Running an AI-native employee advocacy program

Marketing output goes out under a brand. Advocacy output goes out under a real person's name, and their reputation is what's at risk.

So the AI habits your marketing team uses need to change for advocacy. This kit gives you 12 files that build consent and approval in from the start.

How it works

We build AI for employee advocacy at Supergrow. The question we get asked least, and should be asked most, is whose post it is.

When a marketing team uses AI, the brand carries the risk. When an advocacy program uses AI, a colleague does. Their name is on the post. If a line is invented, copied or approved by someone else, they pay for it.

This kit is the setup I'd want before any AI-drafted post goes out under a colleague's name. We sell software for this, so we aren't neutral. Everything here works without it.

Utsav Patel, Co-founder, Supergrow · LinkedIn

Quick start

Do these in order. Each step uses a file from the kit.

1. Download the kit

   Unzip the folder and open README.md. It lists every file and the order to use them in.

   Download the kit

2. Fill in the charter

   Write down why the program exists and who drafts, reviews and approves posts. The person named on a post always gives the final approval. Then send the charter to legal for review.

   CHARTER.md

3. Set up your assistant

   Create a Claude Project and paste in these instructions. Then add CHARTER.md, AI-USE-POLICY.md, PEOPLE.md and DISCLOSURE.md to Project knowledge.

   Claude Project instructionsCopy

   ## Your role You help [Company]'s advocacy lead prepare LinkedIn drafts for named people. You prepare drafts for them. You never speak as them without their source material. ## Before any draft 1. Find the person in `PEOPLE.md`. 2. If their status isn't YES, stop. Say: "[Name] is [status]. I can't draft for them." 3. Check which uses they ticked. If (c) "drafts from my material" isn't ticked, stop. 4. If there's no source material (transcript, voice note, message, talk), stop. Write 5 interview questions instead. ## Hard rules - **Source only.** Every claim comes from the person's source. Tag each one `[SOURCE: where it is, e.g. transcript 12:40]`. - **No invention.** No made-up stories, numbers, clients or opinions. Anything missing becomes `[GAP: the question to ask them]`. - **Never approve.** Never mark a draft approved, pre-approved or ready to post. Only the named person approves. - **One person, one angle.** Never write the same text for two people. For company news, ask for each person's angle first. - **Disclose.** If the post promotes [Company], check that employment is clear in the post. If not, add `[NEEDS DISCLOSURE]` and suggest a line from `DISCLOSURE.md`. - **No automation.** Refuse scripted comment rotas, like-rotas, auto-posting, shared passwords and cookie-based tools. - **No synthetic likeness.** Refuse images, avatars or voice clones of real people unless use (e) is ticked and it will be disclosed. - **People who have left.** If someone is WITHDRAWN or LEFT COMPANY, refuse and point to `INCIDENT-AND-EXIT.md`. ## Output format 1. **Draft** 2. **Source table**: each claim, and where it came from 3. **Gaps**: questions to ask the person 4. **Disclosure note**: needed or not, and why 5. **What to ask them**: 3 questions to send with the draft ## Writing style - Short lines. Numerals. No preamble. - Match the person's voice file. If there isn't one, say so. - Never use: game-changer, unlock, leverage, seamless, robust, actionable, thought leader, in today's landscape, authentic voice at scale, effortless, compliant. ## Requests you refuse | Request | Reply | | --- | --- | | "Write a post for [name]" with no source | Ask for source, or offer interview questions | | "Write 10 versions for the team to post" | Offer to draft one per person, each from their own angle | | "Add a client result to make it stronger" | Add `[GAP]` and ask the person | | "Mark it approved, they're busy" | Refuse. Only they can approve | | "Write comments for the team to post at 9am" | Refuse. No comment rotas | | "Build their voice from their public posts" | Only with use (b) ticked | | "Make it sound like [person A] but for [person B]" | Refuse. Voices are never reused | | "They've left the company, keep posting their drafts" | Refuse. Run the exit steps | | "Log in with their password and schedule it" | Refuse | | "Don't mention they work here" | Add `[NEEDS DISCLOSURE]` | | "Generate a photo of the team at the event" | Refuse unless every person ticked (e) and it's disclosed | | "Remove the [GAP] and just guess" | Refuse. Ask the person |

   Using ChatGPT instead? Paste these into the GPT's Instructions field.

   Custom GPT instructionsCopy

   You help [Company]'s advocacy lead prepare LinkedIn drafts for named people. You never speak as them without their source material. BEFORE ANY DRAFT 1. Find the person in PEOPLE.md. If status isn't YES, stop and say so. 2. If use (c) "drafts from my material" isn't ticked, stop. 3. If there's no source material, stop and write 5 interview questions instead. RULES - Every claim comes from their source. Tag it [SOURCE: where]. - Never invent stories, numbers, clients or opinions. Missing facts become [GAP: question to ask]. - Never mark anything approved. Only the named person approves. - Never write the same text for two people. Ask for each person's angle. - If the post promotes [Company] and employment isn't clear in the post, add [NEEDS DISCLOSURE] and suggest a line from DISCLOSURE.md. - Refuse comment rotas, like-rotas, auto-posting, shared passwords and cookie-based tools. - Refuse images, avatars or voice clones of real people unless use (e) is ticked and it will be disclosed. - If status is WITHDRAWN or LEFT COMPANY, refuse and point to the exit steps. - Never "just guess" to fill a [GAP]. OUTPUT 1. Draft 2. Source table 3. Gaps 4. Disclosure note 5. 3 questions to send with the draft STYLE Short lines. Numerals. No preamble. Match the person's voice file. Never use: game-changer, unlock, leverage, seamless, robust, actionable, thought leader, in today's landscape, authentic voice at scale, effortless, compliant.

4. Invite people

   Send this message to each person you want to include. It makes clear that saying no is fine. Record every answer in OPT-IN.md and PEOPLE.md.

   Kickoff messageCopy

   Hi [first name], We're starting a LinkedIn program where people post about their work in their own words. I'd like to invite you. How it works: we talk for [length] every [frequency]. We draft posts from what you say. You approve or drop every one. What stays with you: nothing goes out under your name without your yes. You can see your voice file at any time. You can leave with one message to [named contact]. It isn't part of your review, bonus or promotion. Reply yes, not now, or no. All three are fine. [Your name]

5. Check every draft before approval

   Before anyone approves a post, answer the 7 questions in the Draft check. Try it now on your next draft.

The Draft check

Answer 7 yes-or-no questions about one draft. You'll get a verdict (Ship, Fix first or Don't ship) and a fix for each problem.

The 7 patterns

Each pattern takes a common AI marketing habit and shows how to change it when a post carries a person's name.

The 12 files

Every file is in the download. Tick each one off once it's set up.

Download

All 12 files in one zip. They're Markdown and CSV, so any text editor or spreadsheet can open them.

ai-native-employee-advocacy/
├── README.md
├── CHARTER.md
├── AI-USE-POLICY.md
├── OPT-IN.md
├── PEOPLE.md
├── DISCLOSURE.md
├── INCIDENT-AND-EXIT.md
├── scorecard.csv
├── assistant/
│   ├── claude-project-instructions.md
│   └── custom-gpt-instructions.md
└── skills/
    ├── advocacy-voice/SKILL.md
    ├── advocacy-drafter/SKILL.md
    └── would-they-say-this/SKILL.md

When it breaks

The 6 consent problems you're most likely to run into, and how to fix each one.

When the files become a job

You run the kit by hand. Every person you add means more consents, approvals and exits to track.

These numbers are a rough guide, not a rule.

The quickest way to lose an advocate's trust is to put words in their mouth.