By writing a working voice document. A working voice document is a plain-language description of how your practice sounds, what you say, what you don’t, and the perspective underneath all of it. It’s the input every AI tool needs to amplify your voice instead of averaging it into the same content everyone else is publishing.
Most practices skip this step. They open the AI tool, type a prompt, accept the draft, and publish. The output is competent and forgettable, because the voice work never happened before the writing work did. The order matters. Voice gets defined first, or the AI defines it for you, badly, by averaging the internet.
What is a working voice document?
A working voice document is a living, plain-language reference that describes how your practice communicates. It is not a brand guideline PDF. Brand guideline PDFs sit in a folder and never get opened. A working voice document gets used every week, by every person on your team who writes, edits, or publishes content for the practice.
It typically runs three to seven pages. It uses real examples, not abstract principles. It is written in the voice it describes, which is the easiest test for whether it’s working.
Why do you need a voice document before using AI?
Because AI tools default to averaging when given generic input. Without specific direction, they pull from the most common patterns on the internet, which produces the same polished, recognizable AI fingerprints showing up across thousands of practice websites right now.
A voice document gives the AI something specific to start from. The cleaner the input, the cleaner the output. This is the practical version of “what we get out of AI is only as good as what we put into it.” The voice document is the what we put in.
What goes inside a working voice document?
Six components do most of the work:
- Perspective statement. What your practice actually believes about your clients, your work, and what healing looks like. Two to four sentences, written in your real voice.
- Three to five core values with language implications. Not values for a wall poster. Values translated into how you actually write. If you say you value clinical rigor, that should change which words you reach for.
- A “what we say / what we don’t say” list. The vocabulary you use and the vocabulary you refuse. This is where generic empathy phrases get permanently retired.
- Sample sentences with annotations. Three to five sentences pulled from real published content, each with a short note on why it works.
- Tonal range guidance. Where your voice stays consistent and where it flexes by context, including intake pages, clinician bios, and blog posts.
- A “never use” list. Specific phrases, sentence constructions, and vocabulary that should never appear in any content the practice publishes.
How do you actually capture your practice’s voice on paper?
There’s a five-step process that produces a usable document in roughly two to four hours of focused work:
- Pull six to ten samples of your strongest existing content. Blog posts, intake emails, clinician bios, even podcast transcripts.
- Interview the founder or clinical lead for thirty to forty-five minutes about how they describe the work, the clients, and what makes their approach different. Record it.
- Identify the patterns in the transcripts and the samples. What words come up repeatedly? What sentence structures? What perspectives?
- Write the document in plain language, using real examples from your samples and the interview.
- Test the document against three sample pieces of content. If a writer using only the voice document produces content that feels recognizably like your practice, it works. If not, refine.
How do you use the voice document with AI tools?
Four ways, in order of impact:
- Build it into the system prompt for any custom AI assistant your practice uses. This is the highest-leverage move. Every output starts from your voice instead of from the average internet.
- Paste it as context at the start of every session in general-use tools like ChatGPT or Claude. The document is doing the work in real time.
- Train every team member who writes or edits to reference the document before publishing. The voice doesn’t hold if only one person knows it.
- Audit AI output against the document before anything goes live. If a draft doesn’t pass the voice document test, it gets rewritten, not published.
What does this really mean for your practice?
In a content environment where AI now performs roughly 65% of the tasks done in marketing roles in real-world use (Anthropic Economic Index, 2025), voice is one of the most defensible assets a behavioral health practice has. Anyone can publish content. Almost no one is publishing content that sounds like a specific practice run by specific humans.
A working voice document is what makes your AI-assisted content recognizable, citable, and trustworthy. It also makes that content faster to produce, because voice is no longer a debate every time someone sits down to write. If you want to see what this looks like applied end to end, voice is the foundation underneath both branding and content marketing, and it’s where any serious marketing strategy work starts.
Frequently Asked Questions
What is a brand voice document? A brand voice document is a written reference that describes how a practice communicates, including the perspective behind the writing, vocabulary preferences, sentence patterns, and examples of language that does and does not represent the brand. It’s the input behind any consistent voice across channels.
How long should a voice document be? A working voice document typically runs three to seven pages. Long enough to give specific guidance. Short enough that team members and AI tools actually use it. Anything over ten pages becomes a guideline that nobody opens.
Who should write the voice document? Whoever is closest to the practice’s communication should lead the writing, working from interviews with the founder or clinical lead. For most practices, that’s a marketing lead, an internal communications person, or a trusted external partner. The founder or clinical lead does not write it alone, but their voice is the source material.
How often should the voice document be updated? Once a year for most practices. Sooner if the practice goes through a major shift, such as adding a service line, changing leadership, or rebranding. The document is a living reference, not a static deliverable.
Can AI help build the voice document? Yes, if used as an analyst rather than a writer. Feeding existing samples to an AI tool and asking it to identify patterns can speed up the discovery phase. The actual document still needs human judgment about which patterns to keep and which to retire.
What’s one phrase your practice would never use, no matter how often you see it on competitor websites?
