An AI-generated translation can be a useful draft. It can also look unusually convincing while changing a condition, missing a qualifier or choosing the wrong product term. Post-editing is the process of evaluating and revising that draft against the source and the project requirements. For Brazilian Portuguese, the review should consider meaning, language variety, voice and where the content will actually appear.
Define the quality target for AI translation post-editing
Agree on the intended use and the level of review. An internal reference document and a public landing page may have different requirements, but neither should be approved without a shared definition of acceptable quality. State what the editor must check, which source is authoritative and who resolves product questions.
Identify the target locale, audience, brand voice and important terminology. Also confirm whether the source text itself is final. If the source keeps changing during review, track versions so the editor does not approve a sentence that no longer matches the material being published.
Compare meaning before polishing the prose
Read the source and target together. Look for missing ideas, added explanations, reversed relationships and altered conditions. A polished target sentence is not enough if it quietly changes what the product does or when an instruction applies.
Pay particular attention to negatives, exceptions and small qualifying words. A change from may to will can turn a possibility into a promise. A missing only can expand an instruction beyond its intended scope. When the source is unclear, raise a question instead of silently inventing a more specific claim.
Check names, numbers and product facts
Do not assume that a model will keep every repeated detail consistent. Compare important values with the approved source. If a numerical format needs localization, preserve the underlying value and make the presentation change explicit. The editor should not invent a local offer to make the translated copy sound more natural.
- Brand names, feature names and terms that must remain unchanged.
- Quantities, dates, measurements and currency amounts.
- Conditions such as eligibility, availability, limits and exclusions.
- References to a screen, button, plan or document.
- Contact information, URLs and other functional details.
- Variables, placeholders and formatting that the product expects to preserve.
Review terminology as a system
A sentence-by-sentence pass can miss a terminology problem spread across the document. Search for the main product terms and check whether the same concept receives different translations without a reason. Confirm that one Portuguese term has not also been used for two different features.
Use an approved glossary where one exists. If it does not, build a small decision list while reviewing. Record the chosen term, its meaning and any context where another translation is appropriate. Consistency should support understanding; it should not force an unsuitable word into every sentence.
Make the language appropriate for Brazil
Review vocabulary, sentence construction and tone for the intended Brazilian audience. An output can be grammatical but mix varieties or retain an English sentence pattern that makes the message harder to follow. Read headings and calls to action aloud to check whether the intended action remains clear.
Avoid replacing one generic style with another. Business copy does not need slang to sound local. Preserve the brand’s appropriate level of formality and make the explanation direct. Where a phrase needs substantial rewriting, retain the source meaning while choosing a natural way to communicate it.
Read interface text where it will be used
Short strings need context. A translation of plan, post or view can change depending on the screen and whether the word is a noun, verb or product label. Use screenshots, a staging site or string descriptions to understand the action.
Test buttons, line breaks, placeholders and error messages after implementation. Preserve tokens and variables according to the system’s requirements. If space is limited, propose concise wording that still communicates the action; do not remove a necessary instruction simply to fit the draft layout.
Apply an additional SEO pass when the page depends on search
For an SEO landing page or article, confirm that the topic and target expression came from local research. Post-editing should not preserve an awkward keyword merely because it was inserted into the AI draft. A natural rewrite may still need a research decision before publication.
Google’s guidance on generative AI content emphasizes accuracy, relevance and added value, and warns against producing many pages without helping users. The appropriate response is a substantive review of the content, including its titles and other metadata. Making AI output sound human is not a substitute for checking whether it is correct and useful.
Deliver decisions as well as corrected text
Deliver the revised version, terminology decisions and unresolved questions. Separate confirmed corrections from changes that need product-owner approval.