Quality in Game Localization in the Age of AI
By Marina Ilari
Ask ten people in game localization what quality means, and you will get ten different answers. A linguist will talk about accuracy and style. A developer will talk about strings that fit and variables that work. A business leader will talk about ratings and revenue. A player will not talk about localization at all. Instead, they will tell you whether the game felt right.
That is the first lesson I take from years of asking this question: quality is not universal; it is contextual. What counts as "good" depends on purpose, audience, risk, and product. AI has not changed that. What it has changed is how much content we can produce, how fast, and with how much (or little) supervision. That forces us to be far more precise about what we mean by quality, and about who owns it.
Quality depends on what the content is for
Game localization varies widely in scope. A small mobile title may need only UI text in one language. A large release involves many languages and many content types: in-game text, voice-over scripts and recording, subtitles, culturalization and world-building support, bug fixing and LQA. Each of these has its own quality requirements.
Not all content carries the same weight, either. Tiering content by its impact on the player and on the brand tells us where quality is critical and how to measure it.
Ownership matters as much as tiering. Localization project managers own process and delivery. LQA leads own validation and feedback loops. Linguistic leads own consistency and decision-making. Client stakeholders own priorities and trade-offs. Each group has its own KPIs, so teams need a shared point of reference, or they end up working toward different goals.
From getting the words right to what the words do
Our industry has historically defined quality in linguistic terms: accuracy, terminology consistency, grammar and fluency, and adherence to the style guide. All of that still matters. The real question is why it matters, for the game, for the player, and for the business behind it.
Players experience the localization in the game. When quality is high, the player forgets they are playing a localized game. When it is not, immersion breaks instantly. Good localization drives engagement, keeps characters authentic, supports gameplay flow, reduces confusion, and preserves functionality.
It also drives business results. Quality improves reviews and ratings, increases retention, strengthens brand trust, reduces support costs, boosts conversion, and minimizes risk. Localization quality is a growth driver, not only a cost of entering a market.
To evaluate it, we have relied on a mix of methods: MQM (Multidimensional Quality Metrics), LQA scorecards, in-country and stakeholder reviews, player feedback and community signals, data-driven metrics and, more recently, automated and AI-assisted QA. No single methodology defines quality. The strongest programs combine linguistic precision, technical accuracy, and player experience.
Why "no errors" no longer equals quality
Error-based frameworks such as MQM were built for a world where humans wrote every line. Today, multilingual content is generated at scale, and production can be faster, cheaper, and less supervised. We still lack frameworks for the problems typical of AI output, which are especially hard to detect:
Fluent content can still be wrong, and its fluency can trick you into thinking it is right.
The riskiest errors can be invisible: compliance issues, IP-related problems, hallucinations.
Content can be linguistically right and culturally wrong.
Having no errors does not equal quality or a good player experience. The issues players feel most are those that affect meaning, coherence, and brand alignment: context, consistency, and IP-related content. They are also among the hardest to detect.
This calls for a shift in mindset. We are moving from "quality equals linguistic quality" to "success equals content performance." Does the content engage the player? Does it drive retention? Does it resonate in the market? Perfectly translated content that doesn't connect is not high quality. We are moving from measuring correctness to measuring impact.
A three-part quality framework
Evaluating quality today takes three approaches working together.
LQA: Established frameworks assess linguistic and in-context quality.
AI quality estimation for machine and post-edited output: Automated metrics reveal performance patterns at scale. They help us select models by language and content type, improve them over time, and decide which workflow each segment will follow. They signal where to look, but they do not explain what is wrong. Pair them with human evaluation of the weakest segments, and feed them clean data: without clean terminology and translation memories, the metrics are unreliable. Treat them as directional, not definitive, especially for tone, narrative nuance, and cultural fit.
International UX for the player experience. Work with user research and community teams. Add language and culture questions to surveys, run focus groups with LQA testers and player communities, and correlate what players say with bug counts and quality scores. Then connect it to UX metrics (playability, sentiment), marketing metrics (traffic, click-through, conversion) and product metrics (monthly active users, retention and revenue per locale).
One caution on that last point: a weak metric in one locale is not always a localization problem. It may come from product design or internationalization, such as content with religious references that do not travel to certain markets. That is one more reason to consider the target market from the inception of the product.
AI is not culturally neutral
A recent Harvard study asked how close people around the world are to the way ChatGPT "thinks." It found that these models are not culturally neutral. They reflect a specific worldview, mostly aligned with Western cultures, because they reflect the data they were trained on.
For localization, the implication is direct. If we rely too heavily on AI without human oversight, we risk content that feels culturally biased, less authentic, or disconnected from local audiences. Keeping cultural judgment human becomes more important than ever.
Where linguists fit in the age of AI
AI is not eliminating the need for linguistic expertise. It is expanding the places where that expertise creates value. The question is shifting from "Who translates the words?" to "Who owns the language, the culture, the quality, and the player experience?" I see four areas where linguists contribute.
Create. Linguists are still translating, transcreating, and adapting scripts for audio. AI may change how we create, but human creativity and judgment remain essential.
Own and protect. The more content we automate, the more important it becomes that someone owns the outcome. That means language ownership, culturalization, LQA and quality evaluation.
Enable. Linguists make the technology better. They design prompts, evaluate AI output, identify error patterns, manage terminology, and help determine where AI works and where it doesn't.
Lead. Linguists can move upstream into project and program management, workflow design, AI governance, and globalization strategy, where their expertise influences decisions.
The job titles already reflect this. Some localization teams are rebranding as globalization teams, signaling a scope broader than translation. Roles once held by translators are becoming Language Specialists or Language Owners, with responsibility for the whole target-market experience. Translators are also moving into LQA, culturalization and cultural ambassador roles, audio and post-editing, terminology and knowledge management, project management, and strategy and consulting.
This is a transition period for linguists and for the industry as a whole. My advice is to invest in adaptability, continuous learning, and diversification while building areas of specialization. Get better at storytelling: show the value you bring and connect it to ROI and meaningful metrics. The opportunity is to move from producing words to shaping experiences, systems, and strategy. I see a bright future for translators who are willing to keep evolving with the industry and its technologies.
