2026 Localization Trends Webinar

Full transcript

Never know if someone's rating something perfect, and another person would say this is totally wrong or this is, like, completely off basis. So I think risk is the most important factor here, and I do think that this was happening in 2025, but potentially this was happening before 2025 anyway.

Trend 3: AI-generated translation will need to be explicitly labeled

Let's hold up our cards. Yes from Chiara. Sahil's in the middle, Eric's in the middle, and Mario's a hard no. Okay, Chiara and Mario, since you guys were the firmest on your answer, let's hear from you.

Chiara: I can start. I think it's a yes, for different reasons, from different angles. First of all, you need to label it for vendors, for freelancers, for example, who are doing the post-editing. You need to tell them transparently that that was AI output rather than human-generated output. Also, you need to flag it to customers. For example, user-generated content that is translated with AI or Google Translate — I see companies that do that. They show it as a disclaimer: "FYI, this was generated with Google Translate," like Booking.com does for reviews. It's about being transparent on the quality of the content you're putting out there. Also, on the quality-monitoring side, if you flag issues, you want to pinpoint whether it's an AI problem or a human-generated problem, so you can share feedback with the people training the AI or managing the vendors. So, strong yes.

Mario: I think I changed my take on this throughout the week. For me it's almost about: will it need to be explicitly labeled? I don't think this has happened, because we do see it labeled, and we also don't see it labeled. I think it's a good idea for the reasons Chiara mentioned. From a European law standpoint, this might be coming — though I'm not sure I'd trust AI to tell me that. I'm also interested in the almost-clash of localization being created as if it was specifically designed for the user. Why would we explicitly label something? It's like English content — you'd never label "this was created by our content designer, Peter." If it's wrong, and there's risk, and customers are churning, does it matter if a human messed up or the machine messed up?

Mario: At my last company, we launched ten or eleven new languages, and as soon as a user changed the language from the default, a banner would pop up: "We launched something for you quickly, and it's not perfect — give us feedback." The important thing was gathering feedback and saying "you asked, and we delivered, and it's not perfect, but here's something you can use." I thought that was a good idea. I didn't think we needed to explicitly label it, though — that's the difference.

Sahil: I think saying this trend "will need to be explicitly labeled" is a very tall claim for 2025. There are two different angles. From some B2C companies — Booking.com, and a Berlin-based company I know — they explicitly state when content is AI-generated, from a brand point of view. On the other side, from our clients' perspective, labeling something as "translated by AI" is fairly meaningless. What clients actually want is human auditability: if we use AI to translate, can we give them a quality score, and based on that score, decide whether they want a human in the loop? Just labeling something as AI-translated doesn't add much value compared to making sure the quality is good enough and whether human review is needed.

Erik: Aside from EU regulations and brand protection — both good reasons to label content — Sahil, you're bringing up an important point: traceability, and what you do with that information. Saying "this has been AI-translated" doesn't assert quality; it only tells you about accountability — you're saying "I'm adding this metadata because I'm not going to take responsibility for a poor outcome." So we didn't get universal labeling, and I don't think we will, but we are getting traceability, and that's the right direction. Directionally correct without being technically correct: "AI-generated translation will need to be explicitly labeled" — no. But there are good reasons and context for it — yes. That's my mixed assessment.

There's interesting conversation in the chat too. Ricardo mentioned it should be, but it's not going to be — there's pushback from linguists, but also organic readers and clients, and AI-generated content is progressively being mixed with human-generated content. It's not black or white; most of the time there's a mix, so how could you even say something is AI-generated or human?

Erik: Exactly — and it degrades trust when you can't tell. You start doubting whether something is real or AI, and there's an implicit value judgment either way. Companies should do it because it clarifies their accountability. A colleague of mine wrote a book and said "no AI included" — no AI participated in creating it. That's a tag of "human only," which actually enhances brand value in that context — the opposite of an AI-generated label. "No AI" may even be more valuable than an AI-generated label.

Trend 4: Knowledge graphs will move from niche tooling to core infrastructure

Knowledge graphs will move from niche tooling to core infrastructure for controlling AI and content, pulling loc teams into upstream data and architecture decisions. Let's vote: yes from Mario, yes from Chiara, yes from Sahil, and in between for Erik.

Sahil: LLMs perform dramatically better when they're given structured knowledge. For that to happen, localization champions, managers, and teams are being pulled into more of the modeling side of things — style guides, glossaries, how the structure of those documents should look before they're uploaded. That's a value-add a localization champion can bring in, which used to be more of an afterthought relying on human judgment. For these assets to be trainable and usable by AI, that content-modeling work has started to happen. I wouldn't say this is 100% there, but it's a strong trend I observed in 2025.

Erik: If you're taking AI translation seriously at scale, mitigating risk appropriately, and handling accountability correctly, some form of structured semantic knowledge should be implicit — built into your architecture. That said, a lot of companies hide that; there's a lack of governance and clarity around semantic ownership, which separates the people who implement it from the people who have that semantic accountability. So knowledge graphs don't get formally rolled out, but they're implicitly built into many ecosystems. I'll also note a huge chunk of the industry still doesn't use AI translation at all. When AI is involved and done responsibly, there should be semantic structure with appropriate risk management and accountability aligned to it. That's where my "no, in reality it's hidden, not formalized, but it is there" comes from.

Sahil: So implicitly this is happening, or has been happening implicitly — it should be happening, and often is, but nobody's making a big deal about it?

Erik: Especially for RAG approaches — vectorized models will include things like semantic structures, of which knowledge graphs are one. TMs are kind of in that same class too, with the appropriate metadata. It's not a standalone winner flag — it's also just a file in the RAG repository. The responsibility for that semantic structure is often separated from the people accountable for localization; many of us could be dealing with semantic structures that don't exist within our organizations, and the people accountable for classifying them may be out of reach for localization teams. Just a hypothesis.

Chiara: Fair point, Erik, I agree. I also interpreted this prediction more broadly — specifically the role of the localization team, especially coming from an enterprise company. I do see this happening, and where it's not, it's a call to action: localization teams being more involved in AI dataset training but also in business decisions, getting their voice heard. I think it has happened, and where it hasn't, I'd suggest everybody go up there and be louder.

Mario: Agreeing with Chiara. I have a linguistics degree, so semantics came up a lot, and diving into this discussion got pretty technical. To level it up more generally: I think we're moving from niche tooling, where people didn't know what was involved, to localization and language having a moment due to AI — more people talking about TMs, using the data available. I think that is happening, and it needs to happen, moving more upstream into data and architecture decisions. A hundred percent.

Trend 5: Written and spoken language services will converge into one multimodal experience layer

The last trend: written translation and spoken language services will converge into a single multimodal experience layer, making voice, video, and text inseparable parts of the localization strategy. Yes across the board, with Erik on a "yes, but."

Erik: Technically, this already happened a long time ago — users are accustomed to it. On the back end, though, it's still fragmented how we organize ourselves to build these capabilities. Customers expect this and already fuse these things together, but on the back end there's quite a lot of disjointed fragmentation across voice, video, and text workstreams.

Sahil: There's been a strong push toward making sure all these different multimedia formats are put together and localized in one platform, boiling down to shared infrastructure and models. It might not be that every company had it fully in 2025, but there was a surely strong push toward a single source of truth that can cater to different formats, backed by a shared model. Strong trend in 2025.

Mario: I'd agree with that. "Multimodal experience" is one of those tricky keywords, but "inseparable parts of the loc strategy" — a hundred percent, especially with AI producing more and more content (things like ElevenLabs for AI dubbing and voice). Why would customer experience be separated? To Erik's point, are we there yet? Probably not, but we need to think about it as a multimodal experience, with one core centralized infrastructure underneath handling TMs and everything else.

Erik: Some AI capabilities are genuinely new to this ecosystem — automatic transcription, text-to-speech, speech-to-text, automatic translation, synthesizing realistic AI voices. You could have a song, or an operatic aria, written about this exact webinar within seconds. But for brand-contingent activities at scale, delivering that and controlling quality to a level that keeps the brand experience positive still has a ways to go. Powerful new tools, but still complexity to reach that last chunk of quality.

Chiara: It also depends on the value of multimodal content for each company. Some companies are all about video and have more advanced tooling; others have more basic tooling. This started even before 2025 — does every company prioritize it at the same level and give it the same resources? I don't know, just launching that question.

Mario: Great point from Ricardo too — it really depends on a company's maturity in internationalization/localization, and beyond that, its maturity in content creation itself. Are you even creating these types of content, or mostly just text? I think we're at an experimentation phase with this trend, and it definitely didn't happen 100% across the board in 2025–26.

The 2026 Big Bet: panelist predictions

Erik Vogt — Pricing will move away from word count

Erik: My bet is about pricing. The value stack is shifting so fundamentally that it will force fundamental changes in how pricing is structured. Prediction: by the end of 2026, a majority of enterprise localization programs using AI at scale will no longer rely on word count as their primary pricing unit, instead anchoring spend to program fees, risk tiers, or outcome-based constructs. Words have already degraded as a core metric — complex fuzzy tables, AI discounts, MT discounts, light/heavy post-edit, review layers have added enormous complexity. Words are still easy to count and will still be part of it, but tokenization becomes a more interesting number with automatic translation. Risk tiers differ too — who is capturing value in the stack, where AI gets pulled in. Flat fees, minimums, enablement fees, governance fees, and edit distance (or time to fix) could become the core value driver rather than raw word counts, which neutralize who's working on it and neutralize process.

Chiara: From an enterprise perspective, how do you forecast a budget for the whole year on anything other than a simple per-word-rate system? It's fair to consider different ways of paying people, but you need to be able to forecast at scale.

Erik: Budgeting is one thing, reality is another — feedback loops of effort to outcome can be predicted and need to be built into our heuristics. Word count alone doesn't tell you much unless you know what TM you're leveraging against; then you get a weighted word count, a modified version of word count. Word count alone has probably already gone out the door as the only factor you need.

Sahil: If that happens by 2026 I'll be very happy — I'm not a big fan of pricing by word count, but it is by far the easiest value metric for customers to reason about. When you say "outcome-based," what's the final metric a customer would actually pay for?

Erik: Edit distance is an easy one — how much work a trusted resource requires to say "this passes." It varies with the actual requirements of the person you're holding accountable, and that accountability becomes part of it. There are also risk-profile tracks — probability of a negative event times the impact of that event — and pricing workflows that align with each risk profile. Instead of "what's your price per word," it becomes "what's your price per word based on the accountability stack." I need to be right that we've done something different than the standard fuzzy-match/MT-discount price list, which I think is obsolete.

Mario Pluzny — AI won't eliminate localization roles, but it will change the job titles

Mario: Approaching this from a coaching angle — almost ten years in the industry, recently focused on helping people with their careers post-layoff. My big bet: AI won't replace localization work or eliminate roles, but it will quietly remove the traditional job titles we're used to. Localization management, program management, vendor management aren't disappearing, but the standard ladder from five years ago — linguist, project manager, program manager, up to head/director of loc — isn't there anymore. It's shifting toward language/localization specialist gathering AI knowledge, then moving into global experience, data, governance, localization engineering, AI specialist roles. From my coaching practice, people aren't scared of AI tooling replacing them — they're scared of being professionally less relevant or misaligned; they feel stuck and undervalued because showing what localization does is harder now. But the skills we have are more valuable than ever — we need to reframe what the market is hiring for. Storytelling becomes really important: how do you sell what you have to offer, position your value with business impact, speak the language of leaders and stakeholders, and elevate into the data/governance layers at the top.

Erik: So does your bet mean every localization job will need the word "AI" in it somehow, in the title or resume?

Mario: That would be oversimplified. I think titles are shifting into adjacent product and engineering roles, maybe as part of the engineering or design stack — something we saw at my last company. Searching "localization project manager" today versus a year or two ago in Dublin (the European hub), there used to be hundreds of roles and now there aren't many; you have to dig through the job spec to find the skills being used. The label stays the same, but what's under the hood is changing, with AI as part of that.

Mario: For advice to anyone joining or upskilling in the industry: two patterns — either lean in and be at the forefront, not just AI-fluent but not AI-dependent (since everyone can use AI now), shifting toward being a systems architect who speaks business impact and process improvement rather than just words/translations/language. Or, if that's not your thing, take your transferable skills into adjacent industries where AI has less impact. The market as a whole isn't shrinking — it's realigning. Don't be afraid of AI, and don't undervalue what AI can't do: critical thinking, human skills, leadership and management.

Sahil — TMS platforms will evolve into content hubs

Sahil: My bet: in 2026, TMS (translation management system) companies are going to evolve from just handling translations into becoming content hubs — a system of record. Today these tools integrate with marketing tools like HubSpot or different CMSs and email automation providers, which push content into the TMS for translation. Over time, these companies will evolve into storing all content and becoming a single source of truth. Example: if Brittney, a director of brand, wants to change a brand wording, she could change it once in the TMS and have that change propagate into HubSpot, Contentful, and everywhere else that content lives — decreasing human error and effort. Because these tools connect to marketing and even HR tools, I think the single source of truth / system of record is what these tools evolve into — becoming a content hub very soon.

Erik: The logic is sound, but it shifts the question of who controls these assets and where capabilities get stacked. You could tack a translation layer directly onto a content management system and eliminate the TMS entirely, or the TMS could act like a content management system in proxy-based delivery models where content lives in and gets distributed from that platform. The value stack has shifted, so there's a different conversation about where to orchestrate from. Adding AI capability changes multiple systems at once — e.g., AI-generated content needing to be localized on a separate track, kept apart from human translation TMs used for leveraging or training.

Sahil (summary): The TMS becomes a content hub — a single source of truth / system of record for content that's otherwise scattered across fragmented tools in the company. Consolidation of the content stack.

Chiara: Sounds like a dream — the dream of a client-side loc manager is to have a centralized tool. In reality, I don't think this happens this year; it would make other tools (HubSpot, Zendesk, content platforms) obsolete for marketers and designers as a source of truth. I love the idea of one source of truth, I just don't see it happening this year, unfortunately.

Mario: Should be happening already, don't know if it's going to. There are always some technical blockers. But the idea of moving ownership of content-related decisions to localization is good — we can build something from that.

Chiara Scaldaferro — Localization teams will run multilingual AI operations

Chiara: We've been through a phase of experimentation with GenAI for content production. I think 2026 and the following years will be about governance and control, and this will shape the role of localization managers and teams within enterprises: they'll run AI operations, specifically multilingual AI operations. Their role won't just be archiving/project management — they'll set quality guardrails, maintain and evaluate language datasets for AI training, and enable other teams to use AI safely and consistently in content production. This is backed up by the CSA research prediction about knowledge graphs, and that's how I interpret it: AI ops teams are emerging, and localization should play a big role in them.

Sahil: It's quite plausible — with the rise of ops functions across departments due to AI, you need a central ops function, and there's a higher likelihood localization is part of it.

Erik: The governance angle nails it — control, and who is accountable for a particular outcome, becomes central: risk management, process design, pricing all hinge on it. What's the concrete, measurable bet here — more governance than now, or something measurable we could bet on?

Chiara: Somewhat connected to what Mario said about new job titles. It may be hard to measure, but if you see localization represented at the C-level, that would be the ultimate goal — achieved by participating in the AI trend and important company functions, and the more you do, the higher you can climb.

Erik: The strategic framework has been destabilized by a fundamental shift in how the value stack gets created. Think about the ROI or business design of any function, including your own job — how are you adding value, and what's the scale of that impact? That will determine what jobs remain, what tools people buy, what vendors they select. There are four levers: risk reduction, cost reduction, differentiation, and enablement. We often default to cost reduction because it's the lowest-hanging fruit, but differentiation and enablement are the game-changing ones. Whoever gets control of those strategies and can articulate them best will win — accountability and governance are a huge part of it.

The panel closed the betting table and opened a live audience poll on who would “win big” in 2026 with their prediction. After a close vote between all four panelists, Chiara Scaldaferro was voted the audience favorite. The panel plans to turn the 2026 predictions into written content so the industry can revisit them in 2027 and check which bets actually paid off.