Hockey games live and die by rhythm. A clean faceoff win, a breakaway, a rebound that bounces into chaos at the crease—each moment needs a call that feels immediate without becoming a loop players can predict before the puck reaches the blue line. That is the challenge EA Sports appears to be tackling with generative AI in NHL 27.
Veteran broadcaster John Buccigross, who is set to appear in the game, has said that the development process included AI-generated versions of his voice for additional commentary lines. The remarks provide a relatively specific glimpse at how the technology may be applied: not necessarily as a replacement for the commentator, but as a way to produce or extend voiceover for the enormous number of situations a modern sports game attempts to recognize.
It is also a reminder that sports commentary is more difficult than reading a stat line in a convincing voice. In a sport with an especially devoted vocabulary, tiny word choices can make an arena feel right—or make the broadcast sound as though it was written by someone who has never watched a period from the press box.
One small word, one very hockey-sized problem
Buccigross described hearing an AI-generated call that celebrated a goal by saying the crowd had risen from its “chairs.” His correction was straightforward: at a hockey venue, the customary word is seats. That may sound like a minor distinction outside the sport, but it is precisely the sort of detail fans notice instantly.
The anecdote matters because it frames the practical limits of generative voice systems. A synthetic performance can sound remarkably close to the person whose voice it models, while still delivering language that is contextually off. The issue is not pronunciation alone. Commentary needs subject knowledge, timing, regional phrasing, league terminology, and an awareness of what sounds natural during a live-style sports broadcast.
In other words, getting the sound of a broadcaster right is only half the task. Getting the broadcaster’s world right is the other half. A line can be technically understandable and still land with the authenticity of a Zamboni trying to take a slapshot.
Buccigross indicated that such details require human review and manual adjustments. That suggests a workflow in which generated material is not simply left to run unattended. But it also illustrates why the human role remains central: someone has to recognize that a plausible phrase is wrong for the sport, provide the correction, and make sure the corrected language carries through future use.
Why a sports game would want more lines
Annual sports games have an unusual voiceover burden. Unlike a tightly scripted story game, they need commentary that responds to a vast web of player actions, teams, venues, scores, penalties, momentum swings, replays, and presentation features. The more bespoke the reactions become, the more obvious repetition can be when the same handful of lines reappear across dozens of matches.
Related coverage includes NHL 27 Commentator Voiceover Includes Generative AI, John Buccigross Says.
Generative AI could theoretically help build a wider catalogue of calls, connective phrases, and situational remarks. A game may be aiming to make broadcasts feel less rigid while giving commentators room to respond to more specific on-ice events. The appeal is understandable: hockey moves quickly, and commentators must make sense of action that can change direction in a second.
Still, a larger pool of dialogue is not automatically better. Sports commentary works because it establishes a recognizable voice. A broadcaster’s cadence, preferred terminology, sense of excitement, and restraint after a routine play all create an identity. If AI-created lines become too generic, too verbose, or too eager to announce the obvious, the result could make a game feel more artificial rather than less repetitive.
The best use of the approach would depend on careful editing and clear guardrails. Commentary cannot merely identify an event. It needs to sound like it belongs in the sequence of play, and it needs to respect the language hockey fans expect to hear.
What remains unclear about NHL 27’s use of AI
Buccigross’s comments establish that generative AI was used for commentary-related voice material, but key details have not been publicly clarified. It is not known how much of the final commentary relies on generated lines, how the performer’s voice rights are structured, what approvals or review steps are used, or whether the technology appears elsewhere in NHL 27.
That uncertainty is important. “AI is used” can describe many very different practices, from limited assistance during production to a much more substantial role in the final game. Fans, performers, and developers have valid reasons to care about where on that spectrum a project sits.
There is also a distinction between generating text and generating a vocal performance. In this case, the discussion is about a recognizable commentator’s voice. That naturally raises questions around consent, compensation, control over how a voice is used, and the ability to correct material that does not fit the speaker’s standards or the sport’s terminology.
Buccigross’s own comments reflect a pragmatic side of that conversation: less recording work may be appealing if compensation remains fair. At the same time, his story about “chairs” versus “seats” demonstrates why reduced recording time does not mean reduced oversight. Someone still needs to listen closely.
Disclosure is becoming part of the conversation
Generative AI has become a flashpoint throughout the game industry, particularly when its use is visible to players or connected to creative labor. Recent pushback over AI-generated cosmetic art in major games showed that audiences often want to know whether artwork, writing, audio, or other assets were made with generative tools.
EA Sports’ 2026 editions of College Football, Madden, and FC carry generative-AI notices on their Steam pages. NHL does not have a Steam release, so it does not offer players that same storefront disclosure route. That does not establish that the implementations are identical across games, but it does make clear information around NHL 27 especially valuable.
Disclosure is not a substitute for good creative practice, but it gives players useful context. A concise explanation of what technology was used, what it was used for, and how human performers and editors participated would help distinguish a narrowly targeted production tool from a vague marketing buzzword.
The discussion also reaches beyond video games. AI-assisted interfaces and entertainment systems are increasingly becoming consumer-facing features, as seen in the broader debate around Gemini-driven infotainment and familiar in-car platforms. Across products, the recurring question is similar: where does automation genuinely improve an experience, and where does it create new uncertainty about quality, control, or transparency?
NHL 27 arrives with familiar strengths and familiar concerns
NHL 27 launches September 11, 2026, for PlayStation 5 and Xbox Series X/S. It has received a 6/10 review score, with the assessment finding the game enjoyable in the fundamental ways hockey games tend to be—big hits, sharp wrist shots, and the inherent thrill of the sport—while arguing that the annual entry does not deliver substantial changes over its predecessor.
That context makes the AI commentary discussion more interesting. For a series often judged on incremental improvements, expanded or more responsive broadcast presentation could be a meaningful quality-of-life advance if it helps matches feel fresh. But if generated lines introduce incorrect terminology, awkward phrasing, or a voice that no longer feels distinctly human, they could become another point of friction in a release already being measured against familiar foundations.
There is no reason a tool has to erase a performer’s contribution. Used with permission, compensation, editorial care, and transparent limits, it may help a sports game address the sheer scale of live-style commentary. Yet hockey is a game of details, from line changes to puck battles along the boards, and its language deserves the same precision.
For NHL 27, the test will not simply be whether an AI voice can sound convincing. It will be whether the game knows the difference between a chair and a seat—and whether it understands why players care.
