Translation services are usually outsourced once the website is completed. It is normally not considered a necessary part of the website creation process from the early days. Once you use AI for website translation, the task becomes a continuous rather than a one-off task that will complement the regular website updates.
It’s not about avoiding translators; it’s about bypassing quality control. It’s all about taking the manual requirement, the exports, the email threads, and the waiting out of the way, so that real translation work can occur at a quickened pace, with human oversight to correct anything the model may get wrong.
Why teams are rethinking their translation process
Several years ago, it took several seconds to pull the web pages into a spreadsheet, submit to a translation company, and wait a week to two for a spreadsheet back. Several years ago, pulling web pages to a spreadsheet, submitting it to an agency, and waiting 1 to 2 weeks for a translation to come back was several seconds of pulling and a few days of waiting. The process also applies to a one-time project, but fails when a site changes frequently.
- The Products list is subject to frequent updates for the price and stock information.
- The blog content is on a weekly or even daily posting basis.
- Quick changes to legal/policy pages are required if there are changes in regulations.
Step 1: Recognize what AI translation is
A single AI translation model can extract the meaning and context of various utterances within a text and render new ones that retain the meaning and structure of the target text as a whole, rather than translating individual words verbatim. It’s a significant improvement over the older machine translation systems, which usually generated robotic, literal output that required extensive editing.
Platforms are further optimized with a learning effect within the new ones, so that the more a team uses the tool, the better the translations become. This is one of the reasons that localization for people on the border of your service is possible even today with the help of AI.
Step 2: Link the tool to your real website
The most significant time savers lie in linking translation directly to a website, rather than transferring from and back to files. It’s here that most teams first learn how to incorporate AI into website translation as a real-time, continuous process, rather than being a batch job.
An ideal platform already extracts and pulls content directly from a CMS, automatically translates it, and then pushes it back without affecting the content’s layout, links, or eye-catching design. This is essentially the model behind Smartcat’s usage of AI for its website translation tool, which, apart from the major difference of being a platform that connects to a website, translates content into over 280 languages and offers a live preview before content is published, does pretty much the same thing.
Step 3: Build a review step into the process
AI produces the initial translation in seconds, and it only takes a few seconds.
Ends with verification of accuracy by a Reviewer (in-house or hired Linguist) and verification of tone.
Step 4: Prioritise pages by impact, not just volume
They will help you prioritize which pages to act on based on their importance, not just volume.
Not all pages have to be handled equally. For pages with high traffic, checkout flows, and the most popular products, the AI-generated output provides an excellent foundation but still requires some human scrutiny; for less prominent pages like older blog posts, the AI-generated output is a great foundation and can be lightly reviewed.
Ways to make common mistakes
Teams that have trouble with localization using AI tend to have a couple of recurring problems:
- Not recognizing the variation of regional languages, such as Spanish vs. Latin America.
- Not updating translated documents when the source document changes, leading to misaligned versions over time.
The key to avoiding these problems is to choose the right platform and establish a review program from the beginning instead of relying on one you add once the game is finished.
Why this approach pays off for SEO
The search engines recognize each language version individually, and AI-based search results now often appear in language versions that best match the searcher’s language. It makes no sense to have a website that only attracts English-language visitors, even if the rest of the world is out of the equation. No matter where a website positions itself in the ranking arena, one needs to understand that it suffers significantly in terms of the number of visitors it can attract beyond English-language users.
Prompt language updates to the versions means that if a version is not updated within months of the source site, you might experience reduced visibility and/or indexing in various regions.
Surprisingly, a lot of teams are surprised by the long-term search performance being more significantly impacted by fast and frequent updates.
Conclusion
Leveraging AI tools for website translation isn’t about shortcutting the localization process; it’s about freeing up some of the slower elements of the process so that all the vital ones are treated more seriously, rather than being pushed aside. The best teams consider AI to be the first draft engine and review by humans as the quality gate, rather than the other way around. It’s common to first apply it for a single page or language to determine whether it can fit a given process before introducing it to a whole site. When this initial test has shown that there is real time saved in the workflow and there has not been any decrease in quality, it’s less risky than it would have seemed at first to scale this process to each market the brand serves.
Featured image credit
