ChatGPT Plugin Directory: Ranking After DevDay 2026

The ChatGPT plugin directory lists the plugins people can add to ChatGPT, and since DevDay on September 29, 2026, OpenAI says improved ranking and recommendations surface them in the directory and inside conversations. OpenAI has not published ranking criteria. Its plugin guidelines name the only stated signal for extra placement: real-world utility and user satisfaction.

That leaves a gap that guesswork will fill. This guide sorts OpenAI's own statements from inference, then turns its published guidelines into a listing checklist you can act on.

What did OpenAI announce for plugins at DevDay 2026?

Four plugin items in OpenAI's DevDay 2026 recap, one of which covers the directory: "Improved plugin creation, submission, and discovery". The table lists all four, with that item split into its three parts.

Announcement	What OpenAI says	Availability
Plugin Creator	Helps developers build a plugin	All plans
Redesigned submission flow	Gives clearer feedback on submissions	All plans
Improved ranking and recommendations	Help people find relevant plugins in the directory and in conversations	All plans
Plugin extensions	Sidebar placement, interactive panels and file viewers inside ChatGPT	All plans
MCP events	Support for the proposed MCP Events specification, so plugins can start automations	All plans
Sites can host plugins	Supported plugins inside the Sites teams build	Business, Enterprise, Healthcare, Edu

The same item adds that users choose which plugins to use and approve the access each one receives. TechCrunch's coverage of the plugin changes confirms the directory details and adds no ranking specifics. Its analysis of the app-store angle reports two submission changes: developers can track a review, see required fixes, request a human review, and update tools without resubmitting. It also puts ChatGPT at 1.2 billion weekly users, which is the audience the directory sits in front of.

What is known about plugin ranking, and what is not?

OpenAI has said that ranking improved and where recommendations appear. It has not said what drives them. The table separates the two, with the source for each known item.

Known	Source
Ranking and recommendations work in the directory and inside conversations	DevDay recap
ChatGPT suggests apps in the flow of a conversation when one could help with the task	TechCrunch, citing OpenAI
Plugins with "strong real-world utility and high user satisfaction" may be eligible for directory placement or proactive suggestions	Plugin guidelines
Plugins must not include fields that manipulate how the model selects other plugins or tools	Plugin guidelines
ChatGPT decides when to call a tool from the metadata you provide	Optimize metadata guide

Unknown	Why it matters
Which signals rank the directory, and their weights	You cannot target a factor nobody has named
How utility and satisfaction are measured	Installs, ratings, retention and repeat calls are all plausible; none is confirmed
What triggers an in-conversation suggestion	Metadata matching is likely to play a part, but OpenAI has not said so for suggestions
Whether paid placement exists in the directory	The guidelines bar plugins from serving ads; they say nothing about paid ranking

Be wary of anyone selling "ChatGPT plugin SEO" as a ranking formula. The guidelines go further than silence: they prohibit metadata written to steer the model's choice between plugins.

What can you infer from OpenAI's documentation?

Two things, both from the developer docs rather than from the announcement.

First, relevance runs through tool metadata. The metadata guide says well-written metadata increases recall on relevant prompts and reduces accidental activations. A recommendation system that suggests plugins for a task needs a similar relevance signal, so tool names, descriptions and parameter docs are the most likely input you control. This is an inference: OpenAI documents metadata for tool calls, not for directory ranking.

Second, placement follows use. The guidelines tie "enhanced distribution opportunities" to utility and satisfaction, which means a plugin that people install, use and keep using should have the better chance. You influence that through the product, not the listing.

Everything past those two points is speculation, including any claim about how much each factor weighs.

How do you make a plugin easy to recommend?

Write a listing that passes review on the first try, and metadata that triggers your tools on the right prompts and stays quiet on the wrong ones. Both come from OpenAI's published guidelines.

Listing element	What the guidelines require	What gets rejected
Plugin name	Tied to your brand	Generic dictionary words; "MCP", "MCP Server" or "Plugin" appended to the name
Description	"Clear, accurate, and straightforward"	Comparisons, disparaging alternatives, unverifiable claims, prices, trials or discounts
Tool names	Unique, plain-language	Promotional or comparative words such as "best" or "official"
Tool descriptions	Match the schema and actual behavior	Unclear or incomplete descriptions
Example prompts	Help users get started; screenshots no longer show in the directory	Not stated
Review access	A demo account with sample data	Sign-up steps or 2FA the reviewer cannot complete

Then tune the metadata against a test set, following the method in OpenAI's metadata guide:

1. Name tools by domain and action. For example calendar.create_event, so the name says what the tool touches and what it does.
2. Start each tool description with "Use this when…". Add when not to use it, such as "Do not use for reminders."
3. Document every parameter. Give examples, and list the allowed values for constrained inputs.
4. Set annotations to match behavior. Mark readOnlyHint on tools that only read data, destructiveHint: false on tools that do not delete or overwrite, and openWorldHint by whether the tool reaches the public internet.
5. Build a golden prompt set. Include direct prompts that name your product, indirect prompts that describe the outcome, and negative prompts where another tool should handle the request.
6. Measure precision and recall. Precision is the right tool running; recall is the tool running when it should. Fix false activations on negative prompts before chasing more recall.
7. Change one field at a time. Log each revision with a timestamp and its results, and replay the prompt set after launch, reviewing tool-call analytics weekly.

The guidelines also reject trial or demo plugins, plugins whose main purpose is advertising, and unofficial connectors to other companies' services. Commerce is limited to physical goods: plugins may not sell digital products, subscriptions or credits.

How do a brand's plugin and MCP server relate to its AI visibility?

They are two different kinds of presence. AI visibility is whether ChatGPT names and cites you when someone asks about your category. A plugin is whether ChatGPT can call your product to do something once the user has chosen it. OpenAI has not said that a plugin changes how ChatGPT's answers describe a brand, and you should not assume it does.

The two connect through the MCP server. OpenAI's plugin documentation describes a plugin as skills, an MCP server and optional UI. The MCP server holds the tools, and the same server can serve other clients, such as Codex, Claude Code or Cursor, through their own setup. One well-documented server can make your product callable in several assistants, with the ChatGPT directory as one route among them.

Citlyze's own MCP server shows the read-only pattern: its tools let Claude Code, Cursor, Codex and other MCP clients list your tracked prompts, compare visibility against competitors and review citations, without changing anything in your workspace.

Visibility still comes from the web. A buyer asking ChatGPT for the best tool in your category gets an answer built from sources ChatGPT trusts, as how ChatGPT picks sources explains, whether or not you have a plugin. The AI search optimization playbook covers that side surface by surface, and ChatGPT rank tracking covers how to measure it.

Should every brand build a ChatGPT plugin?

No. Build one if users would finish a real task with your product inside ChatGPT: book, search, configure, analyze, order. A plugin whose job is to put your logo in the directory is the kind of plugin the guidelines reject, since they turn away advertising vehicles and plugins submitted as trials or demos.

For most marketing teams, the better question is whether ChatGPT recommends the brand in answers. That needs measurement before it needs a plugin.

Where to start

If you have a product that fits, write the golden prompt set before the plugin: 20 direct, indirect and negative prompts, with the behavior you expect for each. It doubles as your review test plan. If you want to see how ChatGPT and other assistants describe your brand today, Citlyze's MCP server puts that data in Claude Code, Cursor or Codex, and Ask Citlyze answers the same questions inside the app. The pricing page shows which plans include API and MCP access.