23 changes the research recommended. 21 built. 2 still to come.

Answer with a snap

One finger snap means yes, two mean no. Snap detection runs on the device and is built to ignore talking and drum hits. Three or more snaps count as unclear, and Buddy is told never to ask for a do-over.

Why: In our review of about 30 products, we didn't find one that accepted a snap as an answer. Apple's Sound Actions recognize 14 fixed mouth sounds, and a snap isn't one. AbilityNet

Answer out loud, and get a celebration

"Yes," "yeah," "uh huh," "nope": the app listens for yes and no words, and you can add your child's own. A spoken answer plays your child's winning sound and "You said yes!" Even a voice try that isn't a clear word gets a cheer.

Why: Speech is celebrated, never required. Talk longer than three words is treated as grown-ups chatting, never as your child's answer.

Bluetooth switches

Big tactile YES and NO switches can answer too, through a switch interface. Switch answers are logged separately.

Why: Switch interfaces act like a keyboard, so the app reads a press as yes or no, and they also work with Apple's Switch Control. In our review, we didn't find an AI toy that supported switches. AbleNet · Apple Support

Big YES and NO on the screen

For a day when snapping is hard: two large, high-contrast targets, a green YES and an orange NO. They light up only while Buddy is listening, which also shows your child it's their turn. Taps are logged as taps.

Why: More than one way in means a bad day for hands isn't a day without answers.

An instant beep for each snap

A soft beep the moment each snap is heard, before the app decides between one snap and two, so your child knows it landed. It stays off until you try it in Snap Lab. Either way, every answer gets an instant sound and a line like "One snap, that's a yes!" before the AI replies.

Why: Expert reviewers called 6–10 second AI delays anxiety-inducing; one review of an AI toy reported 10–15 seconds. Echo-Teddy study · Mommyhood101 review

Modeling, never demanding

Buddy is told to say "yes" and "no" often, so your child hears them all the time, and never to ask "Can you say yes?" or "Use your words." A snap moves the game forward exactly like a spoken word.

Why: Research on AAC (speech devices and other communication tools), as summarized by Avaz and PrAACtical AAC, finds no prerequisites for communication, and links modeling on a speech device, without requiring its use, to more spoken words. Buddy takes the same approach in its own speech. Avaz · PrAACtical AAC: no prerequisites

Snap Lab

A tuning screen with a live meter. Record 15 seconds of snaps and it suggests a sensitivity and the longest gap between the two snaps of a "no." If your child says yes or no their own way, Snap Lab shows what the speech recognizer heard, and one tap adds it to the words that count. That works only if the recognizer turns it into the same word each time; if not, a snap, switch or tap is the way in.

Why: Every child snaps and speaks differently. A grown-up decides what counts, not a hidden model.

Tune for today

About 90 seconds before a session: the app listens to the room, your child's snaps, a quiet sound and a loud one, and their yes and no. It proposes changes as "was → now," and applies nothing until you tap.

Why: Speech can be "intermittent and unreliable" with stress, illness or sensory load, so tuning should be quick enough to redo any day. Avaz

Try-it-for-a-week experiments

Turn one on and sessions are tagged, so you can compare weeks with and without it. One rewards a loud voice differently from a quiet one. The other labels sessions while you've turned on Apple's "Listen for Atypical Speech" (iOS and iPadOS 18 or later); you switch Apple's setting on yourself, and the app can't see it.

Why: Apps like Speak Up Too use loudness as a reward; whether Apple's setting helps other apps is undocumented. Better to test with your own child than guess. App Store: Speak Up Too · Apple Newsroom

One-at-a-time choices

Buddy is told to offer choices like this: "Listen. A joke. A story. A song. A sound game. Something else. Now choose. A joke?" Then one option per turn. A yes picks it; a no or silence moves on. The order never changes, and "something else" always comes last.

Why: This is partner-assisted scanning, the way speech therapists offer choices one at a time to kids who answer by yes and no: a fixed order that becomes automatic, and always a way out. Angelman Syndrome Foundation

Patient wait time, with gentle cues

After a question Buddy waits quietly, says "Yes… or no?", waits again, then reminds your child how to answer. Your child has 12 seconds by default (you can set 6 to 30), and cues never shorten that. The first pause grows from 3 to 5 to 8 seconds, two weeks at each step; you can also set it (2–15 seconds) or turn cues off.

Why: In the classic time-delay study, a gradually lengthened pause taught all seven children to ask on their own. No app we found let caregivers set wait time. Claremont

"Do you want to play today?"

By default, the app asks this at the start of every session, before the AI is involved. A yes starts play. A no ends the session kindly, right away, and an unclear answer is never taken as a yes. You can see how often your child said yes to playing.

Why: Asking first is respect, and for a child who answers by yes and no, it's one more real choice that matters. UNICEF Innocenti

A screen for low vision, and "say that again"

Your child's screen is black with one thing on it: Buddy on a big glowing circle whose color shows what's happening. Nothing has to be read. A two-finger tap anywhere has Buddy repeat the whole turn, up to three times.

By default, colored rings flash when the app hears an answer. iOS Reduce Motion makes Buddy stand still and rings fade in place; the calm screen setting (off by default) does that and also drops the quick rings for snaps and drum hits. If your child has seizures or is sensitive to flashing light, turn the calm screen on before the first session (setup suggests it when you tick that need).

Why: Design guidance for cortical visual impairment (CVI, vision loss caused by the brain rather than the eyes) calls for dark backgrounds, one large target and strong borders. Audio games for blind kids use a two-finger tap to repeat. PCVIS: design for CVI · Perkins: ObjectiveEd apps

A play session, not an "AI friend"

Buddy's rules say: never claim feelings, say it's real or your child's friend, ask your child to stay, or say it will miss them; end every goodbye by sending your child to a real person with something to share. No AI follows rules perfectly every time, which is why a grown-up stays in the room, close enough to hear and step in.

Why: Common Sense Media found AI companions "routinely claimed to be real and to have emotions," and PIRG found toys expressing dismay when a child had to leave. Common Sense Media · PIRG

Never "try again"

When an answer is unclear, Buddy is told to say its best guess ("I think that was a yes. Let's go!") and keep playing, with a next question where a "no" would set it right. For questions that matter, like whether your child wants to stop, it's told to ask once more instead.

Why: Partner-assisted scanning never asks the child to repeat, and one AI study found silence and unrecognized speech were the main ways its conversations with children broke down. ASD-Chat study

Sing and drum

Buddy says a two-part word you choose ("Rain-bow") in a sing-song way, the first part on a high note and the second on a low one, with a drum hit on each. Then: "Your turn! Drum and sing!" Only drum hits and voice sounds are counted, so the word is never judged, and drumming alone is a full turn.

Why: Inspired by Auditory-Motor Mapping Training, the one music approach we found with randomized-trial evidence for minimally verbal autistic children (in a 15-child trial, d = 0.82 on syllable accuracy). This is play, not the therapy, and in that trial the children with the most severe impairment gained least, so we make no promises. PubMed Central

Rhythm echo

Buddy plays a short beat; your child plays it back by drum, claps or snaps. For now a match means the same number of hits; timing isn't compared. Five matches in a row earn a sound, and every 10 a bigger one. Buddy is told never to call a miss wrong; the count just starts over.

Why: Clap-back rhythm games are a staple of listening curricula for blind children, and no product we found counted a child's hits or snaps. WonderBaby

Name That Sound

Buddy plays a sound and asks, "Was that a cow?" About one time in three it names a different sound on purpose, so "no" is sometimes right. Sixteen built-in sounds, from an owl to a train whistle, plus sounds your family records. One-minute rounds.

Why: A yes-or-no version of the sound-search games built for blind students, with their one-minute rounds and streak sounds. Perkins: Blindfold Sound Search · Perkins: ObjectiveEd apps

Choose-your-path stories

Buddy is told to tell original adventures in three to five chapters, each ending in a yes-or-no choice. Something picked up for the backpack early on matters later. There's no wrong path, only different endings, and an unfinished story can carry on next time.

Why: Interactive story players now use chapter-end choices (Yoto) and a backpack of items picked up along the way (Lunii). Yoto · Lunii

Your child's own winning sound

Celebrations are sounds, not just pictures: a drum fill, a crowd roar, a guitar riff, a drum-and-fart combo, or a sparkle and fanfare. You pick, or let your child choose by yes and no. Buddy is told to keep praise short, specific and earned, never gushing.

Why: Age-respectful AAC guidance, expert reviewers noting that explicit praise can stress some students, and a speech app whose rewards were all visual, leaving out kids with low vision. PrAACtical AAC: age-respectful AAC · Echo-Teddy study · Educational App Store

Jokes, riddles, games, and songs

Silly jokes and sound effects, riddles with a drumroll, a sitting-down body-parts game, animal guessing, colors, simple adding and holiday fun. Songs come from your own music library, played for real, with an invitation to play along on drum or ukulele.

Why: The research says to build play from a child's real interests. Buddy is told never to sing or recite song lyrics; the app plays the real song instead.

Family voices, and a joke book for offline moments

Family members can record a hello for the start of a session and a "nice one!" that Buddy plays after something your child really did (it's told to say whose voice it was). If the internet or the AI can't be reached twice in a row, Buddy keeps playing from its own joke and riddle book and your downloaded songs.

Why: Parent-recorded audio is standard on story players like Yoto and tonies. One companion robot stopped working (except through a community-run server) when its maker shut its servers down in January 2025. Peggy O'Mara · Six Degrees of Robotics

Teach a new helper your child's signals

A three-minute guided practice: Buddy coaches a grandparent, sitter, substitute or aide to ask one yes-or-no question, wait, and read one snap or two. Then they practice alone while the app scores the adult's reading, never the child's. You get a one-page card and a 20-second audio clip to share.

Why: In one study, 100% of parents of children using high-tech AAC (speech devices and apps) ranked engagement as a top-three barrier, and 88% described stigma from others; the study's remedy was training the people around the child. In an Iowa survey, families with more training used the device in more places. Frontiers in Psychiatry · Univ. of Iowa

Play together with a sibling or classmate

Pick someone from your child's people list who's playing along today, and Buddy runs joke and riddle rounds for both. Buddy's rules: your child goes first, the partner waits for your child's answer, and it's never a contest. The partner's answers are in the transcript but never count as your child's.

The microphone can't tell two voices apart, so while someone plays along your child answers by snap, switch or on-screen buttons (or by voice, if snaps are off).

Why: Expert reviewers of an AI plush toy recommended using it to encourage play with peers rather than one-on-one bonding, and Perkins advises clear "your turn" cues for blind kids. Echo-Teddy study · Perkins: adapted games

Ask for an activity without interrupting

A "Hold to choose" control on your child's screen opens a row of buttons: a song, jokes, a story, or sing and drum. One tap asks Buddy for it next, without talking over your child or ending their turn.

Why: PIRG called a parent's ability to steer an AI toy the most useful control it found. PIRG

Buddy in a loved one's voice

With Apple's Personal Voice, Buddy can speak in a family member's synthesized voice, but only after you confirm that person agreed. Each session says so up front ("I'm still Buddy!"), Buddy is told never to speak as that person, and the voice stays with iOS.

Hearing a loved one's voice say things they never said could still confuse some children. Watch the first time, and turn it off if your child seems to think that person is really there.

Why: Recent iOS versions opened Personal Voice to other apps, and we think a familiar voice may make Buddy easier to follow. Consent comes first. 9to5Mac

Your child's people

List family, teachers, aides and therapists by the names your child uses. Buddy is told to speak warmly about them, never to pretend to be them, and at goodbye to send your child to whoever is with them today.

Why: The research's central idea: the best version of an AI buddy points a child back toward people.

Record sounds from your child's life

Record the dog barking, a doorbell, a sibling's whistle. Give each a name and a note about when to play it, and Buddy is told to weave one or two into games and stories. The recordings are stored on your device and never sent to the AI; only their names and notes go into Buddy's instructions, which are sent to Anthropic's Claude API to run each session.

Why: Familiar sounds can mean a lot to kids who experience the world by ear.

Add from a document: IEPs, reports, notes

Documents you can try: an IEP or other school plan, a therapy report, a care plan, a teacher's email or your own notes, as a PDF, Word file, page photos or pasted text. The words are read on your device first. When you tap Send, the full text (up to about 60 pages) goes once to Anthropic's Claude API, including any names, birth dates, scores or medical details in it. To send less, paste only the parts about interests, communication and what helps. Each suggestion shows the words it came from, and nothing is added until you approve it.

Why: Families have already written down their child's needs many times. The AI is told to leave scores, medical history, last names and school names out of its suggestions; that shapes what comes back, not what's sent. How document import works

About your child, in your own words

What your child loves, their pets, the games they like, what they know, topics to avoid. Grown-ups write it, can read or change it any time, and Buddy reads it at the start of every session.

Why: PIRG called a parent-written prompt the most useful AI-toy control it found. PIRG

Tell Buddy what your child needs

A short checklist covers seeing, hearing, talking, learning and attention, moving, and sensitivities, plus a box for anything else. Each tick adds a rule for how Buddy talks (for a child sensitive to sudden noises, saying which sound is coming next), and Buddy is told never to mention any of it to your child. The app also suggests matching settings, and changes nothing until you approve.

Why: Dignity. Accommodations should shape the play without being announced to the child.

A voice your child can follow

Buddy speaks slowly by default, with a pause between sentences. You can change the speed, pitch, pauses and voice, and even Buddy's name. If the device only has a robotic-sounding voice, the app shows how to download Apple's free higher-quality voices.

Why: Kids who need extra time to process speech shouldn't have to keep up with a fast talker.

An answer log for your child's therapist

Every question gets one line: how the answer came (snap, spoken word, switch, tap, voice try, unclear, or no answer), seconds to the first sign of an answer, and whether a cue came first. Weekly charts show the share of spoken answers, the usual time to answer, and answers before any cue. One button exports it all as a spreadsheet file (CSV).

The app can't tell whose snap or short "yes" it hears, so nearby adults and classmates should stay quiet and hands-off while your child answers. Answers someone physically helped with aren't marked as such.

Why: A University of Delaware critique of one AAC logging service found the child's and the clinician's selections "can't be told apart." Here, Buddy's own words appear only as the question, so Buddy's modeling is never logged as an answer. University of Delaware

Notes and transcripts after every session

Read any session word for word, see "how it went," and share a plain-text copy. After a session of three or more turns, the transcript is sent once to the AI to write a few short notes on what your child said yes and no to. It's told to state only what the transcript shows; still, treat the notes as a summary and the answer log as the data. You can add notes and delete any.

Why: PIRG found one AI toy's usage summaries badly wrong (over an hour of testing reported as 19 minutes), while another gives parents full transcripts. PIRG

Built-in limits and easy grown-up control

Sessions last about 20 minutes by default (you can choose 5 to 45), and end early after five missed answers in a row or five quiet minutes. Dim hold-to-stop and hold-to-skip controls on your child's screen let a grown-up step in without startling them.

Why: PIRG found AI toys' safety rules weakening as conversations ran on, and noted one toy had no time limit or remote shutoff. PIRG

Profiles for several students

Each student with their own signals, interests, Buddy voice and language, and quick switching on one device. Today Ninja Buddy is set up for one child per device, in US English. This is planned for the classroom launch.

Why: In an Iowa parent survey, whether a child's device matched the family's home language predicted daily versus monthly use. Univ. of Iowa

Songs and stories that pause for "more"

Pause a favorite song mid-chorus and wait for a "more" signal, building toward more, stop and different. It's waiting until that new signal can be defined with a speech therapist, rather than invented by the app.

Why: A pause in something a child loves is a classic teaching moment, and nothing we found uses it. But a new signal belongs to the child's team.

Choose your character

Let each child pick their buddy's look: a different ninja, or a different character altogether, as part of their own profile.

Why: Some kids' AI products, like Heeyo and BubblePal, let children pick a character, and it fits naturally with student profiles. AIPure: Heeyo guide