Translate with AI assistant using your Terminology, Translation Memory, and Style Guide

You can translate content directly from an AI assistant such as Claude or ChatGPT and still follow your company’s approved terminology, Translation Memory, and style guide. Connect the assistant to TextUnited through MCP, ask it to look up your language assets, and let TextUnited’s translation tool produce the result. Then check the output against those same assets.

Use this practical workflow to test a short piece of your own content with your company’s language assets. You will also learn how to check what the translation result actually proves.

What you need to know before you start

Three company language assets work together in this workflow. Each one controls a different part of the translation:

• Terminology: A terminology database stores your approved words and phrases. In TextUnited, each entry is a concept that holds equivalent terms in several languages, with definitions, examples, notes, and a status such as Preferred, Accepted, or Rejected.
• Translation Memory: A Translation Memory (TM) stores sentences your company has translated before. TextUnited can reuse a relevant saved translation when a match is available, helping repeated content stay consistent.

For more on how these assets reduce repeated work, learn how Translation Memory and terminology reduce rework.

• Style guide: A style guide is a set of rules for tone, formality, formatting, and wording. In TextUnited, a style guide can have variants for a specific language, domain, or both. Only a Published style guide is applied during translation. Draft guides are not applied.

MCP (Model Context Protocol) is an open standard that lets AI assistants use tools in other software. The TextUnited MCP server gives your assistant access to your terminology, style guides, and text translation, so it works with your real company assets instead of guessing.

For a closer look at the available tools, see what the TextUnited MCP server can do.

You also need a working MCP connection between your AI assistant and TextUnited. If you have not set it up yet, follow our MCP connection tutorial. Learn how to configure the connection, add your access token, and check that it works.

Step 1: Which content should you use for your first test?

Use a short product update or support instruction of around three to six sentences. A short text makes it easy to see whether each company asset was applied correctly. Pick content that includes all three of these elements:

• A company-specific term: a product name, feature name, or word your company always translates in a fixed way.
• A sentence already in your Translation Memory: a line you have translated before, such as a standard support sign-off.
• New text that needs your company’s tone: a sentence that has never been translated, so the style guide has to do the work.

Here is an example source text you can adapt. Replace the bracketed parts with your own content:

“[Feature name] is now available in your [product name] account. To turn it on, go to Settings and select [option]. The change applies to all team members right away. If you have any questions, contact our support team.

In this example, [Feature name] and [product name] are the company-specific terms. The last sentence is a typical candidate for a Translation Memory match. The middle sentences are new text that should follow your style guide.

What a successful translation must preserve:

• The approved target-language term for every company-specific word, never a rejected alternative.
• The saved wording for any sentence already stored in Translation Memory.
• The tone, formality, and formatting rules from the applicable style guide.
• The original meaning, plus any placeholders, product names, and UI labels that should stay unchanged.

Step 2: How do you check your terminology in TextUnited?

Before you translate, confirm that the terms in your text exist in TextUnited and have the right status. Open your terminology in TextUnited and find the relevant concept. Check four things: the concept itself, the source and target terms, each term’s status (Preferred, Accepted, or Rejected), and the domain the concept belongs to.

A domain is a subject area, such as a product line or industry, that groups related terminology. The domain matters because the same source word can have different approved translations in different domains.

Read how concepts keep your terminology consistent for more on organizing terms by meaning and domain.

You can also ask your AI assistant to run this check for you. The prompt below only reads your terminology and changes nothing.

Suggested prompt:

“Using TextUnited MCP, look up the terminology relevant to this text for translation from [source language] into [target language]. Show the preferred terms, rejected alternatives, definitions, and domains where available. Flag anything missing or ambiguous. Do not change any records. [Paste source text]

If a term is missing or has no approved translation in your target language, add or fix it in TextUnited before you translate. That way, the fix benefits every future translation, not just this one.

Step 3: How do you check your style guide in TextUnited?

Confirm that the right style guide exists, is published, and has a variant that matches your target language and content type. Open the style guide in TextUnited and review its rules, its publication status, and its language or domain variants. Only a Published guide is applied during translation, so publish a Draft guide first if you want to use it.

A style-guide variant is a version of the guide scoped to a specific language, a specific domain, or both. For example, a German variant for marketing content can have different formality rules from a German variant for support content.

Suggested prompt:

“Using TextUnited MCP, find a style guide suitable for [target language] and [domain/content type]. Read the guide and relevant variant. Summarize the applicable rules and check whether it is published. Flag any language or domain mismatch. Do not create, update, or publish anything.

The last line of the prompt matters. TextUnited MCP can also create and edit style guides, so an explicit instruction keeps this step read-only.

Step 4: How do you translate from your AI assistant with all three assets?

Send one request that brings terminology, Translation Memory, and your style guide together. The assistant looks up your approved terms, then calls the TextUnited translation tool with your target language, domain, published style guide, and the relevant term pairs. TextUnited translates the text and reuses your company Translation Memory where it can.

The TextUnited MCP translation tool works on text, one piece at a time. It does not translate whole files or projects. Besides the required target language, it accepts:

• Source language: optional, but recommended for short or mixed-language text.
• Domain: the subject area your terminology belongs to.
• Style guide: a published guide to apply.
• Terminology pairs: source terms with their preferred target translations.

Suggested prompt:

“Use the TextUnited MCP translation tool to translate the text below from [source language] into [target language]. Use the [domain name] domain and the applicable variant of the published [style guide name] guide. Look up the relevant company terminology and supply the approved source-target pairs to the translation tool. Use company Translation Memory where available through the service. If any required asset is missing or ambiguous, tell me before translating. Return the translation produced by TextUnited without rewriting it afterward. [Paste source text]

The last instruction is important. AI assistants often polish text on their own, and a well-meant rewrite can replace an approved term or break a Translation Memory match. Asking for the original TextUnited output keeps the result faithful to your company assets.

Step 5: How do you check what the translation actually demonstrates?

Review the translated text against each asset separately. A translation that reads well is not automatically a translation that followed your company rules. Annotate the result with three checks:

• Terminology: Does the approved target term appear for every company-specific word? Is any rejected alternative used?
• Translation Memory: Does the repeated sentence match the saved translation? Keep in mind that matching wording alone does not prove Translation Memory was used. The sentence may simply have been translated the same way again.
• Style guide: Does the new text follow the selected rules for tone, formality, and formatting?

You can ask the assistant to run this review, as long as it reports evidence and does not overclaim.

Suggested prompt:

“Review this translation against the approved terminology and applicable style-guide rules retrieved from TextUnited. List specific issues with supporting evidence. Do not rewrite the translation yet. Separate confirmed findings from anything you cannot verify. Do not claim a Translation Memory match was used unless the tool response provides that evidence.

Step 6: How do you fix recurring translation problems at the source?

When the same problem keeps coming back, fix the company asset, not just the translation. Use this checklist to find where each problem starts:

• Wrong term: inspect the concept, the term status, the domain, and the terminology pairs that were supplied to the translation tool.
• Unexpected repeated wording: inspect the saved translation in Translation Memory and the context it was saved in.
• Wrong tone: inspect the style guide’s publication status and whether the right language or domain variant was selected.
• Assistant changed the result: ask for the original TextUnited output and compare it with what the assistant showed you.

Which corrections belong in your company assets?

Fix the asset in TextUnited when the problem would affect future translations too. Typical examples are a missing concept, a wrong term status, a term filed under the wrong domain, an outdated saved translation in Translation Memory, or a style rule that no longer reflects your brand. One fix in TextUnited improves every translation that follows.

Which corrections belong only in the current translation?

Edit only the current text when the issue is specific to this piece of content. Typical examples are a sentence that needs different wording because of its context, a one-off campaign phrase, or a layout limit such as a maximum character length.