Writing With AI Without Losing Your Voice
Summary
A nonprofit leader needed to write a highly competitive fellowship application—10,000+ applicants, 20 spots. They had an incredible story but writing isn't their thing. Together, we figured out how to use AI to write faster without losing their unique voice + perspective! Here's what we learned about keeping your writing human when you're using AI tools.
What’s in it for you:
You love using AI for some things but worry everything's starting to sound the same
Writing is hard for you (or someone you're supporting) and you need AI to help—without losing the soul of your message
You want practical steps for training AI to sound like YOU, not like everyone else
You're figuring out how to use AI as a tool that brings forward your unique brilliance instead of replacing it
Helia’s Perspective AND The Story (all in one!)
Here's the thing — I love writing. But I see how hard it is for so many people, and how it becomes this massive barrier to accessing opportunities. And even for those of us who love it, we get stuck. We stare at blank pages. We second-guess every word.
When a nonprofit leader who works with Helia asked us to help them with a fellowship application (not our normal work together!), they were really clear: writing isn't their strength AND they had to compete with thousands of other applicants. The application required deep vulnerability, personal story, lived experience — everything AI supposedly can't do. They needed to move fast, but they also needed it to sound like THEM.
This was our first time helping someone else write something this personal - and I was really excited. I’ve gotten pretty sophisticated at leveraging AI - including for writing (e.g. grants, reports, articles) AND when it doesn’t work/sound right (e.g. social media posts, Helia’s perspective, all personal writing). There's a time and place. I’ve also learned the key that as long as I start from MY language and ideas, it's great (whereas if I start with AI writing, it's a disaster!).
And this time was different. This was about leveraging AI to make writing accessible for someone else, especially someone that doesn’t love writing!!!! I wanted to help them communicate their brilliance without the writing barrier getting in the way.
What we discovered: Leveraging AI for writing doesn’t mean you have to decide between your voice and getting words on the page. But, if you want your authentic voice to come through, you have to be really intentional about how you use AI. Here's the system we built.
(Want to skip the story and just do it? Jump to the Quick Guide. And if you want to see how we figured this out and what we learned along the way, keep reading.)
What this looks like in practice
-
Before we dig into the specific steps, here's something important: As ALWAYS, there's more than one way to make this work!!!!!
Path A: Build the system first, then write
Create a comprehensive style guide upfront (4-5 hours)
Then use it to draft (1-2 hours per project)
Best for: Multiple projects ahead, organizational writing, want consistency from day one
Path B: Write first, build the system after
Jump into drafting with voice materials (1-2 hours)
Create style guide after your first project (adds 2-3 hours)
Best for: One important thing to write, want to try this before committing, learning as you go
We are going to walk through Path A (what we did with Person Y!) because they had a high-stakes application and we knew we'd work together on future projects. Creating the comprehensive style guide upfront was worth it.
What most people might want to do: Start with Path B. Do one project successfully, see if this works for you, THEN invest in creating the style guide.
The guides below give you both paths. Pick what fits where you are right now.
-
Before AI writes anything, you need to teach it how this person actually sounds. Not how they think they should sound in a formal application, how they actually talk in real life.
What we did: We had SO much to work with. Months of transcripts from our conversations, interviews they'd done, conversations with community members – so much of their own thinking was already captured in recordings. We uploaded everything to the project chat and asked AI to analyze it all and describe their voice back to us.
The first couple rounds were NOT RIGHT. The responses weren't focusing on what mattered most. We had to give it more guidance. So we stepped back and got specific: “Here's what makes their voice THEM — they lead with truth not performance, story before strategy, lived experience IS expertise, and they’re honest about not having it figured out. That's their whole approach.”
With a little more guidance, AI finally described back to us a voice that was authentically theirs.
Why this mattered: I learned that you can't just tell AI "write professionally" or "use grant language." That's how you get generic. You have to train it on the actual person's rhythms, their go-to phrases, AND what they never say.
How to replicate:
Gather 5-10 pieces of their writing OR record them talking about their work (transcripts are gold)
Upload everything to Claude or ChatGPT
Ask: "Analyze these documents and describe this person's writing voice. What makes it distinctive? What patterns do you notice?"
Review what it tells you. Does that capture them? Keep refining until it does. Try a prompt like this: “This captures [what works], but it's missing [what's missing]. Specifically: I [do this thing] all the time and it's core to my voice. The description of [X] doesn't feel right—it's more like [Y]. You didn't mention [pattern I see in my writing]. Please revise.”
-
Once AI can describe the voice, you need to codify it. Not with vague instructions like "be authentic" but with specific, usable guidelines.
What we did: We built a comprehensive style guide. And look, I'm not gonna lie — it is LONG. Really long. But that's because we were trying to capture something complex: how an actual human sounds.
We gave detailed bullet outlines for what mattered in each question of the fellowship application. We told AI: pull the language and backup from the project files to support these points. Once we had that foundation figured out, everything else got so much easier.
The guide covered:
Core principles (lead with truth not performance, story before strategy, lived experience as expertise)
Sentence structure patterns (mix of short and long, when to use em dashes, how to create rhythm)
Specific word choices ("folks" not "stakeholders," "capacity for truth" not "openness to feedback")
What to avoid (over-professionalizing, hiding the mess, generic diversity language)
Power phrases to use often (their actual language that sounds distinctly like them)
Why this mattered: AI needs constraints to be creative. The style guide isn't about limiting it — it's giving AI enough specificity that it can sound like this actual person instead of defaulting to bland grant-writing mode.
How to replicate: Create a style guide with these sections then upload it to your chat/project:
Core principles: What matters most in how this person communicates? (3-5 max)
Voice & tone: How do they want to sound? What are they NOT?
Sentence patterns: Short and punchy? Long and flowing? A mix?
Word choices: What words do they use all the time? What words would they never use?
Common pitfalls: What should you always avoid in their writing?
Power phrases: What are the phrases that are distinctly THEIRS?
-
Okay, here's where we stumbled into something brilliant. We'd been working in one chat, and we ran out of memory. So I thought: What if we use a completely separate chat to review?
What we did:
Prompt 1 (Drafting): In the project chat with all the files, we used AI to draft based on the style guide and source materials
Prompt 2 (Review): Started a BRAND NEW chat (fresh eyes!), uploaded just the style guide and the draft, and asked AI to review it against the voice guidelines
The second prompt? GAME CHANGER. Absolutely everything got better.
Why this mattered: The two-prompt system creates a check-and-balance. The first chat is trying to write well based on everything it knows. The second chat is asking, ‘But does this actually sound like THEM?’ The two chats serve two different and necessary functions. Breaking them apart not only solved the memory problem but actually created better writing!
How to replicate:
For drafting (Prompt 1 in project chat): "I'm applying for [GRANT/FELLOWSHIP]. Here are the application questions: [paste]. Please draft responses using: 1) The writing style guide, 2) All relevant content from project files, 3) Authentic voice—vulnerable, story-driven, centered on lived experience. For each question: Start with a hook (usually a specific moment or quote), stay within character limits, use actual voice and stories, make sure it reflects building this WITH community through co-creation."
For review (Prompt 2 in new chat): "I've attached a style guide that captures the authentic voice for this application. Here are the draft responses to the application questions: [paste]. Please review and help me: 1) Voice check—Does this sound like them or like a grant writer? 2) Impact check—Will this move people or just inform them? 3) Clarity check—Is their role and approach clear? 4) Uniqueness check—Does this show what ONLY they can do? 5) Suggest specific improvements while keeping voice intact."
-
The first four questions on the fellowship application? HARD. Getting AI to focus on the right things, making sure we had the right documents uploaded (we literally forgot the interview transcript at first!), hitting that mark between personal/human and formal enough for the application.
What we did: We went back and forth. A lot. "This section feels too formal." "Can you make the opening stronger?" "That doesn't sound like something they'd actually say." Each time, we got more specific about what was off and what we wanted instead.
Once we figured out those first four questions, the next eight were easy. We had the rhythm. We knew what worked.
Why this mattered: AI rarely gets it perfect on the first try. But here's the thing: neither do humans. The magic is in the iteration. We did about 5-6 rounds on those first questions, then only 2-3 on the others.
The whole process took about 3 hours—and that's because we had SO much thinking, background, and transcribed material already. If we were starting from scratch, it would've taken days if not weeks. And in the future
How to replicate:
Read every AI-generated sentence. Does it sound like the human writer is talking?
Mark anything that feels "off" and ask AI to revise with specifics: "This feels too corporate" or "This needs more vulnerability"
Don't be afraid to manually rewrite sections. AI is a tool, not a replacement
When something works really well, save it as an example in the style guide for next time
The hardest part? Word counts and balancing authentic language with what a formal application needs. Just keep pushing on it.
-
The final step — and maybe the most important — is making sure you haven't polished away what actually matters in the writer’s message.
What we did: We went through the final application looking for places where we might have smoothed out important roughness. Where could we add back a more specific detail? Where could we be more honest about struggle? Where could we let the vulnerability show through?
We kept phrases like "I don't have all the pieces figured out yet. That's exactly the point." We made sure the personal story moments were visceral and specific, not generic and sanitized.
Why this mattered: AI's instinct is to clean things up. But sometimes the power is in the mess — the admission that you don't know everything, the specific struggle that shaped you, the moment that didn't go as planned.
The result? An application that stood out because it sounded like a real human with lived experience. Exactly what was needed to compete with thousands of other applicants.
How to replicate: Before submitting:
Find 3 places to be MORE specific (real names, actual moments, concrete details)
Find 2 places to be MORE vulnerable (admit uncertainty, name the hard parts)
Find 1 place to break a "rule" (start with "And," use a fragment, let it be imperfect)
Read the whole thing and ask: Would this move ME if I read it? Or would I skim past it?
The number of hours I’ve spent playing “tug” on what to do cannot be overstated!
Secret Sauce & Takeaways
What made this work:
You can TALK your way through - don’t love writing? Record + transcribe and let AI help shape it
Two-prompt system prevents generic responses (draft in project chat, review in fresh chat — the second chat was absolutely the game-changer)
You can customize based on WHO you’re writing for -
Writing for yourself? Use your own transcripts, your own writing samples, your own voice. The style guide captures how YOU communicate.
Writing for your organization? You need transcripts/materials from leadership (ED, founder, key spokespeople) to capture ORGANIZATIONAL voice—not any one person, but the consistent voice across everything your org produces.
Writing for clients as a consultant? Create a separate style guide for each client using THEIR voice. Get transcripts from all your client conversations. Now you can write in their voice while bringing your strategic expertise. (This is Person K's approach and it works beautifully.)
Once you have a style guide, use it in ALL things you’re writing!!!
PRO TIP - Have different types of things you’re writing - e.g. articles, proposals, reports, other? You can make a custom guide OR custom section for each.
Common pitfalls we learned to avoid:
Don't skip the voice analysis step — without it, you're guessing
Don't try to do everything in one conversation — the review step is critical
Don't accept the first draft — iteration is where it gets good
Don't polish away all the rough edges — that's where the humanity lives
The bigger thing: This isn't about AI replacing writing. It's about making writing accessible for people who have brilliance to share but writing is a barrier. Give AI lots of real, true-to-you information. Then ask it to help you get there.
When one human or department just isn’t quite working as is - helps us to remember we’re all pieces/part of the greater org and the work we’re doing.
Questions to ask yourself
If someone read this without seeing my name, would they know who wrote it?
Are you using AI to help find the right words, or to avoid figuring out what you want to say?
Where in the writing process do you most need support? (That's where AI can help most—but you still have to show up and engage)
What are the things in the writing that ONLY this person can do? (Those need to stay human-crafted)
When you read the AI-assisted draft out loud, does it sound like them talking to someone they trust or like reading from a script?
Want to Try This?
Templates & Guides:
Try the Guide to Writing with AI Without Losing Your Voice to teach AI how YOU sound, then use it to get your words out faster. There’s a 3 minute version for quick help, a 1-2 hour version for a full project, and 4-5 hour investment that will set you up to complete future projects in half the time.
Still thinking on it????
Writing for your organization (not just yourself)? Start with DO THIS WELL, then read Section 1 of TAKE THIS FURTHER about organizational voice before deciding if you need the full style guide.
Planning to write multiple things? Path A: DO THIS WELL for first project → TAKE THIS FURTHER to build style guide → Future projects take 1 hour Path B: TAKE THIS FURTHER first to build comprehensive system → Every project after is easier
The LA Más AI Use Case Library - Real examples of prompts, inputs, and outcomes from organizational work using AI while maintaining authentic voice
Recommended Reads:
Natalie Bergstrasser's "From 'I Don't Have Time for AI' to 'Hold My Beer'" — her approach to using AI for fundraising while maintaining organizational voice
Exploring AI: From Wordsmithing to What's Possible — hear Diane's journey with AI and how she works with/leverages it
Connections
Helia can make connections to folks like Natalie to help bring this to life (one-time set up OR ongoing)
This article comes from Helia's client work in Fall 2024, working alongside a nonprofit leader who needed to write a highly competitive fellowship application that taught us that when you give AI enough of the real, true-to-you material and then ask it to help you get there—not write for you, but help you find your words— good things happen. These conversations form the heart of the Helia Library, because we learn the most from doing and from working alongside other doers willing to share what they're figuring out.
As always, take what's helpful, leave what's not, and make it your own.
Article & Resource Tags
