Research · ChatGPT plugins

How ChatGPT plugins get discovered.

OpenAI documents that a published ChatGPT plugin is found by link or by name, and that suggestions and directory placement go only to plugins it selects. KailxLabs is measuring what that looks like in practice. So far we have one pilot: on 9 Oct 2026 the directory front page held 96 slots on 16 shelves, and a directory search for CRM returned 47 listings without HubSpot. The studies on suggestions inside conversations have not run yet, and this page shows no numbers for them.

First published 9 Oct 2026Sources checked 9 Oct 2026

Last updated 9 Oct 2026

Key numbers

What we counted so far.

One day, logged out, on the public directory. Exploratory. Every number below is in the data files.

96

slots on the directory front page, across 16 shelves, filled by 91 different listings.

KailxLabs snapshot · chatgpt.com/plugins, logged out · 9 Oct 2026

45 of 47

results for the search CRM had CRM in the listing name or short description.

KailxLabs pilot · 1 query, run twice, same results · 9 Oct 2026

0

results for "customer relationship management" and for "crm for small business".

KailxLabs pilot · 2 queries, 1 run each · 9 Oct 2026

Nine queries cannot show how directory search works. They show it can be read from outside, the same way twice, without an account.

Documented

What OpenAI says.

Read at the source on 9 Oct 2026. OpenAI's developer pages show no date, so the source date for them is the day we read them.

What OpenAI documents about plugin discovery, with source, source date and date checked
DocumentedSourceSource dateChecked
ChatGPT and Codex share one plugin directory.OpenAI, Developer docs, plugin conceptsNot dated9 Oct 2026
A published plugin is found by a direct link to its listing or by a search for its name.OpenAI, Developer docs, publish and reviewNot dated9 Oct 2026
Directory placement and proactive suggestions go only to plugins OpenAI selects, which it calls enhanced distribution. Developers cannot request it.OpenAI, Developer docs, publish and reviewNot dated9 Oct 2026
The stated criteria are strong utility in real use and high user satisfaction. No signals, weights or thresholds are published.OpenAI, Developer docs, publish and reviewNot dated9 Oct 2026
ChatGPT can suggest a plugin during a conversation, and what a user sees depends on the account and where ChatGPT is used.OpenAI, Help Center, Plugins in ChatGPTUpdated early Oct 20269 Oct 2026
For installed plugins, ChatGPT and Codex decide when to call a tool based on the metadata the developer provides.OpenAI, Developer docs, optimize metadataNot dated9 Oct 2026
Plugin text must not tell the model to prefer one plugin over others or interfere with fair discovery.OpenAI, Developer docs, plugin guidelinesNot dated9 Oct 2026
Plugins must not serve advertisements, and may currently conduct commerce only for physical goods.OpenAI, Developer docs, plugin guidelinesNot dated9 Oct 2026
On 29 Sep 2026 OpenAI said ranking and recommendations for plugins were improved in the directory and in conversations.OpenAI, DevDay 2026 recap29 Sep 20269 Oct 2026
Three kinds of claim

Known, measurable, unknown.

We keep these apart on every page about plugins.

Known

Documented by OpenAI

One shared directory. Found by link or name by default. Suggestions and shelves for selected plugins only. Tool selection for installed plugins rests on developer metadata. No ads inside plugins.

Measurable

Can be counted from outside

Directory shelves and directory search, with no account. Which tool runs once a plugin is installed. Whether any plugin is suggested on an account with nothing installed. ChatGPT against Codex.

Unknown

Not established by anyone

The recommendation method and its signals. What user satisfaction measures. How often suggestions appear. Whether memory, plan or region change them. How keywords are used. How many plugins exist.

Pilot

The directory search pilot.

Method: each query typed into the search box on chatgpt.com/plugins, logged out, on 9 Oct 2026. We recorded how many listings came back and their order.

Directory search pilot: query, kind of query, number of results and what was observed
QueryKindResultsObserved
HubSpotPlugin name2HubSpot listed first
GustoPlugin name2Both results are Gusto listings
CRMCategory word4745 have CRM in the name or short description. HubSpot is not among the 47. Run twice: same 47, same order.
customer relationship managementCategory phrase0No results
crm for small businessBuyer phrase0No results
payrollCategory word3Gusto twice, then PayCanvas
keyword researchCategory phrase4Semrush listed first
email marketingCategory phrase65 have the phrase in the name or short description
Insights to actionWords from one listing1HubSpot, whose short description starts with those words

Limits

  • Nine queries, one day, one location, logged out. Exploratory under the standard we use for client work.
  • A logged in search may return different results. Not tested.
  • Only the name and short description are visible in results, so we cannot say why a listing with no matching words appeared.
  • A match between query words and listing words is correlation. It does not show that changing the words changes the results.
  • Nothing here describes suggestions inside a conversation.

Data files

Both files are free to reuse with credit to KailxLabs and a link to this page.

Planned

Four studies, in order.

Questions, samples and counting rules were written down before any data was collected. No study shows a finding until its data file is ready.

Planned KailxLabs studies on ChatGPT plugin discovery, with status, surface, sample and what each needs
StudyStatusSurfaceSampleNeeds
1. What the plugin directory shows, and for which wordsPilot doneDirectory, logged out100 fixed queries and every front page shelf, daily for 14 daysNothing. Starts next.
2. Does ChatGPT suggest plugins nobody installed?PlannedChatGPT, logged in, nothing installed60 prompts across five kinds of request, 10 runs eachA ChatGPT account kept only for measurement
3. ChatGPT against Codex on the same taskPlannedChatGPT and Codex, reported apart30 tasks, 5 runs on eachStudy 2 first
4. Does changing listing text change discovery?PlannedOne plugin we controlA worked prompt set against an untouched control set, one field changed per roundA plugin we control. The only design here that can show cause.

Results will be reported as counts, such as 7 of 10 runs, per surface. We will not publish a rank for a plugin, and we will not say a listing change caused a recommendation unless study 4 supports it.

Answers

Plugin questions, answered.

Answer

What are ChatGPT plugins in 2026?

Plugins are packages of skills and connected apps that people install in ChatGPT and Codex from one shared directory.

Read the answer
Answer

Can ChatGPT recommend plugins automatically?

Yes, but only for plugins OpenAI selects. A developer cannot request it, and most published plugins are found by link or name.

Read the answer
Answer

How does ChatGPT decide which plugin to use?

For installed plugins, from the names and descriptions the developer wrote. For plugins not installed, OpenAI publishes no method.

Read the answer
Answer

Can you optimize a ChatGPT plugin for discovery?

Partly. You can test and tune when ChatGPT calls an installed plugin. You cannot request, buy or guarantee a recommendation.

Read the answer
Answer

Do plugin and MCP tool descriptions affect discovery?

For installed plugins, yes: OpenAI says tool metadata decides when a tool is called. For suggestions, nobody has shown it.

Read the answer
Answer

How do you measure ChatGPT plugin visibility?

In three layers kept apart: the directory, tool selection after install, and suggestions before install, each with repeated runs.

Read the answer
Answer

Does getting into the plugin directory guarantee traffic?

No. A published plugin is found by link or by name, and reaches the main directory pages only if OpenAI selects it.

Read the answer
Answer

Can a ChatGPT plugin pay for placement?

Not as of 9 Oct 2026. OpenAI documents no paid placement, developers cannot request promotion, and plugins cannot serve ads.

Read the answer
Answer

Do ChatGPT plugins work in Codex?

Yes. ChatGPT and Codex share one plugin directory, though the Codex IDE extension does not support plugins.

Read the answer
Answer

What changed for plugins at DevDay 2026?

Plugin extensions, a new creation and submission flow, improved ranking and recommendations, MCP Events and plugins inside Sites.

Read the answer
FAQ

About this research

Can I cite this page?

Yes. Cite it as: KailxLabs, "ChatGPT plugin discovery: what is known and what we measured", 9 Oct 2026, with a link to this page. Please include the date, because the platform changes and this page will too.

Is the pilot a finding?

No. It is nine search queries and one snapshot of the directory front page, on one day, logged out. It shows the directory can be measured from outside. It does not show how directory search works, and it says nothing about suggestions inside conversations.

Why are there no numbers for suggestions?

Because we have not counted them yet. Suggestions depend on the account, so they need a logged in account kept in a clean state, and many repeated runs. We will not publish a rate before the runs exist.

Do you build plugins?

No. KailxLabs measures and researches discovery. Building a plugin, its MCP server and its submission to OpenAI is engineering work for a product team or a development studio.

Will you study other AI apps?

Later. Claude, Gemini and Microsoft Copilot each have their own directory of connected tools. We are keeping the first studies on ChatGPT so the results are narrow enough to check.

Last updated 9 Oct 2026
  • 9 Oct 2026: hub published with what OpenAI documents, a directory pilot of nine queries and one shelf snapshot with both data files, and four planned studies.
Book a call

Talk to the founder.

20 minutes with Kailesk Khumar. You leave knowing where AI names you today, and what it would take to become the answer.

  • A live look. We ask five of your buyer questions on ChatGPT, Google and Perplexity while you watch.
  • The reasons. The three biggest reasons AI names a rival instead of you.
  • The plan. A clear next step and its price, whether or not you hire us.

Pick a time that suits you. The calendar invite arrives straight away. Prefer a link? Book on Cal.com.

KK Kailesk Khumar AI visibility strategy call
20 min Video call Your time zone
Prefer email? Get a free written read instead
What you sell
How did you find us?

No newsletter. A reply from a person within two working days.

Got it. Kailesk will reply within two working days.
Talk to the founderBook a call