The popular version of this story says a generation raised on voice notes and smart speakers is done with typing, so voice AI wins by default. The polling doesn’t support that. Even 18- to 24-year-olds, the age group that likes voice notes most, prefer a text for long messages by roughly three to one.
There is a better case for voice AI, and it comes from classrooms rather than messaging habits. When children can speak instead of type, many of them say more: longer answers, richer vocabulary, more reasoning. That’s a reason to build voice in, not as a replacement for text, but as a way to let people express a thought before they can polish it.
- Young adults like voice notes more than older adults do (43% of UK 18- to 24-year-olds vs. 11% of those 55 and older), but 71% of 18- to 24-year-olds still prefer a text for sending a long message (YouGov, 2022).
- In a German study, fifth and sixth graders' spoken answers averaged about 33 words with 1.8 explanation arguments, compared with about 10 words and 0.6 arguments typed. Different groups and settings mean the input method can't take all the credit.
- A voice assistant can transcribe a child correctly and still miss the point: in one lab study, child-led speech was transcribed with 84% accuracy, yet devices responded meaningfully only about half the time.
- 67% of U.S. 9- to 17-year-olds have used an AI chatbot (Common Sense Media, 2026). Conversation with AI is already familiar, which makes voice a natural next input. That part is a forecast, not a finding.
- Voice doesn't make AI safer or healthier on its own. A four-week randomized study of 981 adults found no significant difference between text and voice chatbot conditions on psychosocial outcomes.
The popular story: a generation done with typing
The argument has real evidence behind it, which is why it spreads. In YouGov’s May 2022 survey of UK adults, 43% of 18- to 24-year-old smartphone users said they liked receiving voice notes, compared with 11% of those 55 and older (YouGov, June 2022). Teenagers have also taken to conversational AI quickly. Pew Research Center found that 64% of U.S. teens aged 13 to 17 had used an AI chatbot, and about three in ten used one every day (Pew Research Center, December 2025).
Put those two trends side by side and the conclusion seems obvious: young people like talking, they like AI, so they’ll talk to AI instead of typing to it. The trouble is that each data point measures something different. Liking voice notes is not a preference for dictation. Using a chatbot says nothing about whether you typed or spoke to it. And adult polling can’t stand in for what school-age children prefer.
What the polling actually shows
Text still wins, in every age group. The same 2022 YouGov tables that show young adults warming to voice notes also show that 71% of 18- to 24-year-olds would rather send a long message as a text, against 24% who’d choose a voice note. For short messages the gap is wider: 91% text, 6% voice note.
Newer data points the same way. In March 2026 polling of Great Britain adults, 91% of adult Gen Z respondents regularly used messaging, while 15% of all adults regularly used voice notes (YouGov, April 2026). Voice notes are growing at the margins, and messaging remains the default. Neither survey asked whether messages were typed or dictated, so neither measures keyboard use directly.
We also looked for a representative study that asks today’s children whether they’d rather dictate or type everyday messages, and then connects that to how they use AI. We didn’t find one. That doesn’t prove it doesn’t exist, but it does mean that “Gen Alpha prefers talking to typing” is, for now, an assumption.
Where speaking does help: saying more
The strongest evidence for voice isn’t about preference. It’s about expression. When children speak, many of them produce more complete thoughts than when they type, at least in the tasks researchers have tested.
In a Swedish study of 81 students in Grades 4 and 5, mostly from regular classrooms, each child wrote stories both with speech-to-text and with a keyboard, in counterbalanced order. With speech-to-text they produced longer texts, in less time, with more varied vocabulary and longer sentences (Almgren Bäck, Nordström, and Svensson, Reading & Writing Quarterly, 2025). Because every child used both methods, this is the most direct comparison available.
A German study of fifth and sixth graders explaining math problems found a bigger gap. Spoken answers, given by dictation or audio message, averaged about 33 words and 1.8 explanation arguments. Typed answers averaged about 10 words and 0.6 arguments (Hankeln et al., Technology, Knowledge and Learning, 2025).
The authors are explicit about the limit: spoken answers came from one group of students working one-on-one with a test administrator, and typed answers from different students in whole-class settings. The difference can’t be attributed to the input method alone. The finding is about expression in a specific task, not intelligence. Speaking didn’t make anyone smarter. It made more of what they already understood visible.
Speech-to-text can also help children who find writing hard. In a small Swedish study of 16 children aged 10 to 13 with reading and writing difficulties and 12 comparison peers, final texts written with speech-to-text had fewer errors, with a bigger benefit for the difficulties group. Overall text quality didn’t differ between methods, and recognition mistakes still had to be corrected (Kraft, Frontiers in Education, 2023).
Hearing words isn’t understanding them
A system that transcribes a child correctly can still fail to help. In a lab study of 28 children aged 5 to 10 talking to a commercial voice assistant, child-led conversation was transcribed with 84% accuracy, but the device responded meaningfully and on topic only about half the time (Kim et al., International Journal of Child-Computer Interaction, 2022). Responses improved as children got older and their phrasing became more standard.
Recognition itself is still uneven for the youngest speakers. A 2025 study tested two versions of Siri and Alexa on speech from children aged 2, 3, and 5, and found that human listeners far outperformed the assistants at every age, especially for the youngest children, even though Siri had improved (Bradley, Yu, and Johnson, JASA Express Letters, 2025). These are specific systems, not every voice AI, and newer models may do better. They’re still a good reason to doubt claims that any child can simply talk to any device.
The often-quoted “three times faster” figure needs the same care. It comes from a controlled study of 48 university students (24 in English, 24 in Mandarin) transcribing short messages on an iPhone 6 Plus: English speech entry ran at 153 words per minute against 52 for the keyboard. Speech also left slightly more uncorrected errors in the final text, 1.30% versus 0.79% (Ruan et al., ACM IMWUT, 2017). Adults, ideal conditions, copied text. It is not a speed estimate for a child composing their own thoughts.
Dictation, assistants, and agents are different things
Much of the confusion comes from treating every kind of speaking to a phone as the same behavior. They aren’t, and the research above applies to some of them and not others.
| Term | What it does | Don’t confuse it with |
|---|---|---|
| Voice note | An audio recording sent to another person | Dictation, or a live conversation with AI |
| Dictation (speech-to-text) | Turns your speech into editable written text | AI writing for you, or AI talking back |
| Voice assistant | Answers questions or runs a fixed set of commands, like setting a timer | An agent with access to your information and tools |
| Voice agent | Takes a spoken request and uses context or tools to help finish a task, asking follow-up questions when needed | An AI companion built to simulate a relationship |
The school studies are about dictation. The voice-assistant studies are about assistants. Neither tests a voice agent, which does something different with the words it hears. For a closer look at that line in a text context, see our explainer on the difference between an AI agent and a chatbot.
Why voice agents could matter more to the next generation
Conversation with AI is already ordinary for most kids. Common Sense Media’s 2026 census found that 67% of U.S. 9- to 17-year-olds had used an AI chatbot, rising with age from 58% of 9- to 12-year-olds to 77% of 16- to 17-year-olds.
Kids are also bringing personal questions to these tools. Among 9- to 17-year-olds who use AI, 57% have used it for information or advice about their health or body. Most would still go to a trusted adult first with a health question (73%), but 12% would turn to an AI chatbot first.
From here on, this is our reading of the evidence, not something any study has measured. Three things point toward voice agents mattering more for people growing up now:
None of this is unique to children. A caregiver with a toddler on one arm, or an older parent who finds a phone keyboard hard to use, may get the same benefit from saying a question out loud. Those are plausible use cases, not populations the child studies measured.
What voice doesn’t fix
Speaking to an AI doesn’t make the experience safer or better for wellbeing by itself. In a four-week randomized study of 981 adults assigned to text, neutral-voice, or engaging-voice chatbot conditions, researchers found no significant effects of the conditions on the psychosocial outcomes they studied. Participants who chose to use the chatbot more, in any condition, tended to have worse outcomes, including more loneliness and emotional dependence (Fang et al., preprint revised October 2025). That last finding is an association, not proof that heavy use causes harm, and the study did not include children.
Content matters more than the input method. Among 9- to 17-year-olds who use AI chatbots, 17% reported seeing something they felt was inappropriate for their age, and only a third of them told a trusted adult (Common Sense Media, 2026). A voice interface carries that same risk into a format that’s harder for a parent to glance over.
“Voice changes how easily a thought gets out. It doesn't change whether the answer that comes back is right.”
Text also keeps doing jobs voice can’t. You can reread it, edit it, search it, and send it in a quiet room. The YouGov numbers suggest people already know this: voice for some moments, text for most. The products worth building offer both, and let the person choose.
What this means for families
If you’re a parent, the research supports a few practical moves, none of which require believing that typing is finished:
For adults managing family health, the case for voice is the same one the classroom studies make, just with different hands full. It’s often easier to say “what did Mom’s cardiologist change at the last visit?” than to type it. For how agents handle that kind of question across a family’s records, start with our complete guide to AI agents in family health, and before connecting any tool to that data, read whether it’s safe to share family medical records with AI.
Kai, Kaizen's AI assistant, now takes live voice conversations. Ask about a parent's records out loud, follow the transcript as you talk, and come back to the saved conversation later. Kai is built for adults managing their family's health, not for children.
Talk with KaiSee everything that shipped with voice in the Kaizen 1.17 release notes.
Frequently Asked Questions
For some writing tasks, it helps children say more. In a 2025 Swedish study of 81 students in Grades 4 and 5, dictated stories were longer, took less time, and used more varied vocabulary than typed ones. That measures one kind of task performance. It doesn't show that speech should replace typing or literacy instruction, and dictated text still needs reviewing and editing.
Young adults like voice notes more than older adults do, but they still prefer text for most messages. In YouGov's 2022 UK polling, 43% of 18- to 24-year-old smartphone users liked receiving voice notes, compared with 11% of those 55 and older. Yet 71% of 18- to 24-year-olds preferred a text for sending a long message, versus 24% who preferred a voice note. No representative study we found measures the same preference among children.
A voice assistant answers questions or runs a fixed set of supported commands, such as setting a timer. A voice agent takes a spoken request and uses context or tools, like your saved records or a web search, to help finish a task, and it can ask follow-up questions when the request is unclear. Dictation is different from both: it only turns speech into editable text.
It depends on the product and the supervision around it, not on voice itself. Recognition of young children's speech is still uneven, and a 2026 Common Sense Media survey found that 17% of 9- to 17-year-olds who use AI chatbots had seen something they felt was inappropriate for their age. Use products designed for a child's age, keep a trusted adult involved, and treat AI answers about health as a starting point for a conversation with a clinician.