Write the AI Memo Before the Enterprise Seats Arrive

Picture the week your team's new seats go live. On Monday everyone logs in. By Wednesday someone has built a "Brand Voice" GPT from a two-year-old slide deck and shared it with the whole department, and someone else has pasted a customer's complaint thread into a chat to draft a reply. Nobody broke a rule, because nothing was written down. That is the week to prevent, and a one-page memo sent before rollout does most of the work.

The memo is the step before anything in our ChatGPT Enterprise playbook for marketing teams, which goes deep on custom GPTs, knowledge files and governance. The memo is shorter. It tells people what they may do on day one.

What the vendor has already settled

Start with what OpenAI commits to for business workspaces, because some of the questions your team will ask already have answers. On its enterprise privacy page, OpenAI states that it does not train its models on business data by default, that workspace admins control how long ChatGPT Enterprise data is retained, and that members can build and share GPTs internally within the workspace. The same page says admins control which apps are enabled, and that ChatGPT Enterprise workspace admins can view, access, export and delete end user conversations.

That last point belongs in your memo in plain words. People write differently when they know an admin can read a conversation, and telling them up front is fairer than letting them find out later.

What only your company can decide

The contract covers the vendor's side. Your side is a set of choices about behavior, and they differ from one company to the next. A marketing team usually needs to settle these before the first login.

1. What goes in. Name the sources people may paste or upload, such as published pages, approved brand guidelines and anonymized campaign data. Name what stays out even inside the workspace, such as unannounced product plans or anything a client asked you to keep confidential.
2. Who owns each shared GPT. Every GPT shared beyond its builder needs an owner and a review date. When the owner leaves, the GPT gets reassigned or retired instead of running on stale instructions.
3. What gets reviewed before it leaves the building. Any AI-assisted copy a customer will see gets a human read for facts, claims and tone. Say who does that read for each channel.
4. When you disclose AI use. Decide whether clients, partners or readers are told when AI helped produce something, and where the disclosure goes. Some client contracts will already have an answer.
5. What to do when it is wrong. Give people one place to report a bad output, especially a confident factual error in a shared GPT, so it gets fixed at the source instead of patched in every draft.
6. Who can connect apps. If admins enable apps that reach into your drive or CRM, say which ones marketing may use and who approves new ones.

Keep it to one page

A memo nobody finishes is worse than no memo, because everyone assumes it said something it did not. One page, numbered, in the language your team already uses. Put the date at the top and a named person at the bottom, so questions have somewhere to go. Resist listing forbidden prompts. Rules about categories of information age well, while rules about specific prompts go stale within weeks.

A useful test before sending: hand the draft to the newest person on the team and ask what they would do with a client's sales figures. If they cannot answer from the page in under a minute, the page is not done.

The opening lines of a workable memo might sound something like this:

From Monday you have a seat in our AI workspace. Use it for first drafts, for research on published sources and for summarizing our own approved documents. Do not paste client data that is not already public. Anything a customer will read gets a human check before it goes out. Every shared GPT lists its owner in the description; if one gives you a wrong answer, tell that owner. Admins can see workspace conversations. Questions go to the marketing lead.

Who writes it, and who signs it

The marketing lead should write the memo, because the rules are about marketing work and they need to sound like the team. It should not be signed by marketing alone, though. Whoever owns security or IT needs to agree with the lines about data and connected apps, and whoever handles client contracts needs to agree with the lines about disclosure and confidentiality. In a small company that may be the same two people who run everything else. Get both names on the page anyway. A memo with only one department behind it gets treated as that department's opinion.

Leave a few things out on purpose. Prompt libraries, measurement and the custom GPTs you plan to build all belong in the rollout plan, not in the day-one rules. If the memo tries to carry the whole program, it stops being readable.

Send it before the invitations

Timing matters more than polish. If the memo arrives a week after the licenses, it reads as a correction, and people defend what they have already built. If it arrives the day before, it reads as the plan. Send it, ask each person to reply with one question, and answer the questions in a short follow-up. The questions show you which lines were unclear.

After the first month, reread the memo against what actually happened. Which rules did people ask about most? Which shared GPTs got used, and which were abandoned after a week? Update the page, change the date and send it again.

I will concede the obvious objection: a memo does not make the output good. Thin knowledge files still produce off-brand copy, and a team without a prompt structure still gets inconsistent results. The memo is the floor. It keeps a quality problem from turning into a confidentiality or trust problem while you work on the rest.

Then build the system

Once the memo is out, the real work is making the workspace useful. The free Complete ChatGPT Enterprise Playbook for Marketing Teams walks through the knowledge architecture behind custom GPTs, the five core custom GPTs it recommends and how to configure each, a four-layer prompt structure, the GPT Librarian role that keeps shared tools current, a three-pass review system for AI output, integration points with HubSpot, GA4 and Slack, and a 90-day rollout plan. Your memo's line about GPT owners is the Librarian role in miniature, which is a good sign you are starting in the right place.

← Back to all articles