This article is sponsored by Acclaro and was written, edited, and published in alignment with our Emerj sponsored content guidelines. Learn more about our thought leadership and content creation services on our Emerj Media Services page.

AI tools have made it faster and cheaper to translate marketing content, but an accurate translation does not guarantee that a campaign engages buyers or converts in each market.

Language remains important to the online customer experience. The European Commission reported in April 2025 that a study across six European countries found 44% of web users considered it especially important for e-commerce websites to be available in their preferred language.

The Oxford Internet Institute found in 2025 that stronger multilingual skills do not reliably help AI models reflect local cultural values. In April 2025, 66% of US adults were very or extremely concerned about people getting inaccurate information from AI. Pew reported in September that 76% considered it very or extremely important to distinguish AI-generated content from human work, while 53% had little or no confidence they could.

For global marketing leaders, the measure extends beyond translation accuracy to whether localized content delivers results in each market.

Emerj Editor Yolandi de Weerdt hosted a conversation with Andrew Thomas, Head of Marketing at Acclaro, about why localized content underperforms as AI takes on more translation.

This article examines four practical strategies for improving the performance of AI-translated content across markets:

• Use market performance data to guide content revisions: Connect marketing results with localization workflows to identify underperforming content and prioritize improvements.
• Test market interest before expanding local operations: Publish AI-translated content to gather early engagement signals that can inform local investment.
• Preserve brand voice across AI translation tools: Build your style guide into the AI tools employees use.
• Direct human review where it matters most: Use automated quality checks and market performance data to flag content that needs linguistic or cultural review.

Listen to the full episode below:

Episode: How to Make AI-Translated Content Convert in Every Market – with Andrew Thomas of Acclaro

Guest: Andrew Thomas, Head of Marketing at Acclaro

Expertise: Global Content Strategy, Localization and Translation Technology, B2B Product Marketing, AI-Enabled Content Operations

Brief Recognition: Andrew Thomas leads marketing at Acclaro. He previously served as VP of Marketing, Linguistic AI at RWS, leading the Trados translation software suite, following earlier roles at SDL. He has more than 25 years of experience in the language industry.

Use Market Performance Data to Guide Content Revisions

Andrew Thomas’s starting premise is that content properly adapted for each market should perform about as well in every market. If a landing page converts in Germany but stays flat in Brazil, leaders should not assume Brazil is simply a weaker market.

If the product or service does not fit a market, better content will not fix the gap. All other things being equal, however, he argues that a lasting gap between markets signals content that has not yet been adapted well enough for local buyers.

Thomas says few organizations can say whether a localized page, such as a Japanese landing page, is driving the engagement or conversions it was built for. In his account, the marketing operations team sees performance data, while a separate team or vendor manages translation, and the two rarely share information.

Thomas describes the fix as a feedback loop:

• Marketing operations shares results for each localized asset with whoever owns localization, whether an internal team or an outside vendor.
• The localization team reviews those results alongside the translation data for the same content.
• The team revises underperforming versions in market and measures them again against the same goal.

He compares the approach to search engine optimization (SEO), where teams keep adjusting a page until it performs. Applied to each language version, the same discipline makes localization an ongoing performance process rather than a one-time production step.

Connecting marketing and localization data changes how teams manage content after publication:

• Before: Content is translated, approved for accuracy, published, and left alone unless someone reports an error.
• After: Each market’s version is tracked against its intended goal and revised when it falls short.

When language versions of a campaign underperform without investigation, that investment may deliver less value than intended. Shared data shows leaders which markets are falling short and which assets to revise first.

When one market lags, first ask whether the offer fits that market. If it does, investigate whether adapting the content could improve results rather than accepting the gap as inevitable.

Test Market Interest Before Expanding Local Operations

Thomas also argues that cheaper, faster translation should change the order in which companies enter new markets.

In the traditional approach Thomas describes, a company sees potential in a new country and hires a local salesperson to test interest. One of that person’s first requests to headquarters is for content, and the request reflects the salesperson’s view of what is needed, not necessarily what local customers want.

Once the material is in use, the salesperson often finds that it is the wrong content or that it does not perform, and new requests follow. Thomas notes that companies can spend many cycles on this back-and-forth.

His recommendation is to reverse the order. He says to test localized content before hiring sales or support staff. His proposed market-entry process uses early engagement to guide investment decisions:

• Publish AI-translated content in the target market.
• Measure engagement as an early signal of demand.
• Invest where interest appears by hiring local staff and bringing in human reviewers to make the content fit for a full launch.
• Hold back investment where interest does not appear, having spent far less than a local hire would cost.

A local hire commits salary, recruiting, and management time before demand is proven. A content test commits a relatively small translation budget and produces evidence that can guide the larger investment decision.

Thomas recommends using straight AI translation to gauge interest, then bringing people into the process once the company decides to invest in a market. Their role is to refine the content and make it fit for purpose.

Leaders should weigh one tradeoff: unreviewed content becomes the company’s first impression in that market. The approach suits demand testing with lower-stakes material, not flagship campaign launches.

As a rule of thumb, use engagement with localized content as an early signal of market interest. Let people, both local teams and culturally fluent reviewers, decide how to win that market.

Preserve Brand Voice Across AI Translation Tools

General-purpose AI assistants such as ChatGPT, Claude, and Gemini can all translate, so anyone in the company can quickly produce content in almost any business language.

Marketers already know the downside in their own language: employees create marketing content without the marketing team, so brand guidelines and preferred terminology never enter the process. Thomas says the same pattern now affects translation and calls it shadow localization, where teams, including marketers, translate content with AI tools because the official process feels slow or bureaucratic.

A Cybernews survey of more than 1,000 US employees, reported by the Journal of Accountancy in November 2025, found that 59% use unapproved AI tools to get their work done.

In Thomas’s account, the cost is a loss of quality control and governance, which shows up as weaker content performance. He also explains why AI translation without company-specific instructions produces bland output:

“If you don’t give it your specific company’s brand terminology in those languages, what’s going to happen is the content that it produces is going to be reduced to whatever the average is. The way that LLMs work, they reproduce the most obvious patterns, which means by default you’re always going to get a generic response.”

— Andrew Thomas, Head of Marketing at Acclaro

In practice, a company that sounds distinct in English can sound like every competitor in Japanese or German.

Rather than banning these tools, Thomas recommends making the approved route the easiest one. Companies can build custom assistants inside the AI tools employees already use and connect them to the localization team’s systems, so every translation request draws on approved terminology and past translations by default.

He says one of the first steps Acclaro takes with a new client is converting the existing brand style guide into instructions that AI applies to every translation. Enforcement should vary by content type. Marketing content needs strict adherence, while help content and product documentation follow brand rules more loosely, because their purpose is different and the teams that own them may have rules of their own.

Leaders can embed brand guidance into AI translation workflows through four steps:

• Translate the style guide into the target languages. Brand rules written only in the source language will not shape output in other languages.
• Maintain approved terminology in each language, including product names and brand terms.
• Set enforcement levels by content type, strict for marketing and lighter for support and documentation.
• Connect employees’ AI tools to the localization team’s terminology and past translations.

Potential benefits include less rework, fewer brand inconsistencies reaching customers, and localized content that sounds like the company rather than the market average.

Direct Human Review Where it Matters Most

Thomas attributes AI’s difficulty with cultural nuance to the gap between evolving cultures and the historical content models learn from. By their nature, he says, these models are “always going to be generating content for a culture from the past.”

According to Thomas, Google Translate tended to treat the word “mask” as a beauty product before 2020, because that was its most common use in the text the system had learned from. When “mask” suddenly came to mean protective medical equipment, the translations were wrong.

Thomas says a human reviewer would have caught the error, while involving language experts in the initial data setup could have prevented it. He recommends human involvement at both stages:

• At setup: Professional linguists and people who understand the target culture help build the terminology, instructions, and data the AI works from.
• At the final stage: Once AI has taken content 80% to 90% of the way, reviewers refine the pieces that need cultural nuance, most often marketing content.

The remaining question is how to decide which content needs final review without sending everything through it. Reviewing everything gives up the speed AI provides, and reviewing nothing leaves the company hoping regional results hold. Thomas starts from an assumption about failure:

“You have to make the assumption that at some point there will be a failure. If you do not plan for that failure, it’s going to eventually bite you. But if you do plan for it and can understand how to quickly resolve it, then you can basically let the AI run free.”

— Andrew Thomas, Head of Marketing at Acclaro

Thomas describes using AI to check AI: separate automated quality checks review each translation and estimate how likely it is that a person needs to step in. Translations flagged by these checks go to an internal reviewer or an outside partner. Performance data provides another reason for review when a market’s results lag.

The workflow shifts from reviewing everything or nothing to reviewing only what the checks or the results flag. The intended payoff is less time spent on review and a lower risk of a damaging error reaching customers.

Thomas explains that content quality should be judged by the return it generates, with expectations varying by content type, market, and audience. He calls this “quality of outcome.”

To assess content by its intended outcome and determine when human review is needed, leaders can follow this three-step process:

1. Define the outcome each content type must achieve in each market.
2. Let content owners set the quality thresholds that decide what can be published on AI alone.
3. Send content below the threshold or content with lagging results, to an in-country expert.