Your plugin passed review and the directory listing is live, but few people connect it. Most advice about getting found in ChatGPT was written for the App Directory, which OpenAI replaced with the Plugin Directory on July 9, 2026. This guide covers what changed, the ways people actually find and start using a plugin, a listing checklist with OpenAI's own limits, the rules that block approval or monetization, which channel fits which kind of plugin, a 30-day plan, example prompts and a first-run message you can copy, and the numbers to watch each week. It is useful without a Soloop account.
The short answer
To get users for a ChatGPT plugin, treat the directory as one channel, not the plan. Write the listing and example prompts around one job, make the first connection and first result quick, and bring the people who already have that job to the plugin: your existing users, the workspace admins who can install it for a team, and the communities where the job is discussed.
OpenAI says plugins with strong real-world utility and high user satisfaction may get more prominent placement or be suggested proactively, but there is no application for featuring. The parts you control are a clear listing, a setup that works on the first try, and a plugin people keep using. Measure connections, first successful tasks and repeat use in your own product, and change one thing at a time.
What changed for ChatGPT plugins in 2026
The word plugin is back, but this is not the 2023 plugins beta, which ended in 2024. In 2026 a plugin is a package that can bundle skills, connected apps, app templates and interactive extensions, and it works across ChatGPT and Codex.
| Dimension | What changed | What it means for getting users |
|---|---|---|
| Directory (July 9) | The App Directory was replaced by the Plugin Directory. Existing app connections were not affected. | Guides about the App Directory are out of date. Check your listing in the Plugins section of ChatGPT and update links that point to old pages. |
| What a plugin contains | Skills (instructions), connected apps (an MCP server), app templates for admins and interactive extensions. A plugin can be skills only. | A skills-only plugin needs no account connection, which removes the biggest setup step. Add an app only where the job needs your data or actions. |
| Listing assets | Screenshots are no longer shown in the directory; you provide example prompts instead. | The example prompts are your storefront. Write them as real tasks your users ask for. |
| Interactive extensions (September 29) | Plugins can add panels, sidebar views, file viewers or editors and forms beside the conversation. | A visible interface can shorten the first result, but check which ChatGPT clients and plans your users have before relying on it. |
| Custom GPT retirement | OpenAI plans to retire custom GPTs and offers a migration path to plugins. New GPT creation ended on October 26, 2026 for affected Enterprise workspaces. | If you had a public GPT, plan its move to a plugin and tell its users where to find the new version. |
July 9, 2026: the App Directory was replaced with the Plugin Directory and existing app connections were unaffected; September 11: planned custom GPT retirement with migration to plugins; September 29: interactive plugin extensions.
OpenAI Help Center: ChatGPT release notesScreenshots are no longer shown in the directory; developers provide example prompts instead.
OpenAI Developers: plugin guidelinesHow people find and start using a plugin
There are more entry points than the directory's search box. Each one asks something different of your listing and setup.
| Dimension | How it works | What you can influence |
|---|---|---|
| Browsing the directory | The Plugins section lists public, workspace, personal and installed plugins, with categories such as Popular and New & Noteworthy. | The name, short description and example prompts. Placement in curated categories is OpenAI's decision. |
| A suggestion in a conversation | ChatGPT can suggest a plugin while someone works on a related task. | No criteria are published. OpenAI ties enhanced distribution to utility and user satisfaction, so a plugin that solves the task well is the only lever. |
| Automatic use after install | ChatGPT uses installed plugins when they are relevant, and people can call one with an @ mention. | Clear tool descriptions and a narrow purpose, so the model chooses your plugin for the right requests and not the wrong ones. |
| Workspace installation | Admins of Business and Enterprise workspaces can make a plugin available or install it for members, and can import plugin marketplaces from GitHub. | For a team product, one admin conversation can reach a whole workspace. Give admins setup notes and a data-handling summary. |
| OpenAI Verified badge | OpenAI works with selected developers to review plugins for quality, reliability and usefulness. | There is no public application; treat it as a result of quality, not a launch tactic. |
Directory with Public, workspace, Personal and Installed views and categories such as Popular and New & Noteworthy; plugins suggested during a conversation; installed plugins used automatically when relevant or by @ mention; admins set Available or Installed policies and can import GitHub marketplaces; OpenAI Verified badge for selected developers.
OpenAI Help Center: Plugins in ChatGPT (updated Oct 28, 2026)Plugins with strong real-world utility and high user satisfaction may be eligible for directory placement or proactive suggestions.
OpenAI Developers: plugin guidelinesWhat one launch shows
In an October 16, 2026 case study, OpenAI reports that Stack Overflow built a ChatGPT plugin for its Stack Overflow for Agents product so people could discover and set it up in fewer steps than a manual MCP setup. Its VP of Product says registrations tripled overnight and agent creation rose up to tenfold.
Treat this as one reported example from a well-known brand in a vendor case study, not a benchmark. The useful lesson is the mechanism: the plugin removed setup steps for people who already wanted the product. A plugin rarely creates demand that did not exist; it makes existing demand easier to act on.
Reported: registrations tripled overnight and agent creation up to 10× after launching the plugin; the plugin made discovery and setup take fewer steps than an MCP setup.
OpenAI: Stack Overflow for Agents triples signups with ChatGPT plugin (Oct 16, 2026)ChatGPT plugin listing checklist
Check each item before you send anyone to the listing. Several are review requirements, so missing them delays the launch itself.
- Verified identity: complete individual or business verification so the plugin is published under your name or company.
- Display name: up to 30 characters, specific to the job, not a generic word, and without "MCP", "MCP Server" or "Plugin" added to the product name.
- Short description: up to 30 characters that say what it does for whom.
- Long description: up to 4,000 characters that describe the functionality accurately, without comparing to or disparaging other products, unverifiable claims, or promotion of pricing, trials or discounts.
- Example prompts: real tasks in your users' words, each one showing a result the plugin can deliver on the first try.
- A square icon of at least 48×48 pixels, plus an optional dark variant.
- Website, support, privacy policy and terms pages on HTTPS, with a privacy policy that matches what the plugin actually collects.
- For an MCP app: one stable public HTTPS server, OAuth or no authentication (API keys are not supported) and the domain verification token in place.
- Review materials: a test account with sample data that works without MFA or email codes, five positive and three negative test cases you have run, a walkthrough video and release notes.
- Country availability set to the markets you can support.
Display name and short description up to 30 characters, long description up to 4,000; website, support, privacy and terms URLs; square icon of at least 48 px; domain verification; one remote HTTPS MCP server with OAuth or no authentication; reviewer test account without MFA; five positive and three negative test cases; walkthrough video; country codes; publish after approval.
OpenAI Developers: Submit pluginsNo generic names or MCP/Plugin suffixes; descriptions must be accurate and may not compare, disparage, make unverifiable claims or promote pricing, subscriptions, trials or discounts.
OpenAI Developers: plugin guidelinesRules that block approval or monetization
Some growth habits from web apps break OpenAI's plugin guidelines. Check your plan against this list before you build the launch around it.
- Submitting a trial or demo plugin, or one that only repeats what ChatGPT can already do.
- Copying another product, implying OpenAI endorsement, or adding fields meant to steer the model toward your plugin or away from others.
- Selling digital goods, subscriptions, tokens or credits inside the plugin; only physical goods may be sold.
- Showing plans, starting subscriptions or promoting upgrades inside ChatGPT; you may say a feature needs another plan and link to an informational page, not to checkout.
- Making the ChatGPT version worse than your own product or adding a ChatGPT surcharge.
- Collecting data "just in case", or collecting payment card data, health information, government IDs or passwords.
- Leaving out the read-only, destructive and open-world annotations on each tool.
Plugins must add functionality beyond built-in capabilities; no trial or demo plugins, copycats or model-steering fields; only physical goods sold; no subscription promotion or checkout links; parity with the external version; data minimization and restricted data; explicit readOnlyHint, destructiveHint and openWorldHint annotations.
OpenAI Developers: plugin guidelinesChoose a channel by what the plugin does
A plugin is a shortcut to a job inside ChatGPT. The fastest first users are people who already do that job, often with your product.
| Dimension | Where the first users are | First move |
|---|---|---|
| Companion to your SaaS | Your existing users and trial sign-ups. | Tell them in the product, your docs and onboarding email which tasks they can now do from ChatGPT, with the exact prompt to try. |
| Team or B2B tool | Workspace admins and the team lead who owns the workflow. | Offer admins setup notes, a data-handling summary and a GitHub marketplace they can import; ask for one team pilot. |
| Developer tool | Codex users, GitHub, Hacker News and developer communities. | Show the plugin working in Codex and ChatGPT with one real task, and publish the setup steps where developers look. |
| Skills-only plugin for a profession | Communities, newsletters and accounts for that profession on Reddit, X or LinkedIn. | A short clip of the before and after for one task, posted where the community's rules allow it. |
| Former custom GPT | People who used the GPT. | Migrate it, then update every link to the GPT and explain what the plugin adds. |
A 30-day launch plan
- Days 1–3: write one sentence naming who the plugin is for and the task it finishes. Rewrite the name, short description and example prompts from that sentence.
- Days 4–7: time a new user from the listing to the first useful result. Remove steps until it takes a few minutes, and make the first answer show what else the plugin can do.
- Days 8–14: invite 20 people who already do the task, one at a time, mostly from your own users. Ask each to try one example prompt and tell you where they got stuck.
- Days 15–21: fix the step where most people stopped, and contact two workspace admins who could make the plugin available to a team.
- Days 22–30: make one public announcement in the channel that produced the best conversations, then compare connections, first successful tasks and repeat use with the first two weeks.
Copy the example prompts and first-run message
Example prompts now do the job screenshots used to do in the listing, so each one should describe a real task and lead to a result the plugin can deliver on the first try. Avoid prompts that only work with sample data.
The first-run message turns one successful task into a second one and sets expectations about changes to the user's account. Keep it free of pricing and upgrade prompts, which the guidelines do not allow inside ChatGPT.
Measure the funnel in your own product
The OpenAI pages reviewed for this guide do not describe directory analytics for developers, so measure the steps you can see: authorization starts on your OAuth page, completed connections, the first successful tool call and use again in the second week. For a skills-only plugin with no connection, track sign-ups or visits that arrive from the links inside the plugin's answers.
Illustrative example, not a benchmark: in one month 120 people start authorization and 84 complete it, a connection rate of 70% (84 ÷ 120). Of those, 42 finish a first task (50% of connections) and 21 use the plugin again in their second week (50% of first tasks), or 17.5% of everyone who started (21 ÷ 120). If most people stop at authorization, shorten sign-in and permissions; if they connect but never finish a task, fix the example prompts and the first answer before announcing anywhere else.
Where Soloop fits
Soloop does not build, submit or manage ChatGPT plugins, and it does not read plugin usage. It helps with the outreach around the launch. According to the Soloop manual, it can research where a group of people discusses a task and return sources, post on X after you connect it with OAuth, prepare Reddit posts for you to publish yourself, send cold email through Gmail (beta) and prepare Google Search Ads for your review and payment. Actions wait for your approval by default.
A bounded first request: My ChatGPT plugin [listing link] helps [audience] [do one task]. Find where [audience] discusses [task], with sources, suggest one channel for a 30-day test and draft five short messages asking people to try one example prompt and tell me where they get stuck. Do not post or send anything.
Sources, limitations and refresh
OpenAI's ChatGPT release notes, the Plugins in ChatGPT help article, the plugin submission page and plugin guidelines, and OpenAI's Stack Overflow case study were read on October 29, 2026. OpenAI does not publish how suggestions or directory placement are decided, and availability varies by plan, workspace, client and region. The channel table, launch plan, templates and funnel example are editorial guidance; the funnel figures are illustrative. Nothing here guarantees approval, placement, suggestions or users.
This guide will be reviewed by January 29, 2027, or sooner if OpenAI changes the directory, its listing rules or the custom GPT retirement schedule.
The Soloop manual describes research with sources, X publishing with OAuth, Reddit preparation with manual publication, beta cold email through Gmail and Google Search Ads after review and payment; approval comes first by default. It does not build or manage ChatGPT plugins.
Soloop product manual