How to Review AI Menu Translations: Source Text, Terminology, Pricing and Change Records

Before publishing an AI menu translation, check that two things are in order: the dish facts match what the kitchen confirms, and the target-language wording has not changed what customers understand when ordering. A translation being generated, grammatically correct or viewable on the page does not replace either of these checks.

Eurostat reported on 2025-12-11 that, in 2025, 20.0% of the surveyed enterprises in the EU with at least 10 employees or self-employed people used AI, compared with 13.5% in 2024. The survey also covered only specified industries, so its findings do not represent every industry or micro-restaurants.

The survey measures enterprise technology adoption; it did not assess the quality of restaurant translations. Whether a tool suits a small restaurant depends on the source menu information, the availability of reviewers and the capacity to maintain the content. Increased adoption is not proof of translation accuracy or business results.

Start with kitchen-approved source text

Document the actual ingredients, preparation, portion sizes and options, not just the dish names. You can keep a distinctive name and explain what your restaurant actually serves. Forcing different dishes into a familiar foreign dish name may give customers a different expectation of what they will receive. Ask the kitchen about anything unclear in the source text; do not let the translation decide what the original means.

• Measurements and specifications: clarify what a single item, half portion, full portion and sharing size mean.
• Included items and extras: note whether sides, drinks and add-ons come by default or cost extra.
• Availability and substitutions: specify what customers can choose directly, what requires asking the kitchen first and what is unavailable that day.
• Factual basis: record who confirms recipes and supplier information, and which actual conditions to check when answering dietary questions.

Keep the original meaning and approved wording in the terminology record

• Content covered: enter the dish or category and the target language actually used.
• Original meaning and evidence: retain the confirmed source text and information about ingredients, preparation or specifications.
• Identified ambiguities: specify words that could be misread to mean something is complimentary, a main dish, a full portion or always available.
• Approved wording: record the wording adopted after review and who approved it, so it can be reused next time.
• Scope: note which changes to recipes, portions or combinations would mean this record needs to be reviewed again.

Reviewers need to understand both the target language and the dishes. If different people cover these areas, they can review them separately and consolidate their findings. One person does not have to do everything, and someone checking grammar alone should not decide on the kitchen’s behalf whether a food is suitable for a particular diet.

Check each meaning that could affect a customer’s choice

Prioritize negations, optional versus required items, and what is included by default versus charged extra. “Available as an option” must not be translated as “included by default,” and “please ask” must not become “always available.” No matter how definitive the wording sounds in another language, it cannot create a safety assurance for customers that the source text does not support.

Have a person review all text generated by any AI service, checking dish facts and language meaning separately. Keep a way for customers to ask staff about the published menu. The restaurant remains responsible for confirming the facts; do not treat platform-generated content as proof that every description has been verified.

After reviewing the text, check the page customers actually see

On a phone, review the categories, dish details, specifications and add-ons in each relevant language. Check whether long dish names obscure prices, whether key information appears before customers make a selection, and whether the source text is still understandable when a translation is missing. A complete review record does not mean the page’s meaning and reading order are clear.

Reading the menu and submitting an order are separate steps. Before enabling QR-code ordering, check in a separate environment that does not disrupt service how order submission connects to the restaurant receiving the order and the kitchen preparing it. If someone needs help during service, first check whether an order has already been placed, then help resolve the issue. Do not place a duplicate order to address a language misunderstanding.

When the source menu changes, identify which languages need another review

• Scope of changes: list the affected dishes, recipes, specifications, prices or combinations.
• Affected languages: check whether the approved wording still applies and send it to the relevant reviewer. Do not carry over the conclusion from an old version.
• Published content: correct visible information as soon as an error is found, then retain the original meaning, the differences and the approved wording.
• Handover responsibility: when maintenance changes hands, pass on the source text, terminology and review records. Keep unverified items separate from content that has been approved.

Sort subsequent questions into categories such as missing information in the source, terminology errors and page wording. Maintain a small number of languages for which you have review capacity, then expand based on actual use. Do not count fewer questions about a dish that has been discontinued as evidence of improved translation.