Every business has a private language. Product names, fare classes, account tiers, loyalty programs, industry shorthand. It's the vocabulary your customers use every day, and it appears in no dictionary.

Generic AI gets that language wrong in two ways.

Translation failures. Product names get translated when they should stay untouched, and domain terms come out as nonsense in the customer's language. For many enterprises, this is the single thing blocking a multilingual launch.

Comprehension failures. When a customer uses shorthand only the business understands, the agent misreads the intent, retrieves the wrong knowledge, and answers the wrong question. This one hurts every conversation, even in English-only deployments.

Until now, the workarounds were stuffing terminology into the company description or writing one custom instruction per term. Both break past about 20 terms. Enterprises have thousands.

A managed vocabulary, applied on every turn

Glossary is now generally available in Ada. It gives your AI agent a managed vocabulary of your terms, your translations per market, and your definitions. On every turn of every conversation, Ada retrieves the terms relevant to what the customer just said and applies them to how the agent understands the request, searches your knowledge, and responds — in the customer's language, in your words. It's built for thousands of terms.

Here's what that includes:

• Custom translations per language. Map what customers type, in any language, to the exact wording the agent should use in responses and knowledge lookups. One term can carry a different translation per market.
• Brand and product preservation. Mark the names that must never be translated. They survive intact in every conversation language.
• Business definitions. Attach a short definition to any term so the agent understands what it means in your business, not the dictionary.
• Availability rules. Scope terms by audience or market, so the same term can behave differently for different segments.
• Built for scale. Manage terms in the dashboard or in bulk via CSV. Typos, plurals, and word variants are matched automatically, so you upload base terms only.
• Visibility. The conversation view shows which glossary terms were applied on each turn, so your team can verify adherence and spot gaps.
• Voice and tooling. Terms can be flagged for Voice so transcription recognizes brand names it would otherwise mishear, and technical teams can manage the glossary from their own AI tooling through Ada's MCP tools.

What it feels like when it works

Travel. A traveler writes in Spanish: "Quiero cambiar mi vuelo, tengo tarifa Flex Plus." Before, the agent would translate "Flex Plus" into Spanish, miss the fare-rules article, and quote generic change fees. Now the fare name stays intact, the right policy comes back, and the answer arrives in Spanish.

Retail. A shopper in Danish asks about a product with an English name. Before, the name mutated in translation and the shopper couldn't find that product anywhere. Now the product name is preserved exactly, and the order lookup matches.

Financial services. A credit union's members are always "members," never "customers." And "Everyday Growth Account" is a specific savings product with a specific rate, not four generic words. Definitions make the agent answer precisely, and the conversation view gives your team an audit trail — which matters when wording carries compliance weight.

Mobility. An Italian rider writes "Ho un problema con la mia corsa." Before, "corsa" read as an amusement ride and the knowledge search came back empty. Now it maps to the company's canonical term, and the issue gets resolved in Italian.

Media. A subscriber says "I'm having trouble with cooking." A definition tells the agent that Cooking is the recipe subscription product, and it routes straight to the right account flows.

The teams who tested Glossary before GA were blocked on multilingual launches by exactly these problems, running real glossaries — some with 150+ terms across 20 languages. Not anymore.

Getting started

If your terminology lives in your company description or in a stack of custom instructions today, move it to Glossary: it's the supported, measurable home for your language, and the conversation view shows you it's working. And if translation failures have been holding back a language launch, this is the unblock.