How do ChatGPT plugins get discovered?.

How ChatGPT Plugins Get Discovered.

Discovery is the first problem any plugin has to solve. It doesn't matter how well your ChatGPT plugin works if no one knows it exists. But discovery inside ChatGPT works differently from traditional plugin stores, and that difference matters for how you build, describe, and position plugins.

There are three distinct ways a user can encounter a ChatGPT plugin:

• ChatGPT Plugin Directory.
• Direct mention.
• Indirect prompting.

Each one has a different trigger, a different audience, and different things you can do to influence it. This guide walks through each one so you know exactly where to focus.

The ChatGPT plugin Directory

The Plugin Directory is available at chatgpt.com/plugins and through the tools menu inside any ChatGPT conversation. It's organized into categories — Featured, Lifestyle, Productivity, and others — and it's where most new users will encounter your plugin for the first time.

How browsing and search work

Most users don't open the directory knowing which plugin they want. They search by what they're trying to do — "plan a trip," "track my expenses," "write a contract." That means your plugin's description needs to map to how users phrase tasks, not how you describe your product internally. Task-oriented and specific beats product-oriented and vague every time.

One thing worth knowing about is that the search algorithm is still fairly basic. It handles individual keywords well but struggles with compound queries. "Edit" and "designs" return results individually, but "edit designs" may not. Until search improves, simpler and more direct language in your description is the safer bet.

What featured status actually means

Plugins that meet a higher standard of design and functionality may be selected by OpenAI for Featured placement. Most developers assume this just means a better spot on the directory page, but it goes further. According to OpenAI's submission guidelines, Featured status also affects how often your plugin gets surfaced inside conversations, outside of the directory entirely.

Enhanced distribution is earned based on real-world utility and user satisfaction — there is no application process or paid promotion option for Featured placement. You build toward it by delivering a quality experience, not by applying for it.

Direct mention and re-engagement

Once a user connects your plugin, they can call it by name in any future conversation using an @ mention. When you start a message with the name of an available plugin — like "@Spotify, make a playlist for my party this Friday" — ChatGPT automatically surfaces the plugin in your chat and uses the relevant conversational context to help.

This path relies on the user already knowing your plugin exists. It's valuable for retention once users are connected, but it's not a discovery mechanism for new users.

That makes two things especially important.

• Your plugin's name. A name that's memorable, specific, and distinct makes direct mention more reliable. If users can't recall it, they won't invoke it. OpenAI recommends combining your brand with your best use case in the name, something like "BrandName Apartment Search" rather than just "BrandName."
• Your external promotion. Developers can use deep links on other platforms to send users directly to their plugin's page in the directory, reducing the friction between discovering your plugin and actually connecting with it. A link from your website, an email, or a social post can reach users who aren't actively looking.

Working through ChatGPT submission, we spot that by default, ChatGPT uses what it calls "Important actions" — it reads from plugins automatically, but asks before taking actions that may have a meaningful effect outside the interface, expose sensitive information, or be difficult to undo. Designing your plugin with this in mind makes the experience feel trustworthy, which directly influences whether users come back and invoke it again.

Indirect prompting

This is the most powerful discovery mechanism, and the one most specific to how AI platforms work. ChatGPT can suggest plugins when they're relevant to the conversation. If you're talking about buying a new home, ChatGPT can surface a real estate plugin as a suggestion so you can browse listings right inside the chat.

The most immediate signal is what the user is saying. The model reads the conversation, identifies the intent, and evaluates whether any connected plugin can help with that intent right now. OpenAI has described the signals driving this as conversational context, historical plugin usage patterns, and user preferences.

How the model reads your tool descriptions

This is the part developers have the most direct control over, and it matters more than most realize.

Every ChatGPT plugin is built on MCP tools, discrete capabilities the model can invoke. Each tool has a name and a description, and those descriptions are effectively the model's instruction manual for when and how to use your plugin.

As OpenAI states, tool descriptions must explain what the tool does in plain language, without promotional language, without comparisons to other plugins, and without instructions that encourage the model to select the tool beyond its actual purpose.

A tool called search_apartments with a description that reads "searches available apartments by location, price range, and move-in date" gives the model precise criteria for when to reach for it. A vague or promotional description makes that call harder. OpenAI has stated that incorrect or missing tool annotations are a common cause of plugin rejection.

In short, indirect prompting visibility is partly a writing problem. If you want to go deeper on how to structure tool metadata for this, we covered it in detail in this guide to setting up metadata for indirect prompting.

Quality and user satisfaction as a ranking signal

Tool definitions get your plugin considered. Performance determines how often it keeps getting surfaced.

OpenAI's submission guidelines state directly that plugins demonstrating strong real-world utility and high user satisfaction may be eligible for enhanced distribution opportunities, including proactive suggestions inside conversations. Plugins that deliver consistent value and that users return to over time will earn more prominent placement.

This means indirect prompting is a system that learns which plugins earn their recommendations by completing tasks reliably, and which ones don't. Plugins that deliver on the first interaction tend to get recommended again. Plugins that don't tend to get surfaced less. The platform surfaces plugins it trusts to help, and trust is built one completed task at a time.

How the three mechanisms compare

As you can see from all the above, each discovery path rewards something different. To help you sum it up, here's a quick breakdown.

What this means for how you build

None of the three mechanisms work in isolation. An plugin with vague metadata won't get selected by the model in indirect prompting situations because the model can't tell when it's relevant. An plugin with a great description but a poor experience will stop getting surfaced as performance signals accumulate. An plugin that works well but has no external presence will struggle with cold-start discovery until contextual suggestions eventually kick in.

The practical implication is that you need to think about all three as a single system. Your metadata shapes how the model understands you. Your name shapes whether users come back. Your plugin's performance shapes how often the model bets on you in the future.

The developers who figure this out early will have an advantage that compounds. The ecosystem is still new enough that there's real room to learn how the signals work and design around them, and that window won't stay open forever.