Why your Claude or ChatGPT skill isn't triggering, and how to fix it

Claude and ChatGPT decide whether to use a skill by reading its name and description, not its instructions. If the description doesn't match how people actually ask, or overlaps with another skill, the AI skips it. Fix the description first: say what the skill does, when to use it, and when not to.

How the AI chooses a skill

When someone starts a chat, the AI only sees a short summary of each installed skill: its name and its description. It reads the full instructions only after it has decided a skill fits. So a skill with brilliant instructions and a weak description never gets opened.

That's also why typing the skill's name works but normal requests don't. Naming it removes the guesswork. Your team won't do that. They'll type "reply to this" and hit enter.

The six usual reasons a skill doesn't fire

1. The description says what the skill is, not when to use it

"Email templates for client communication" describes the contents. The AI needs to know which requests it's for.

Too vague

Client email templates and guidelines.

Gets picked

Use when someone asks to write, reply to or rewrite an email to a client, including follow-ups and schedule changes. Not for internal messages or sales outreach.

2. Two skills overlap

If "Client email" and "Customer reply" both cover replying to a customer, the AI has to guess, and it often picks neither. Merge them, or give each a clear boundary that names the other.

3. It's too broad or too narrow

"Use for all writing" competes with everything. "Use for replies to Dana at Northwind about launch dates" almost never matches. Aim for one real job your team does every week.

4. The name doesn't help

"Skill v2 final" gives the AI nothing. Use a short name that matches the job, like "Client email reply".

5. It isn't turned on for that person

In Claude, organization skills need skills and code execution turned on for the organization, and each plugin has an availability setting (Required, Installed by default, Available or Not available). In ChatGPT, admins choose whether a plugin is Available, Installed or Disabled. A skill that's only "Available" won't run until someone installs it.

6. The model changed

A description that worked last month can stop working after a model update. If a reliable skill suddenly goes quiet, test it again before rewriting the instructions.

How to write a description that gets picked

Start with "Use when" and name the job in the words your team uses.
List three or four ways people actually ask for it.
Add a "Not for" line naming the closest neighboring jobs.
Keep it to two or three sentences. Put the detail in the instructions.

How to test it

Write ten requests the way a colleague would type them, without the skill's name.
Add three requests that are close but shouldn't trigger it.
Try each one in a fresh chat, so earlier messages don't help.
Count how often the right skill fired, and how often it fired when it shouldn't have.
Change the description, then run the same requests again.

Do this in each tool your team uses. A description that works in Claude doesn't always work the same way in ChatGPT.