Choose a narrow purpose and a source set
A terminology guide is a short reference for words your team needs to use consistently. It might explain product names, internal abbreviations, project stages or terms that have a specific meaning in customer communications. Unlike a general dictionary, its job is to clarify how your team uses each term in a particular context.
Start with one audience and one type of work, such as onboarding documents or customer support replies. Then gather materials your team already treats as authoritative: an approved style guide, current product descriptions, published help content or maintained process documents. Note who owns each source and whether it is still in use. This gives you a way to resolve disagreements later rather than accepting whichever definition an AI system writes most confidently.
Decide what each entry should contain before generating anything. A compact format might include the preferred term, a plain-language definition, an example sentence, terms to avoid, the source document and a review status. You do not need every field for every entry. The important distinction is between wording taken from an approved source, a draft interpretation and a decision the team has explicitly made.
Use AI to extract candidates, not settle their meaning
Provide only documents you are permitted to use with the chosen tool, following your workplace’s data-handling rules. Ask the AI system to identify candidate terms and return the passage or document location associated with each one. Request that it mark any term whose meaning is not stated directly. If the tool cannot show where a proposed definition came from, treat that definition as a suggestion to investigate, not as an established fact.
A useful instruction is: “From these documents, list recurring terms that a new teammate might misunderstand. For each, provide the source location, a proposed plain-language definition and any conflicting uses. Write ‘definition not found’ when the documents do not define it.” This narrows the task to extraction and drafting. It also makes gaps visible instead of encouraging the system to fill them with plausible-sounding language.
Review the candidate list for clutter. Common words do not need entries unless your workplace uses them in an unusual way. Conversely, a term used only once may deserve an entry if misunderstanding it could cause a handoff or communication problem. Aim for a guide people will consult, not an exhaustive list of every noun in the documents.
Resolve conflicts with a named decision maker
AI can help you spot that one document says “account,” another says “customer” and a third uses both. It cannot determine which wording your team should adopt unless that choice is already documented. Record conflicting uses side by side and ask the person responsible for the relevant work to decide. Where different meanings are intentional, label the contexts rather than forcing one definition onto all teams.
Check each proposed entry against its source. Does the definition preserve important conditions? Does the example use the term correctly? Is an abbreviation expanded accurately? Pay special attention to words such as “approved,” “complete” and “urgent,” which may sound straightforward but carry workflow-specific consequences. Mark entries as reviewed only after a person with appropriate context has checked them.
Avoid turning uncertain wording into a rule. If a definition is still under discussion, label it as a draft and keep it out of instructions that others are expected to follow. A small set of confirmed entries is more useful than a polished-looking guide that quietly mixes decisions with guesses.
Make the guide usable and keep it maintained
Test the guide with a real task. Give a teammate an existing document to revise or a new document to draft, and ask where the guide helped and where it left a question unanswered. Look for entries that are hard to find, too vague to apply or overloaded with internal language. Short examples often make a definition more useful than a longer explanation.
Store the guide where people already look for writing or process guidance. Assign an owner to receive proposed changes and record when an entry was last reviewed. When a product name, process or approved source changes, revisit the affected entries rather than asking AI to regenerate the whole guide without context. Keep a brief record of decisions so the next reviewer can tell why a term was chosen.
The strongest role for AI here is to reduce the effort of finding candidate terms, comparing usage and drafting clear wording. People still need to decide what the organization means. That division of work makes the guide easier to trust—and easier to correct when the work itself changes.
