Guide
How to teach your AI your company's terms
An AI uses your company's words when it reads a short glossary, rather than when you correct it chat by chat. For each concept people mix up, pick one preferred term (the word customers see in the product, if there is one), write a plain definition that says how it differs from the nearest term, list the words not to use and why, add how to write it, and name an owner who decides changes. Then put a short Use / Don't use list where the AI reads it every time: custom instructions, a project, or a skill the team shares. Start with the 15 to 30 words that cause real confusion, not a dictionary.
Short answer
- One term per concept, with a plain definition.
- Words to avoid, with the reason.
- An owner per term.
- A Use / Don't use list where the AI reads it.
A worked example
Illustration: a made-up business. A scheduling software company's documents use "account", "workspace", "client org" and "tenant" for overlapping things, and its AI drafts mix them. The glossary, in part:
| Use | Definition | Don't use | Owner |
|---|---|---|---|
| workspace | The space a customer's team works in, with its own members and settings. A customer can have several. | "tenant" (internal), "client org" (old name) | Product |
| account | The billing relationship with one customer. One account pays for one or more workspaces. | "workspace" when you mean billing | Finance |
| booking | One appointment a client makes. Not a "reservation", which is a held slot not yet confirmed. | "appointment" in the product UI | Product |
The list that goes into the AI's instructions:
Use: workspace (the team's space), account (billing), booking (a confirmed appointment).
Don't use: tenant, client org, reservation (unless the slot isn't confirmed), appointment in UI copy.
Write "workspace" in lower case; "Workspace settings" only as the button name.The support lead puts the list in the team's shared reply skill. The next week's drafts stop saying "tenant" to customers.
Build yours
- Collect the words people mix up from your docs, tickets and product.
- Group variants by concept; if two words don't always mean the same thing, they're two entries.
- Pick one term and write why the others are out.
- Define it plainly, saying how it differs from the nearest term.
- Name an owner and put the Use / Don't use list where every chat reads it.
Next step: the same words for the whole team
A glossary in one person's instructions helps one person. Getting your team to use AI the same way covers putting it where everyone's AI reads it. For the rest of the context an AI needs, see what company context to give an AI.
Skills for this
- AI practiceTeam glossary builderBuilds a team glossary from documents: one preferred term per concept, plain definitions, words to avoid, usage notes, owners, and a list for AI instructions.
- MarketingBrand voice guide and rewriteExtracts a voice guide from 3 to 5 sample texts (traits with do and don't pairs, a word list, mechanics) and rewrites a draft in it.
- AI practiceContext brief for an AI assistantWrites instructions for an AI assistant or agent: role, audience, can and can't, sources of truth, tools, format, examples, edge cases, and handoff.
Questions
How do I teach ChatGPT or Claude my company's terms?
Give it a short glossary: one preferred term per concept, a plain definition, the words not to use, and how to write each term. Put the Use / Don't use list in custom instructions, a project or a skill, so every chat reads it, rather than correcting the AI one chat at a time.
What goes in a company glossary for AI?
For each concept: the preferred term (the word customers see in the product, if there is one), a one- or two-sentence definition that says how it differs from the nearest term, the words to avoid and why, capitalisation and an example sentence, and the person who decides changes.
How long should the glossary be?
As long as the terms that cause confusion, not a dictionary. Start with the 15 to 30 words people mix up: product names, metrics, and anything with an old name. The Use / Don't use list in the AI's instructions can be a few lines.
AI skill
Have your AI build your team's glossary
Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. Team glossary builder builds the glossary from your documents: one preferred term per concept, plain definitions that say how each differs from its neighbour, words to avoid with the reason, usage notes, an owner per term, and a Use / Don't use list ready for AI instructions.
14 days free · No card needed
Sources
- GOV.UK: A to Z style guide — one entry per term, with the preferred form and usage (read 2026-11-04)
- Google developer documentation style guide: Word list — use and don't use, with the reason (read 2026-11-04)
- Noy and McGuinness: Ontology Development 101 — define what makes a concept different from its siblings (read 2026-11-04)