AI voice profiles document how you write but your voice is a decision engine profiles can't hold. Here's what to build instead, from the creator of a voice profile tool.
When you said “a voice profile is a mirror that shows you yourself by showing you what you’re not,” that crystallized things in a way that my own response did not.
I love that you took the responses and made tweaks to your service!
Your brilliance shines through in every language. 🙂
I really respect your willingness to pivot and say "this didn't work exactly how I thought it would" and then reshape it. I think that's much more valuable than expecting to get everything perfectly right the first time.
Mia, this is so good!!! AI voice profiles being limited is both fascinating and on-brand. Of course the machine can summarize our voice but then gets soo confused the second we contain more than one mood, one opinion or one tiny emotional plot twist.🤣
God forbid a woman have range.
Thank you for doing the testing and bringing us the receipts, as usual. 🩷🦩
You described perfectly why no "AI voice" tool has ever worked for me but didn't really know why. Not that I want AI to write for me anyway. But at least know I can go ahead and make a new "voice profile" and then use it like you said, as quality gate or a starting point if I'm really stuck.
Didn't know you were fluent in that much languages. I only have English, Spanish, and most Portuguese under my belt. 😅
Looking forward. I noticed a main issue is they change my usual syntax and make it very general. Additionally, they use filler words and often make me sound more brusque than I am. I cheery most of the time, yet they write as if I have an attitude problem, which I instantly correct. It often forgets I enjoy a good laugh.
So keen to give this a whirl, Mia! I actually get so frustrated with the whole "voice skill" promise that many people make. I think a hugely important first step is to figure out your own writing voice first, before AI goes anywhere near it. Get comfy with it, know the shape and feel of it, own it. Only then will you know if AI is smoothing you out and you'll know what wrong feels like. I cringe every time I hear someone swear by a voice skill that is so great it only requires them to give a simple topic paragraph and a perfect post spits out that requires minimal editing - sure dude.
(Non)favourite editing suggestion from Claude is "cut this sentence, it doesn't sound like Dallas at all". It is usually my favourite, one that is classic me and Claude is telling me it knows better, that I would never say that... just did Claudio!
I'm super ashamed to say this but I had 2 editing skills that VERY HEAVILY contradicted each other and I didn't realized it until sooooo much later... They had Claude fight me like "yhis is NOT your voice Mia", I was like what? It is, that's why I want to say this... 🤣🤣
Sometimes I feel like calling old little Mia and telling her how stupid some situations are in 2026 haha
Haha isn't that a crazy thought that this time last year there were no voice skills, we had to build that context every time, now we're like "voice skills are okay but can be problematic", what will 2027 us be saying?! 🫣😅
Yeah, I've learned that if I feel like I'm fighting too much with Claude over something, I can usually blame a skill or project knowledge I forgot about 😂🤦🏻♀️
mine is a work in progress, changing over time. I noticed that sometimes when the model changes, I have to adjust or make additions. Totally agree with you on the importance of the "what you're not" parts. It also depends on what you're using it for, one type of writing may be good but try something else and it's got weird tics. same profile, different application, different results.
Exactly! People usually build one voice profile and then smaller ones for the different use cases. And 100% with you on the model changes, this is a BIG one!
Loved it! Can’t wait to implement these ideas in my workflow.
One thing I forgot to mention and captured in your article clearly:
After I first used my AI profile, it was off much like everyone else has said. So I went over to Claude. I ran a comparison between the draft and another article that I’ve written. I asked how was the first article different. It was interesting what it came back with. Claude said:
The AI draft reads like someone who has read Michael’s voice profile carefully. The Darwin piece reads like Michael. The gap is authenticity vs. construction.
To close it: more personal risk, more friction in the rhythm, and at least one moment where Michael admits he got something wrong.
The bones are right. The marrow isn’t there yet.
That last piece is critical. You need to continue being the human in the loop and make it your writing.
Thank you Michael, one of your comments on Substack actually inspired this piece!
All of these things and observations are why I think the "human" part of this has to stay a little messy. Rough edges are NOT bugs, ever. They're proof that we're still making the decisions.
I like your approach cause it treats the tool as something to work against a little 🙏
I do a lot of contrasting in my writing and when I ask AI to help me edit my original blob of text mess, AI keeps putting AI things--AI phrasing. It's soo much, I can't stop seeing it--even if it make sense. Example, "that's the real shift" "...nobody thinks about..." "...most people..." < I didn't say that, I mean that but...that's not me uh🤯
I think these are some very great tips, thank you Mia again for another great drop.
I am so glad you said AI will never 'sound' identical in the writing because it can't. I used your voice profile and built a new file just a month ago, and when it asked me to share my writing, I gave the files and it said, this is talking, lol. Quite frankly it was a good experience going through it to audit what good enough is for me. The rough drafts dictated to AI do the best in my case.
It's hard to reproduce the writing anyway as we sound different when we text friends and when we send proposals to clients and we read tons of stuff that influences how our language evolves. I think at some point I had both professional and personal voice files but gave that up and consolidated it. Will check out your updated tool, thanks for this article, Mia!
Thanks Anna. I’ll probably publish what my results are. I definitely not a code bro, so I am much more interested in what ‘configuration’ I can do to get better results. Also learning so I can share my Enthusiastic Skepticism with everyone.
So have you tried any of the techniques that seem to be flooding everywhere to use specific skills and projects and a modified Skills.mD master that says they must ask which to apply? That’s what Jenny O and Jenny H and obvs Mia have been describing. I was about to spend time this weekend attempting to experiment with the techniques but wondered what you’d tried and not.
Yes, to some degree, I don't think my voice has two separate modes so that doesn't work for me but might work for you. I'd say try it and see if it suits your work.
I am not at the point where my life is fully automated (and not looking to be TBH) and has been using other tools with a memory layer to follow the context and respond to emails. The rest can be done by me faster than reviewing Claude's output.
Anna, that’s fascinating because I strongly agree. I think one of the mistakes people make is that they write in different modes depending on subject, context, audience etc. I like many writers can write in a loose way that is almost conversational, and yet in a professional mode know how to write like a Sales and Marketing professional. But I also know how to write like a healthcare clinical professional.
I think part of the flaw with the whole voice print approach is that mode switching. If you write/feed a mixed portfolio to the voiceprint builder do you just confuse the output? Or if you only feed it Substack article, will it only ever write in your Substack voice.
I was just reading Jenny Ouyang and Jenny Hong articles on exactly this topic this weekend and was fascinated by the very different writing styles they both have. (I love that diversity even in a topic like using AI tooling here in Substack!)
I’d love to hear what the failure mode or friction you encountered in trying to have a switchable voice print/two channel approach?
Hi Stuart, thanks for reading my comment. That’s exactly what happened in my case. Claude assumed my writing style applied to all tasks, so when I asked it to respond to clients or write cold outreach emails, it used the same tone I use for Substack: very reflective and embodied.
I tried separating those use cases, but the output was still off. At one point it generated something like, “Hi there, as I’m sitting here my hands went cold before writing this email,” which was hilarious but not useful at all for the task.
I realized that giving it very literal and personal material pushed it too far in that direction.
Thank you as ever Mia for your writing. I LOVE your reference to ‘dor’ and I recognize that in my own writing. It’s my leaning back on my Scottish heritage, to my classical education in the 1970’s and mostly in my love of folk songs, stories and the sheer act of connecting with people through the act of storytelling.
It’s fascinating your own article coinciding with my own description of voice last week, using my ‘Little Mermaid’ analogy and linking it to my Competence Framework. I love the way you have now woven the evolution and development of your own Voice tooling as well as the recognition of the limitations of the current models and tooling to accurately reflect US, the information we are looking to share and the stories we tell.
Stuart, “dor” is one of my favorite words and there's no English equivalent that carries the same weight and that’s what makes it (and other words like it) so special. I'm so happy I get to share it with the world! ❤️
Just went through your Substack and found the mermaid post :) Bookmarked it to read, can’t wait! And thank you for your kind words!
This was really great, Mia! It took my understanding to new depths.
I think AI has been oversold in what it can do in some contexts (writing for us and having it be just like what we’d have written) and I’m finding I am continually having to update my expectations. I’ve been doing that from a place of what I call an argument with reality. Meaning, I get really frustrated with AI and throw my hands in the air and say, “Whatever! I guess it can’t do X either. FINE!” All very sulky.
However, this whole process, from doing my original voice builder (and trying to use it) to answering your chat question to reading this article, have given me so much more understanding of what and why I’ve gotten the subpar results that I have. And that has allowed me to settle into reality rather than argue against it.
Since the chat last week, I’ve been reacting much differently to my working sessions - more like “OH! I get it. I’m expecting too much right now.”
As for the updated voice builder, I know what I’ll be doing this weekend!!!
Christin this makes me so happy to read!! I feel like we all go through that sulky phase with AI at some point 😂 So glad the pieces are clicking together now. Have the best time with the voice builder this weekend, can't wait to hear how it goes! 🐕🐈
yeah it’s pretty fun sometimes 🤣 Mostly Romanian, English, Spanish. And with a hint of Portuguese when I’m feeling theatrical and I wanna practice it haha!
This was really great, Mia! It took my understanding to new depths.
I think AI has been oversold in what it can do in some contexts (writing for us and having it be just like what we’d have written) and I’m finding I am continually having to update my expectations. I’ve been doing that from a place of what I call an argument with reality. Meaning, I get really frustrated with AI and throw my hands in the air and say, “Whatever! I guess it can’t do X either. FINE!” All very sulky.
However, this whole process, from doing my original voice builder (and trying to use it) to answering your chat question to reading this article, have given me so much more understanding of what and why I’ve gotten the subpar results that I have. And that has allowed me to settle into reality rather than argue against it.
Since the chat last week, I’ve been reacting much differently to my working sessions - more like “OH! I get it. I’m expecting too much right now.”
As for the updated voice builder, I know what I’ll be doing this weekend!!!
Phenomenal writeup as always, Mia!
When you said “a voice profile is a mirror that shows you yourself by showing you what you’re not,” that crystallized things in a way that my own response did not.
I love that you took the responses and made tweaks to your service!
Your brilliance shines through in every language. 🙂
Ryan!! Thank you so much for the kind words :) and thank you for sharing your experience 💕
I really respect your willingness to pivot and say "this didn't work exactly how I thought it would" and then reshape it. I think that's much more valuable than expecting to get everything perfectly right the first time.
Thank you Ben, appreciate you! 🙏
Mia, this is so good!!! AI voice profiles being limited is both fascinating and on-brand. Of course the machine can summarize our voice but then gets soo confused the second we contain more than one mood, one opinion or one tiny emotional plot twist.🤣
God forbid a woman have range.
Thank you for doing the testing and bringing us the receipts, as usual. 🩷🦩
Hahahahah girl, I read this as "God forbid a woman have rage" and was about to encourage you to have all the rage 🤣
Thank you Pinkie ❤️
Excellent post!
You described perfectly why no "AI voice" tool has ever worked for me but didn't really know why. Not that I want AI to write for me anyway. But at least know I can go ahead and make a new "voice profile" and then use it like you said, as quality gate or a starting point if I'm really stuck.
Didn't know you were fluent in that much languages. I only have English, Spanish, and most Portuguese under my belt. 😅
That's just one less, my friend. Hahahah 🤣❤️
Ahhhh dammit! 😆😆
Looking forward. I noticed a main issue is they change my usual syntax and make it very general. Additionally, they use filler words and often make me sound more brusque than I am. I cheery most of the time, yet they write as if I have an attitude problem, which I instantly correct. It often forgets I enjoy a good laugh.
Oh yes, basically it removes ALL personality and when it tries to add it in, it's fake!
So keen to give this a whirl, Mia! I actually get so frustrated with the whole "voice skill" promise that many people make. I think a hugely important first step is to figure out your own writing voice first, before AI goes anywhere near it. Get comfy with it, know the shape and feel of it, own it. Only then will you know if AI is smoothing you out and you'll know what wrong feels like. I cringe every time I hear someone swear by a voice skill that is so great it only requires them to give a simple topic paragraph and a perfect post spits out that requires minimal editing - sure dude.
(Non)favourite editing suggestion from Claude is "cut this sentence, it doesn't sound like Dallas at all". It is usually my favourite, one that is classic me and Claude is telling me it knows better, that I would never say that... just did Claudio!
I'm super ashamed to say this but I had 2 editing skills that VERY HEAVILY contradicted each other and I didn't realized it until sooooo much later... They had Claude fight me like "yhis is NOT your voice Mia", I was like what? It is, that's why I want to say this... 🤣🤣
Sometimes I feel like calling old little Mia and telling her how stupid some situations are in 2026 haha
Haha isn't that a crazy thought that this time last year there were no voice skills, we had to build that context every time, now we're like "voice skills are okay but can be problematic", what will 2027 us be saying?! 🫣😅
Yeah, I've learned that if I feel like I'm fighting too much with Claude over something, I can usually blame a skill or project knowledge I forgot about 😂🤦🏻♀️
mine is a work in progress, changing over time. I noticed that sometimes when the model changes, I have to adjust or make additions. Totally agree with you on the importance of the "what you're not" parts. It also depends on what you're using it for, one type of writing may be good but try something else and it's got weird tics. same profile, different application, different results.
Exactly! People usually build one voice profile and then smaller ones for the different use cases. And 100% with you on the model changes, this is a BIG one!
Thank you for sharing ❤️
“In Romanian, there’s a word, dor, that carries homesickness, longing, love, and an ache for something that may not exist anymore.”
Such a powerful emotion. “Nostalgia” in Greek. “Saudade” in Portuguese. Universal longing for the loved and lost.
I'd say it's more similar to saudade. Definitely not nostalgia, "dor" is much deeper! 🥰
Cat that barks like a dog -- https://youtu.be/UgIfcmDI9K4?si=PIcHGRMg77x0w24t
🤣🤣🤣🤣
Loved it! Can’t wait to implement these ideas in my workflow.
One thing I forgot to mention and captured in your article clearly:
After I first used my AI profile, it was off much like everyone else has said. So I went over to Claude. I ran a comparison between the draft and another article that I’ve written. I asked how was the first article different. It was interesting what it came back with. Claude said:
The AI draft reads like someone who has read Michael’s voice profile carefully. The Darwin piece reads like Michael. The gap is authenticity vs. construction.
To close it: more personal risk, more friction in the rhythm, and at least one moment where Michael admits he got something wrong.
The bones are right. The marrow isn’t there yet.
That last piece is critical. You need to continue being the human in the loop and make it your writing.
Thank you Michael, one of your comments on Substack actually inspired this piece!
All of these things and observations are why I think the "human" part of this has to stay a little messy. Rough edges are NOT bugs, ever. They're proof that we're still making the decisions.
I like your approach cause it treats the tool as something to work against a little 🙏
I do a lot of contrasting in my writing and when I ask AI to help me edit my original blob of text mess, AI keeps putting AI things--AI phrasing. It's soo much, I can't stop seeing it--even if it make sense. Example, "that's the real shift" "...nobody thinks about..." "...most people..." < I didn't say that, I mean that but...that's not me uh🤯
I think these are some very great tips, thank you Mia again for another great drop.
It ruined writing for us, didn't it? Haha, all jokes aside, don't worry about one-off "expressions" or weird wording.. it's more about the patterns!
And THANK YOU! Really appreciate it 🙏❤️
Yes it got us lol 😂 the patterns do matter 💪🏾
I am so glad you said AI will never 'sound' identical in the writing because it can't. I used your voice profile and built a new file just a month ago, and when it asked me to share my writing, I gave the files and it said, this is talking, lol. Quite frankly it was a good experience going through it to audit what good enough is for me. The rough drafts dictated to AI do the best in my case.
It's hard to reproduce the writing anyway as we sound different when we text friends and when we send proposals to clients and we read tons of stuff that influences how our language evolves. I think at some point I had both professional and personal voice files but gave that up and consolidated it. Will check out your updated tool, thanks for this article, Mia!
That makes me so happy to hear honestly. These tools push us towards more clarity, I don't think they're supposed to be the end all.. 🥰
Also yes to the changing language thing, our voices are ALWAYS a little bit in motion anyway.
Thank you for reading Anna! ❤️
Thanks Anna. I’ll probably publish what my results are. I definitely not a code bro, so I am much more interested in what ‘configuration’ I can do to get better results. Also learning so I can share my Enthusiastic Skepticism with everyone.
So have you tried any of the techniques that seem to be flooding everywhere to use specific skills and projects and a modified Skills.mD master that says they must ask which to apply? That’s what Jenny O and Jenny H and obvs Mia have been describing. I was about to spend time this weekend attempting to experiment with the techniques but wondered what you’d tried and not.
Cheers my friend
Yes, to some degree, I don't think my voice has two separate modes so that doesn't work for me but might work for you. I'd say try it and see if it suits your work.
I am not at the point where my life is fully automated (and not looking to be TBH) and has been using other tools with a memory layer to follow the context and respond to emails. The rest can be done by me faster than reviewing Claude's output.
Anna, that’s fascinating because I strongly agree. I think one of the mistakes people make is that they write in different modes depending on subject, context, audience etc. I like many writers can write in a loose way that is almost conversational, and yet in a professional mode know how to write like a Sales and Marketing professional. But I also know how to write like a healthcare clinical professional.
I think part of the flaw with the whole voice print approach is that mode switching. If you write/feed a mixed portfolio to the voiceprint builder do you just confuse the output? Or if you only feed it Substack article, will it only ever write in your Substack voice.
I was just reading Jenny Ouyang and Jenny Hong articles on exactly this topic this weekend and was fascinated by the very different writing styles they both have. (I love that diversity even in a topic like using AI tooling here in Substack!)
I’d love to hear what the failure mode or friction you encountered in trying to have a switchable voice print/two channel approach?
Hi Stuart, thanks for reading my comment. That’s exactly what happened in my case. Claude assumed my writing style applied to all tasks, so when I asked it to respond to clients or write cold outreach emails, it used the same tone I use for Substack: very reflective and embodied.
I tried separating those use cases, but the output was still off. At one point it generated something like, “Hi there, as I’m sitting here my hands went cold before writing this email,” which was hilarious but not useful at all for the task.
I realized that giving it very literal and personal material pushed it too far in that direction.
Thank you as ever Mia for your writing. I LOVE your reference to ‘dor’ and I recognize that in my own writing. It’s my leaning back on my Scottish heritage, to my classical education in the 1970’s and mostly in my love of folk songs, stories and the sheer act of connecting with people through the act of storytelling.
It’s fascinating your own article coinciding with my own description of voice last week, using my ‘Little Mermaid’ analogy and linking it to my Competence Framework. I love the way you have now woven the evolution and development of your own Voice tooling as well as the recognition of the limitations of the current models and tooling to accurately reflect US, the information we are looking to share and the stories we tell.
As ever KUDOS!
Stuart, “dor” is one of my favorite words and there's no English equivalent that carries the same weight and that’s what makes it (and other words like it) so special. I'm so happy I get to share it with the world! ❤️
Just went through your Substack and found the mermaid post :) Bookmarked it to read, can’t wait! And thank you for your kind words!
This was really great, Mia! It took my understanding to new depths.
I think AI has been oversold in what it can do in some contexts (writing for us and having it be just like what we’d have written) and I’m finding I am continually having to update my expectations. I’ve been doing that from a place of what I call an argument with reality. Meaning, I get really frustrated with AI and throw my hands in the air and say, “Whatever! I guess it can’t do X either. FINE!” All very sulky.
However, this whole process, from doing my original voice builder (and trying to use it) to answering your chat question to reading this article, have given me so much more understanding of what and why I’ve gotten the subpar results that I have. And that has allowed me to settle into reality rather than argue against it.
Since the chat last week, I’ve been reacting much differently to my working sessions - more like “OH! I get it. I’m expecting too much right now.”
As for the updated voice builder, I know what I’ll be doing this weekend!!!
Thanks so much! This has been sooooo helpful.
Christin this makes me so happy to read!! I feel like we all go through that sulky phase with AI at some point 😂 So glad the pieces are clicking together now. Have the best time with the voice builder this weekend, can't wait to hear how it goes! 🐕🐈
You think in four languages!? Wow!
yeah it’s pretty fun sometimes 🤣 Mostly Romanian, English, Spanish. And with a hint of Portuguese when I’m feeling theatrical and I wanna practice it haha!
haha, where did you pick up Portuguese?
Telling you only if you promise not to laugh...
Here you go:
https://www.imdb.com/title/tt0289800/ (watched about 1943975 times, still so close to my heart)
https://www.imdb.com/title/tt0249299/ (watched 2 times only)
And I also had some friends from Portugal but Portugal portuguese is a different beast than the Brazilian one, so I just know the Brazilian! 🤣
Ah, I see it now — it's like how I pick up words in Turkish, haha (giving you arsenal but don't laugh back, okay?)
mhm, nobody's laughing here..... 👀
This was really great, Mia! It took my understanding to new depths.
I think AI has been oversold in what it can do in some contexts (writing for us and having it be just like what we’d have written) and I’m finding I am continually having to update my expectations. I’ve been doing that from a place of what I call an argument with reality. Meaning, I get really frustrated with AI and throw my hands in the air and say, “Whatever! I guess it can’t do X either. FINE!” All very sulky.
However, this whole process, from doing my original voice builder (and trying to use it) to answering your chat question to reading this article, have given me so much more understanding of what and why I’ve gotten the subpar results that I have. And that has allowed me to settle into reality rather than argue against it.
Since the chat last week, I’ve been reacting much differently to my working sessions - more like “OH! I get it. I’m expecting too much right now.”
As for the updated voice builder, I know what I’ll be doing this weekend!!!
Thanks so much! This has been sooooo helpful.