The Short Answer
Voice is the part people judge first, and the part that is hardest to fake.
likeness of the professional voice clone, on Done-For-You builds
languages supported
when a Done-For-You build is typically live
The professional voice clone at 97% likeness is part of the Done-For-You service, and only Done-For-You. On the free tier and on Pro, your clone is text. That is worth settling before you read another word, because it is the single thing most often described wrongly by everyone except us.
On a Done-For-You build the voice is a live conversation. Your client speaks, the clone answers, back and forth. It is not a text-to-speech readout of a chat reply with a play button on it.
The honest version
A number cannot settle whether a voice sounds like you. Ears can.
What 97% Likeness Means
And, more usefully, the three things it does not mean.
97% likeness is the target we build to on a Done-For-You clone, and it is measured rather than asserted. Most platforms quote a likeness number and never say where it came from. Here is exactly how ours is produced, so you can hold us to it.
Source capture and conditioning
Clean reference audio is gathered across multiple sessions rather than one sitting, because a voice recorded on a single day is not a representative sample of how you actually sound. Audio is normalised and resampled to the rate the speaker model expects before anything is measured.
Accent and dialect calibration
Regional accent, dialect and cadence are tuned deliberately, not left to the model. This is the step that most often separates a clone that is technically close from one your own clients accept, and it is the reason a generic voice tool tends to flatten a distinctive accent toward the middle.
Your own baseline, measured first
Before the clone is scored at all, real recordings of you from different days are scored against each other. That produces your personal ceiling. Nobody scores a perfect match against themselves, so an honest target is the range your own voice occupies day to day, not an abstract 100%.
Speaker embedding comparison
The clone and the real recordings are each reduced to a 192-dimension voiceprint using an ECAPA-TDNN speaker-verification model, the same class of model used for speaker identification research, and compared by cosine similarity. This is the headline measurement and it is reproducible by anyone with the same public model.
Pitch stability analysis
Fundamental frequency is tracked across the whole sample to catch the failure that similarity scoring alone will miss: a clone that lands in your register on average while wandering high and deep within a sentence. Median pitch and spread are both checked, because a good average can hide a bad delivery.
Second-model cross-check and batch pass rate
The comparison is repeated against a second, independently trained speaker model, and then run across a batch of matched line-for-line pairs rather than one flattering clip. What matters is the pass rate across the set, which is what stops a single lucky sample being mistaken for a finished clone.
Iterate until it clears the bar
If the batch does not clear the baseline, the voice is retrained and rescored. The measurement is not a certificate issued at the end, it is the thing that decides when the build is actually done.
That is the process. Now take the three qualifiers that go with the number seriously.
It does not mean nobody could ever tell. Somebody who has known your voice for twenty years is a harder audience than somebody who has watched three of your videos. Both are real listeners, and no single figure describes both of them.
It does not describe the answers. Likeness is about sound. Whether the clone says the right thing comes from what it was trained on, which on a Done-For-You build is your books, courses, PDFs and video, curated for the build. Two different problems, solved in two different places.
If you want the wider version of this question, how accurate an AI clone is covers accuracy of content as well as sound.
What Actually Convinces People
A career coach documented her own build on camera. Watch what happens to the pronoun.
Lucy Gilmour is a career coach. She built a Personify clone, filmed the process, and published 25 minutes of it on YouTube. Early on, the clone is a joke with a name.
"We're calling it fake Lucy."
Then her son wandered in while she was testing it.
"I was like, 'Do you want to talk to fake mommy?' ... he was like asking math questions like, 'What's 100 plus two?' And she was going, 'It's 102.'"
Lucy Gilmour, in her build video. Note the pronoun.
Inside a single anecdote the clone goes from "it" to "she", and she does not stop to mark the change. That is the most telling thirty seconds in the whole video, and nobody scripted it.
Her one-to-one clients had early access and spent about two months testing it before launch. This is what they told her.
"They say it's like talking to you, Lucy. I could talk to her for hours. And actually, some of them do."
Lucy Gilmour, reporting client feedback in her build video. One session ran to three hours.
Three hours is not politeness. Nobody spends an afternoon with something they are only testing as a favour. The full case study has the usage numbers behind it.
Judge It By Ear
What that footage proves, and what it does not. We would rather you heard this from us.
We could quote Lucy at more length. She says she is blown away by how good it is, and that every time she opens it she is blown away again. Her clients told her something similar. None of that is going in this page as proof of voice fidelity, because it is not proof of voice fidelity.
What those quotes are evidence of. Authenticity and engagement, from people who know the real Lucy and had two months to get bored of the clone. Feedback that survives two months of daily use is a harder test than most testimonials pass.
What they are not evidence of. A likeness measurement. When Lucy praises the build she names "the interview side", which is content rather than sound. One line, "I tested it last night, it sounds amazing", plausibly refers to the voice, and even that is ambiguous in context.
What would settle it. Your own ears. Voice fidelity is the one claim in this category that a listener can check in about ten seconds and no amount of copy can substitute for, which is why we would rather put you in front of one than write another paragraph about it.
Where to start. The demo at the top of this page shows you what a clone learns from your website in around thirty seconds. That preview is text, like every self-serve clone. Hearing the voice is part of a Done-For-You conversation, which is scoped on a discovery call.
Conversation, Not A Readout
Two quite different products both get sold as voice.
A lot of tools describe voice as text-to-speech: the model writes an answer, then a synthetic voice reads the answer out. That is a spoken transcript of a chat. It can sound perfectly pleasant and still feel like being read to.
A Done-For-You Personify build includes live voice conversation instead. Your client talks, the clone answers, and the exchange runs like a call. That is also what makes a three-hour session possible at all, because nobody listens to a readout for three hours.
It changes what the voice has to be good at, too. A readout only has to sound pleasant. A conversation has to sound like you. Voice cloning in coaching goes further into the category.
Where Voice Is Available
Stated plainly, because this is the answer most often got wrong about us.
| Free | Pro | Done-For-You | |
|---|---|---|---|
| How the clone talks | Text | Text | Text and live voice conversation |
| Professional voice clone | Not included | Not included | Included, at 97% likeness |
| How you get it | Start free | $39/mo, or $29/mo billed annually | Scoped on a discovery call |
That is the whole rule, and it does not have an exception. If you have seen Personify described as doing voice on a self-serve plan, the thing being described is a Done-For-You build. The pricing page sets out the self-serve tiers in full, and build it yourself versus Done-For-You is the honest version of which one you actually need.
How A Build Works
Scoped on a call, live around day 14.
A discovery call
Done-For-You is scoped on a discovery call, because what your clone needs depends on what you already have.
Curated training
Your books, courses, PDFs and video, curated for the build rather than tipped into a folder and hoped over.
Voice and interface
The professional voice clone at 97% likeness, and a branded interface that looks like your brand rather than ours.
Live around day 14
A success manager stays with you, and Done-For-You builds carry no Personify commission on what you charge for access.
Lucy filmed her own scoping call, so you can hear how the voice part gets described before anything is built rather than after.
"we would be cloning your voice so it sounds exactly like you. So we would also match your tone, your teaching styles."
Personify supports 100+ languages. Done-For-You customers include Prof. David Clutterbuck, co-founder of the EMCC in 1992 and widely described as the world's most published author on coaching and mentoring, with more than 80 books and 900 articles across 40 years. He launched his clone, Ask David, on stage at the EMCC Global Annual Conference in Zagreb in May 2026.
Frequently Asked Questions
Does Personify do voice cloning on the free tier or on Pro?+
No. Clones on the free tier and on Pro are text. The professional voice clone at 97% likeness is part of the Done-For-You service only, which is scoped on a discovery call. If you have read somewhere that Personify does voice on a self-serve plan, that description belongs to a Done-For-You build.
What does 97% likeness actually mean?+
It is the likeness figure Personify quotes for the professional voice clone on a Done-For-You build. It is a claim about how closely the cloned voice matches yours, not a claim about whether the answers are right. Answer quality comes from the training material, which is a separate piece of the build. And no number decides this for a listener, which is why the honest recommendation is to hear one before you buy on the figure.
Is the voice just a text-to-speech readout of the chat replies?+
No. A Done-For-You build includes live voice conversation, not a text-to-speech readout of chat replies. Your client speaks and the clone answers, and the exchange runs like a call rather than like a page being read aloud.
Is there proof that a clone sounds like the real person?+
The strongest proof is hearing one, and we say so rather than dressing up testimonials as measurement. What real users report is authenticity and engagement. Career coach Lucy Gilmour gave her one-to-one clients two months with her clone before launch, and their feedback to her was "it’s like talking to you, Lucy. I could talk to her for hours." One session ran to three hours. That is evidence people found it worth talking to, not a fidelity score.
How many languages does Personify support?+
Personify supports 100+ languages.
What is included in a Done-For-You build, and how long does it take?+
Curated training across your books, courses, PDFs and video, a professional voice clone at 97% likeness, live voice conversation, a branded interface and a success manager. It is scoped on a discovery call and is typically live around day 14. Done-For-You builds carry no Personify commission on what you charge for access.
Hear it before you take the number on trust
The professional voice clone lives in Done-For-You, and Done-For-You is scoped on a discovery call. If you want to start with a text clone today and see what yours knows, the free tier is open.
Free tier available. No credit card needed.