Since launching our AI vocal coach in November 2025, singers have typed 14,091 messages to it. Questions, complaints, confessions, jokes, and at least two apologies to a piece of software. The messages come from 1,339 different people practicing in bedrooms, cars, and kitchens across every day of the week and, judging by the languages, most of the world’s timezones.
We read them. (Anonymized, in aggregate — see the method note.) Together they form something we didn’t quite expect: an unfiltered picture of what learning to sing actually looks like in 2026, written by the learners themselves, mid-practice, when nobody polished anything for an audience.
Here is what they told us.
What we found: The three biggest struggles are pitch accuracy (256 singers), vocal range (245), and navigating registers (139). Only 19% of messages are questions. The rest is negotiation, pushback, and dialogue. Singers don’t passively receive coaching; they customize it, argue with it, and use it to rehearse ambitions they’d never voice to a human teacher. Practice volume is flat across all seven days of the week: singing practice in 2026 isn’t a weekend hobby, it’s woven into daily life.
What do amateur singers struggle with most?
Strip away the setup questions and the small talk, and three struggles dominate everything else.
1. Staying in tune. About one in five singers — 256 of the 1,339 — brought up pitch: singing flat, singing sharp, not knowing which. It is the most-discussed problem in the corpus, and it converges with what we found in a separate analysis of 632,000 missed notes: when amateur singers miss, they overwhelmingly miss flat, and mostly on high notes. The singers, it turns out, know exactly where it hurts. One asked a question we found so good we’re writing a whole article about it: “Why does it seem to be harder to hit the pitch when singing soft?”
2. Range and high notes. Nearly as many singers (245) asked about their vocal range, and 182 specifically about singing higher. Not “how do I become Mariah Carey” — much more modest and more poignant than that: “I really can’t sing that high yet.” Note the yet.
3. Registers. 139 singers wrestled with head voice, chest voice, falsetto, and the mysterious “mix” — the vocabulary problem and the physical problem tangled together: “I think I am singing only in head voice, isn’t that too high?”
If you teach singing, none of this will surprise you. What might: how precisely the self-reported struggles match the acoustic data. People asking about flatness are the same population measurably going flat. Amateur singers are not deluded about their voices. They just haven’t been given the tools to fix what they can already hear.
Do singers actually listen to an AI vocal coach?
Here’s a number that changed how we think about the product: only 19% of the messages are questions. The rest is dialogue — reports, pushback, bargaining.
136 singers actively negotiated the difficulty of their session. Downward, often: “still too high.” “Slow it down by 60%.” But also, gloriously, upward: “This is way too easy and quite boring to be honest.”
One singer, unhappy with an exercise choice, delivered what may be the corpus’s finest review: “If you were a human coach, I would say: you are drunk.”
We love this, sincerely. A practice session where the student pushes back is a practice session the student owns. In our seven-mox§nth outcomes data, the singers who talked more with the coach practiced more and improved more. The negotiation isn’t noise around the learning. It might be the learning.
Are people prompt-engineering their singing lessons?
Buried in the longest messages, we found something we did not anticipate: singers treating their vocal coach the way power users treat ChatGPT.
Some paste in fully-written system prompts: “You are an advanced AI vocal coach. Your role is to run a strict, real-time vocal training session focused on improving first-note accuracy, chord resistance, pitch locking, and onset precision…” — and the coach, gamely, complies. Others paste sheet music notation and ask for a breakdown. One user arrived with a complete self-designed 30-minute lesson plan and instructed the coach to run it.
Two years ago, “customizing your singing teacher” was not a sentence. Now it’s a Tuesday. The line between taking a lesson and designing one is dissolving, and the most motivated learners crossed it without asking permission.
The Freddie Mercury clause
Then there is the category we’ve come to think of as rehearsing the dream:
“Okay, let’s pretend I was Freddie Mercury and was singing Bohemian Rhapsody to you — what would you comment on it?”
“What would you tell me of my voice if I was Adele?”
Nobody asks a human teacher this. It’s too exposed. But alone with an AI at 11pm, people audition their fantasy selves — and then, having survived it, go back to practicing scales. If you want a one-line theory of why judgment-free practice matters, it’s this: people will attempt in private what they’d never risk in front of a person, and attempting is where learning starts.
What are people actually practicing for?
The ambitions in the corpus are not fame. They are Sunday.
Singers mention their church praise band, their two-part choir (“I am a Soprano, and in three-part show choir I am Soprano 2”), a musical opening in three weeks, and — repeatedly, devotedly — Adele. “I want to be able to sing Hometown Glory by Adele, can we work towards this?” One singer had exactly 50 minutes before choral practice and wanted to warm up.
The picture that emerges is of AI coaching slotting into the existing amateur musical life of choirs, congregations, karaoke nights, and kitchen performances — not replacing it. The dream, for most people, is to stand in the second row on Sunday and quietly nail the part.
When do people practice singing?
We expected practice to cluster on weekends. It doesn’t. Message volume is nearly flat across all seven days — Wednesday and Sunday peak slightly, but every day of the week carries its share. Singing practice, for these 1,339 people, isn’t a hobby scheduled for Saturday. It’s woven into ordinary days, in whatever slot life allows — which, judging by one message, sometimes means “Yes. I just have to get back in the house. I’m in my car.”
This was always the argument for an AI coach: not that it’s better than a great teacher, but that it’s there — on a Tuesday, at 11pm, in the car, 50 minutes before choir.
Overheard in practice
Finally, some messages defy analysis and deserve only to be shared. Reproduced with minor spelling corrections, and our lasting affection:
“Stop for today, my cat wants to sleep.”
“It’s so hard for me, the higher ones close to C5 — I sound like a dying cat.” (Dying dogs are also well represented in the corpus. The metaphor appears to be species-flexible.)
“Stupid song, please.” — frustration and politeness achieving perfect coexistence
“Do you like Hamilton the musical?”
“What’s your name?”
“How am I doing compared to other students?” — some instincts survive every technological revolution
And one that isn’t funny at all, and that we’d frame instead as the entire reason this product exists. A singer, mid-session, working through a note they kept missing:
“The first time, I let myself be confused by a fear of overshooting the fa. The last time, I just sang.”
The last time, I just sang. If 14,000 messages boil down to anything, it’s the distance between those two sentences — and how many people are quietly, daily, in cars and kitchens, crossing it.
Method note: 14,091 messages from 1,339 distinct singers, November 26, 2025 – July 15, 2026, from Singing Carrots’ production database. Messages were analyzed anonymized and in aggregate; theme counts are by distinct singers (one person asking thirty times counts once), using keyword matching. Quotes are individual messages reproduced with minor spelling corrections only; no quote is attributed and no singer’s messages are combined into a profile. Singing Carrots is an online vocal training platform used by over 300,000 singers.
Also read our quantitative analysis of 7 months data of singers practicing with our AI Vocal Coach. How did singers improve along the way?
How many people have used an AI vocal coach?
Singing Carrots’ AI vocal coach has been used by over 2,000 singers since November 2025, generating more than 14,000 in-session messages and 13,000 coaching sessions. The analysis in this article covers 1,339 distinct singers.
What is the most common singing problem for beginners?
Pitch accuracy — specifically singing flat. In our corpus, 256 of 1,339 singers raised pitch as a struggle, and a separate analysis of 632,000 missed notes found 65.7% of errors are flat, concentrated on high notes. Range and high notes (245 singers) and register confusion (139) round out the top three.
Read more
To get an overview of the AI Vocal Coach app market, check out this comparison article: Top 7 AI Vocal Coaches, based on real data.