Adobe Digital Experience

How to Establish Your Voice with AI Tools

A practical setup guide  ·  Claude Desktop, Claude.ai, and M365 Copilot  ·  May 2026

AI tools write generically by default. You can fix that.

Every AI assistant starts from the same place: trained on the average of the internet, defaulting to hedged language, passive structure, and phrases you'd never use. Left unconfigured, it writes like a committee drafted it.

The fix is not a list of rules ("be direct," "no jargon"). Rules describe what to avoid. What the model actually needs are real examples from your own writing — the same way you'd brief a ghostwriter with pieces you're proud of, not a style sheet.

This guide walks through how to extract those examples, combine them into a voice reference, and give it to Claude so it applies every time you write.

Steps
1
If you use M365 Copilot

Extract your voice profile from M365

Copilot has access to data Claude cannot reach: your sent emails, Teams messages you initiated, and documents where you are listed as author. Running an extraction here first gives the most representative sample of how you actually write — not what you remember writing.

Paste the prompt below into Copilot Notebook or M365 Copilot Chat. Copy to clipboard, or save the output as a text file — you will bring it to Claude in Step 2.

Copilot prompt — paste as-isSearch my sent emails, Teams messages I initiated, and documents where I'm listed as author. Exclude: auto-generated summaries, calendar items, replies under 3 sentences, and anything templated. Build a reusable voice profile. Every rule, pattern, and observation must be grounded in specific examples from my actual writing. Quote or cite the source for each finding. Do not infer, generalize, or fabricate — if you cannot find evidence for something, omit it and flag the gap. Return: 1. Style rules (up to 10) — each with a direct quote as evidence 2. Vocabulary — 5 words I use (cited) and 5 I avoid (cited) 3. Signature move — one structural habit with 2–3 examples 4. Tone in 3 words — justified with evidence, no vague descriptors 5. Register shifts — how my voice changes across executive, peer, and customer-facing writing 6. Value framing — pain-led, outcome-led, or proof-led? Cite examples. 7. Closing patterns — hard ask, soft ask, open question, or implied next step?
2
Required

Build your voice reference in Claude

Open Claude Desktop (recommended) or Claude.ai in a browser. Claude will survey the writing samples you provide — plus the Copilot output from Step 1 if you have it — and produce an annotated voice reference file.

Claude Desktop — recommended

Grant Claude access to a folder containing your writing: emails you've exported, meeting notes, briefs, decks. Claude reads the files directly. No copy-paste needed.

Claude.ai in browser

Upload files or paste samples directly into the conversation. Drag in PDFs, text files, or paste excerpts. Filesystem access is not available in the browser — bring the content to Claude.

Once your samples and Copilot output are available, send Claude this prompt:

Claude prompt — adapt as neededI want to build a voice reference file that captures how I write, so I can give it to AI tools as a standing instruction. [If you have Copilot output — paste it here, or say:] Here is a voice profile extracted from my M365 writing history by Copilot: [paste Copilot output] [Point Claude to your samples, or say:] Here are writing samples from my own work — emails, briefs, meeting notes, transcripts: [paste or attach samples, or name a folder if using Desktop] Build a reusable voice reference. Structure it as: - Hard rules: what never to do - Sentence rhythm: before/after contrast using my actual sentences - Vocabulary: words I use and words I avoid - Specificity standard: how I name people, products, and constraints - Register shifts: how my voice changes for executive, peer, and customer audiences - 2–3 annotated examples pulled from my writing, labeled for why they work Save it as voice_style.md somewhere you can find it — your Documents folder is fine.

Review the output before treating it as final. The annotations — not just the examples — are what make it reusable. If a section feels off, paste a sentence you prefer and ask Claude to recalibrate.

3
Optional — Claude Code users

Make it permanent: have Claude apply your voice every session

Claude Code is an advanced version of Claude that runs as a desktop app or in a terminal. It reads a plain-text instructions file — called CLAUDE.md — at the start of every session. Add a pointer to your voice reference there, and Claude applies it automatically to every writing task without being reminded.

Add this line to your CLAUDE.md:

Before any writing task, read [full path to your voice_style.md] and write in that voice.

The full path is wherever you saved the file in Step 2 — use that exact location. Your CLAUDE.md is at the same location for everyone:

CLAUDE.md location — standard for all usersMac / Linux: ~/.claude/CLAUDE.md Windows: C:\Users\[your name]\.claude\CLAUDE.md

If the file does not exist yet, create it. You can also ask Claude Code to handle this at the end of Step 2: "Save the voice reference and add an instruction to my CLAUDE.md so it applies automatically."

Keep it current

Voice references go stale if they reflect only one type of writing. If you write executive briefs, internal notes, and customer-facing copy, include one example of each. When your voice drifts — new account, new format, new audience — paste a sample you like and ask Claude to update the reference. The model adjusts based on which annotated example most closely matches the current task.