How to move a custom GPT's knowledge files into skills

Where each kind of file goes

What each destination holds

From OpenAI's file uploads FAQ (Wayback copy of 18 September 2026), the Claude Help Center and Gemini Apps Help, read 5 November 2026. AI Skill Pod is this site's product. Pod pricing: Free up to 3 members and 10 skills; Team $5 per seat a month from 3 seats.

Sort the files in five steps

1. List every knowledge file with what the GPT uses it for, from the original files you uploaded.
2. Pull out the rules. Read each file for the lines the GPT applies every time, and write them into SKILL.md as steps. Most style guides shrink to a few lines.
3. Keep references whole. Put templates and product sheets in the skill's folder and name each one in the step that needs it, such as "Fill in template.xlsx".
4. Move changing documents to one place. A price list or rate table goes in a shared drive, linked from the skill with the date it was checked.
5. Test with five real questions the GPT used to answer, and check each answer against the file it came from.

A worked example: one GPT, seven files

Illustration: a made-up business. An insurance broker's "Policy explainer" GPT has seven knowledge files: four product sheets, a claims FAQ, a monthly rate table and a brand guide. The office manager sorts them:

1. The claims FAQ becomes six steps in SKILL.md: the questions clients ask most, with the answer and when to send them to a person.
2. The four product sheets go in the skill's folder, each named in the step that explains that product.
3. The rate table stays in the shared drive, where the finance lead updates it each month; the skill links to it and says to quote rates only from there.
4. The brand guide becomes its own brand-voice skill, since the broker's other skills need it too.

The product sheets also sit in the shared drive, linked from the skill, so teammates who get it from the team's pod can open them too. The rate table is never copied, so a new month's rates reach every answer the day they're updated.

Next step: keep the team on one copy after 11 December

Once the files are sorted, the people who used the GPT by link need the new skill. Keep your custom GPT shared after it retires covers telling your team and clients and keeping one version.

Questions

How many knowledge files can a custom GPT have?

Up to 10 files for the lifetime of the GPT, each up to 512 MB, with text and document files capped at 2M tokens per file, according to OpenAI's file uploads FAQ (its copy of 18 September 2026; older copies said 20). Spreadsheets aren't held to the token cap but are limited to about 50 MB. Check the limit your GPT editor shows as well.

What happens to a GPT's knowledge files on 11 December?

Custom GPTs stop running then. OpenAI's migration FAQ says a migrated GPT's instructions become a skill in a ChatGPT plugin and its knowledge files are copied into the plugin's reference files. That keeps them in ChatGPT only; to use them in Claude or another AI, keep your original files and sort them into the skill, its folder or a shared drive before the date.

Should knowledge files go in a skill or a Claude Project?

Into a skill if the files support one task, such as the template a report follows. Into a Project if people chat with the documents themselves, such as a handbook. Large or changing documents, like a price list, are better left in one shared place and linked, so nobody works from an old copy.