Creator and agency register (listing, site, video): "Open posts, media and settings side by side. Ask the assistant to find anything. Give an agent a job and watch it work, on your desktop, under your rules." Lead visual: an editor, the Media Library and an agent's activity window open at once.
Enterprise and press register (VIP, WP Tavern, SEJ): "WPVibe puts your site inside someone else's AI client through a closed relay. OpenStation puts governed AI inside your admin: read-only Copilot, agents that are WordPress users with a capability ceiling, every tool an Ability you can audit, no API keys or metering. Open source, GPL, by Automattic."

Concrete changes

Listing title and tags. Keep "OpenStation" first. Add the AI surface: ai-agents ai-assistant command-palette chatgpt claude. wp.org search weights title, tags, freshness, installs and support resolution. Today a search for "AI agent" cannot find us.
Screenshot order. Agents (now 4) and Cmd+K (now 3) move to positions 2 and 3. The listing video shows one agent job end to end inside the first 30 seconds.
Readme. The first paragraph stays. The second becomes the AI paragraph with the governance sentence. "For developers" gains one line on Abilities and MCP readiness.
Site. Headline matches the umbrella. The stats block still says "v1.0.0"; drive it from the wp.org API.

5. Acquisition plays, ranked

Effort is team-weeks. Impact is a judgment on net installs per quarter once running.

Tier 1 · Owned Automattic distribution Customize

Play	What it is	Effort	Impact
WordPress.com placement	Marketplace listing with a featured slot, one-click recommendation on Business and Commerce plans, a "Try the new admin" banner in Calypso for eligible sites. It already runs there.	2	Thousands. The only play at this scale.
WordPress VIP co-marketing	Pitch VIP's content team a follow-on to this guide: "The AI-ready admin layer: governed agents inside WordPress." Joint webinar, a case study with a VIP editorial customer, a spot on VIP's recommended plugins.	3	Tens of large multi-user sites, plus the enterprise narrative for press.
Studio preset and Playground blueprint	A "WordPress with OpenStation" starter in Studio and a listed Playground blueprint with sample content, three open windows and an agent pre-configured.	1	Hundreds, compounding through agencies.
Jetpack and WooCommerce channels	A WooCommerce blog post on the store desk. A Jetpack newsletter mention. Both audiences are the multi-screen admin user.	1	Hundreds.

Tier 2 · Listing and search Generate

Play	What it is	Effort	Impact
Listing rewrite	Title, tags, screenshots, video and readme per section 4.	0.5	Hundreds over a quarter. Raises conversion of every other play.
Support hygiene	Answer every wp.org thread within 48 hours and mark resolved. Route threads to GitHub issues.	2 h/week	Trust signal, and the cheapest churn fix.
Review velocity	Ask for a review in the product after a user's 5th session, once. 23 reviews at 5.0 is great and small. 100 changes search rank.	0.5	Rank lift, indirect.
Search content	Five evergreen pages on the queries people type: "wordpress admin multiple windows", "wordpress command palette", "wordpress ai agent plugin", "connect claude to wordpress". Each ends in the demo.	1	Slow, durable.

Tier 3 · Earned media and community Compose

Play	What it is	Effort	Impact
Press pitch, enterprise register	WP Tavern, Search Engine Journal, WPBeginner news, Post Status. Angle: "Automattic's open answer to AI site management: governed, self-hosted, no relay." Time it to a release that ships something concrete; MCP transport is the ideal hook.	1	WPVibe got 6,000 installs on one SEJ piece plus owned media. Ours is smaller without Awesome Motive's properties, but non-zero.
YouTube	A ready-made demo site and a 90-second script to WPTuts, WPCrafter, Jamie Marsland and two Spanish-language channels. Video converts this product.	1	Hundreds per video that lands.
International	happas.jp covered it unprompted. WPMarmite (FR), note.com (JP), Spanish pieces written directly.	0.5	Tens to hundreds.
WordCamp and Make	A talk on "agents as WordPress users", a Make/Core post on consuming the Abilities API, a demo link in both.	1	Developer mindshare, feeds agencies.
AI-client directories	Once the MCP transport ships (the competitive doc calls it "the missing 10%"), list in the Claude connectors and ChatGPT apps directories and mcpservers.org. Distribution outside wordpress.org entirely, and where WPVibe's curve came from.	eng 2 to 4, mkt 0.5	Potentially the second-largest lever after WordPress.com.

Tier 4 · Product-led Customize

Play	What it is	Effort	Impact
"Invite your team"	After a user enables OpenStation, offer to email the other users on the site a one-line invitation with the toggle link. Per-user opt-in is our retention risk; this turns it into spread.	1	Raises activation per installed site.
Shareable artifacts	A gallery for desktop themes, wallpapers and workspaces, uploadable from the product. Every share is a landing page.	2	Slow, community-building.
Multisite network story	1.1.8 shipped the network switcher. Agencies run multisite. One post, one video, one Tavern pitch.	0.5	Tens of networks, each many sites.

6. Retention plays (do these first)

Play	What it is	Why
Deactivation feedback dialog	On deactivate, one optional question: broke something / too slow / didn't understand it / not for me / other, with a text box. Anonymous, opt-in, common on wp.org. Results land in a table we read weekly.	We cannot fix churn we cannot see. The single most important instrument in the plan.
First-run activation	A one-time admin notice with the toggle, and a 20-second in-product tour on first enable: open a window, snap it, hit Cmd+K. Measure "enabled by at least one user within 7 days of install".	An install nobody turned on gets deleted at the next plugin cleanup.
What's New that speaks human	A dialog on first load after an update: three lines, one screenshot, written in the voice file.	Converts the release cadence from noise into a reason to keep the plugin.
Compatibility program	The 20 most-installed admin-heavy plugins (Yoast, Elementor, WPForms, ACF, WooCommerce, Rank Math, MonsterInsights, …). Test each in a window. Fix or document. Publish a compatibility page.	Each incompatibility is a deactivation and a one-star risk.
Support SLA	48-hour first response, thread marked resolved, GitHub issue when it is a bug. Owner named.	The listing's 0% resolved is both a churn cause and an acquisition drag.
Performance as a promise	Put the 3 ms and "third screen" numbers in the listing and the first-run tour.	Speed is the objection people expect from a windowed admin. Pre-empt it.

7. The marketing AI stack

The guide's core advice: audit Generate, connect a few things in Compose, customize only where it differentiates, name owners. Here is that stack for a team of two or three.

Generate: one voice, reusable

A voice file in the warehouse repo: the umbrella message, the two registers, banned words, three paragraphs from the readme that already sound right. Every AI-assisted draft starts from it.
A prompt library for the recurring jobs: PR list to human changelog; changelog to What's New; release to blog draft; blog to social snippets; English to Spanish, French and Japanese; screenshot captions; support first reply.
Approval stays human. Roberto approves anything public. The readme voice is an asset and brand drift is real.
Measure downstream. A post counts if it produced demo launches or installs, not because it shipped.

Compose: three integrations, owned

Release to content. A GitHub Action on tag collects merged PR titles, drafts the human changelog and What's New lines with the voice file, opens a draft post on openstation.blog through the REST API, and opens a Linear task for review. Nothing publishes automatically.
Metrics to review. A daily script pulls the wp.org info and download APIs, GitHub stars, the wp-rankings estimate, site and demo counters, and the deactivation table into one dashboard. A weekly Slack post with the North Star and leading indicators.
Support to triage. A daily check of the wp.org forum opens a Linear issue per new thread with a reply draft attached. First reply within 48 hours becomes mechanical.

Do not connect anything else this quarter. The guide's "connected but fragile" pattern is real at any team size.

Customize: where only we can win

The Playground demo as a guided tour. A real-looking site, three windows pre-arranged, a Space per job, an agent with one visible completed run, a five-step overlay. Instrument launches and step completion. It is the product's best salesperson and today it is a blank sandbox.
The retention loop. Deactivation reasons and first-run activation feed a monthly "why sites leave" review that picks one fix per month.
The Automattic go-to-market. Nobody else can put OpenStation on WordPress.com, in Studio, or in a VIP guide. This is customization in the guide's exact sense.

Ownership

Area	Owner
North Star and weekly review	Daniel (project lead)
Content, voice, blog, international	Roberto
Listing, support SLA, compatibility	One named engineer, rotating monthly
WordPress.com, VIP, Studio relationships	Daniel, with an Automattic sponsor from each product
Metrics pipeline and release automation	One engineer, two weeks, then maintenance

8. Measurement

North Star: estimated active installs, tracked daily from the wp.org bucket plus wp-rankings, and from our own estimate: downloads on days with no release in the previous 7 days, times 7, minus deactivation-dialog count. Reported as a 7-day rolling net change.

Leading indicator, weekly	Why it leads
New installs (quiet-day downloads estimate)	Acquisition, cleaned of update noise
Deactivation submissions and reasons	Churn, and why
Sites with one enabled user within 7 days of install	Activation
Playground demo launches and tour completions	Top of funnel we control
openstation.me visits by referrer	Which channel is working
wp.org reviews (count) and threads resolved (%)	Listing trust signals
GitHub stars, forks	Developer channel

Lagging, monthly: press placements, YouTube views on partner videos, WordPress.com and VIP milestones.

9. Ninety-day plan

Instrument and stop the obvious leaks

Weeks 1 to 2 · by September 30

Deactivation dialog, first-run notice and tour, What's New dialog. Ship in 1.2.0.
Listing rewrite: title, tags, screenshots, readme AI paragraph, site stats fixed.
Resolve the open support thread. Set the 48-hour SLA and the rotating owner.
Metrics pipeline v1 and weekly Slack post.
Voice file and prompt library.

Owned distribution and the demo

Weeks 3 to 6 · October

WordPress.com marketplace conversation started, QA pass on a Business site.
Playground blueprint with guided tour; Studio preset submitted.
Compatibility program: first 10 plugins, page live.
Release-to-content automation live.
WooCommerce post and first Spanish pitch.

Target: net flat by October 31

The AI story and earned media

Weeks 7 to 13 · November to mid December

VIP co-marketing pitch with the "AI-ready admin layer" draft.
MCP transport scoped; if it ships, directory listings and press pitch the same week.
Press pitch in the enterprise register with the WPVibe comparison.
Two YouTube partner videos.
"Invite your team" and the review prompt ship.

Target: 3,000 by December 31

10. What not to do

Do not chase the downloads chart. More releases lift downloads without lifting installs, and we already ship twice a week. Cadence is fine; narrative is missing.
Do not add telemetry without opt-in. wp.org rules and our "nothing leaves your server" message both forbid it. The deactivation dialog and first-run counter are opt-in and anonymous.
Do not build a relay or a SaaS. The self-hosted, no-key story is the differentiator against WPVibe.
Do not rewrite the readme's first paragraph. It is the best-written thing in our marketing. Add to it.
Do not connect more than three systems this quarter. The guide's "connected but fragile" pattern is one weekend of enthusiasm away.

Sources

wp-rankings.com/plugins/desktop-mode (estimated installs, momentum, resolution rate)
wordpress.org/plugins/desktop-mode and its info and download-stats APIs (installs bucket, downloads, ratings, support)
github.com/WordPress/openstation (stars, forks, release tags)
openstation.me and openstation.blog (site and blog audit)
happas.jp review (the one third-party review surfaced)
"Building the Enterprise AI Stack: A Framework for Marketing Leaders", WordPress VIP and Americaneagle.com
Internal: WPVibe growth analysis and code comparison (warehouse, competitive/)

Numbers as of September 16, 2026. The install estimate carries medium confidence per wp-rankings; the wp.org bucket is 2,000+. Targets are proposals to be agreed with the team.