Should eco founders let the team use ChatGPT before they freeze an AI usage policy?
Quick answer
No. Freeze a one-page AI usage policy first: named tools, data that never leaves the desk, the claims kit every chat must paste, and a human sign-off before anything public. A shared consumer ChatGPT with memory on is not a company tool. EmpCo from 27 September 2026 still applies to model drafts. Do not invent a productivity percentage. Ad drafts: LLM ad-copy gate. Email drafts: LLM email gate. CS paste: support macros. Facts the model may know: founder AI facts pack.
Vs other Intelligence guides
| Guide | Job |
|---|---|
| LLM ad-copy claims | Human plus LLM ads before Ads Manager. Needs this policy first. |
| LLM email claims | Klaviyo or ChatGPT emails before Schedule. Same kit, different desk. |
| Support claims macros | CS paste answers. Policy still bans dumping the ticket into a consumer chat. |
| Substantiation folder | Where proof files live. Policy says who may paste the index. |
| AI marketing tools | Stack overview. Not an org fence. |
| First marketer hire | When to add headcount. ChatGPT is not a hire. |
| Brand llms.txt | Public machine file. External assistants, not staff chats. |
| Founder AI facts pack | What the model may know. This page is who may open the chat. |
| This guide | Should eco founders let the team use ChatGPT before they freeze an AI usage policy |
Composite operator guidance. Not legal advice on EmpCo, GDPR, or vendor DPAs. Do not treat a worksheet account count as a typical time-saved lift. OpenAI, Anthropic, and Google do not publish a productivity bonus for skipping a policy.
Three seats. One page.
| Seat | Owns | This week |
|---|---|---|
| Policy owner | Named tools, banned paste, sign-off rule, Team vs consumer | Founder or claims owner. One page, dated. |
| Operator | Pastes the kit and facts pack. Does not invent greener lines. | Anyone who opens ChatGPT, Claude, or Gemini for work. |
| Signer | Human gate before ads, email, GBP, menu HTML, or a guest FAQ | Same person who already signs claims. Not the model. |
Staff line: "The model is a rearranger. The policy is who may open the chat." If you cannot name the signer, do not let the team paste.
Proof ladder (AI usage)
| What is true this week | You may allow | You freeze |
|---|---|---|
| One-page policy signed. Named Team workspace. Claims kit and facts pack exist. Signer named. | Staff chats in that workspace, kit pasted first, human gate before public. | Consumer Plus on a personal Gmail. Memory on. Customer tickets in the prompt. Invented productivity %. |
| Founder-only chats. No staff access yet. Kit exists. | Founder uses Team or a locked consumer account with memory off. Still paste the kit. | A Slack bot that auto-posts model drafts. |
| No claims kit and no facts pack | Research on public URLs only. No ads, no email, no menu copy, no GBP Q&A. | Any public line that came from a chat. |
| Vendor says the draft is "EmpCo-ready" | Ignore that claim. EmpCo is your copy plus proof. EU green claims. | A vendor badge as a policy. |
One-page policy kit
- Allowed tools by name: ChatGPT Team, Claude Team, Gemini for Workspace, or NONE. Not "any AI."
- Account rule. Company workspace. Memory off for brand work. No shared consumer password. Log who has a seat.
- Paste-in. Allowed-claims kit, folder index (filenames only), public URLs, the facts pack, named format.
- Never paste. Customer names and emails, full tickets, unpublished prices, supplier contracts, costed recipes, ad-account passwords, unpublished origin stories.
- Human gate before Ads Manager, Klaviyo Schedule, GBP, menu HTML, Event copy, or a press reply. Ad gate: LLM ad-copy.
- Incident log. Date, tool, what was invented, whether it shipped. Not a hours-saved dashboard.
| Leave out of the chat | Why |
|---|---|
| A live Zendesk or WhatsApp thread | PII plus a claims improvisation. Use macros. |
| "Make it more eco / greener / climate" | That prompt is how invented claims start. Ad-copy gate. |
| Last week's unpublished price list | The model will quote it as live. Facts pack only holds public prices. |
| A vendor "AI CMO" with no scorecard | Headcount is still a hire. Fractional CMO brief. |
Allow vs banned
| Surface | Allow | Kill |
|---|---|---|
| Staff ChatGPT / Claude / Gemini | Named Team workspace, kit plus facts pack pasted, memory off | Personal Plus, browser plugins that scrape the admin, Slack auto-post |
| Ads / email / GBP / menu | Rearrangement of kit lines after a human sign | A chat titled "EmpCo-ready" with no signer. Eco-friendly as a trait |
| Customer data | Macros on your desk. Redacted examples with fake names | Paste the ticket. "Summarise this diner complaint" |
| Training a custom GPT | Only after this policy and the facts pack exist. Public URLs plus kit | Upload the whole Drive. "Train on our brand" with no index |
EmpCo from 27 September 2026 still applies to model drafts that ship in the EU. Score leftover adjectives in the green claims calculator. Do not publish a time-saved rate.
10-check staff-chat log (run this week)
Open the Team workspace. Do not open a personal Plus. Log Y/N, tool, whether the kit was pasted, whether a public line shipped.
| # | Check | Pass |
|---|---|---|
| 1 | Is there a one-page policy with a last-reviewed date? | Dated page, or NONE. |
| 2 | Who is the signer? | Named human, not "the team." |
| 3 | Which tools are allowed by name? | Named list. Not "ChatGPT and friends." |
| 4 | Any consumer Plus still used for brand work? | Leftover accounts go on the worksheet. |
| 5 | Is memory off for brand chats? | Memory on fails. |
| 6 | Did this week's ad draft paste the kit first? | Ad-copy gate. |
| 7 | Did this week's email draft paste the kit first? | Email gate. |
| 8 | Did anyone paste a customer ticket? | Kill. Use macros. |
| 9 | Did a model invent carbon-neutral, plastic-free, or eco feast? | Incident log. Strip before it ships. |
| 10 | Did a vendor promise hours saved or EmpCo-ready copy? | Ignore. No invented productivity %. |
Fix the source that week. They used personal Plus: move the seat. They invented eco-friendly: the prompt still asked for greener. Do not invent a productivity percentage.
Policy freeze (labeled example)
Replace every cell from the signed page and the live tool list. Do not copy the example as a time-saved benchmark.
| Field | Worked example | Your cell |
|---|---|---|
| Signed one-page policy URL or doc (dated) | 1 dated policy example, or NONE | [policy_url] |
| Named Team seats live this week | 4 Team seats example | [team_seats] |
| Consumer Plus or personal Claude still used for brand work | 3 leftover consumer accounts example. Kill or migrate. | [consumer_leftover] |
| Invented-claim incidents in drafts this week (did not have to ship) | 2 leftover invented-claim drafts example. Log them. | [invented_incidents] |
Those cells are worksheet labels, not a productivity quota. If [policy_url] is NONE, staff ChatGPT is blank until you freeze the page.
Go / no-go
| Call | When |
|---|---|
| No-go on staff ChatGPT | No dated policy, no signer, consumer Plus for brand work, tickets in prompts, invented productivity %, vendor EmpCo-ready badge, greener prompts |
| Go on founder-only, skip staff | Founder pastes kit and facts pack. No staff seats yet. Public lines still need a human gate. |
| Go on staff chats plus human gate | Dated policy, named tools, Team seats, kit required, memory off, incident log started, 10-check run once |
| Partial go | Publish the policy today even if Team seats wait a week. Kill consumer leftover this week. Ads and email still use their gates. |
Score signed policy, leftover consumer accounts, and invented-claim incidents. Hours saved is not a KPI.
Policy first. Then the chat.
Name the tools, the paste rules, and the signer. Leave eco-friendly invention out. Do not brief a productivity lift for skipping the page.
Template fields and 7-day run
| Field | Fill with | Do not invent |
|---|---|---|
[policy_url] | Dated one-page URL or doc, or NONE | A Notion dump titled AI strategy |
[tools] | Named Team products, or NONE | Any AI |
[signer] | Named human who already signs claims | The model |
[kit] | Allowed-claims kit plus facts pack, or NONE | Homepage hero as the kit |
[markets] | US / EU / UK / both | One eco policy for every origin |
Brief: "Policy: [policy_url]. Tools: [tools]. Signer: [signer]. Kit: [kit]. Markets: [markets]. If policy is NONE, staff ChatGPT is blank. Kill greener prompts. Do not invent a productivity %."
- Day 1: Export every consumer Plus, personal Claude, Slack GPT, and leftover eco-friendly prompt. Fill [consumer_leftover] and [invented_incidents].
- Day 2: Write the one-page policy. Name [tools] and [signer]. If kit is NONE, freeze public lines until the folder exists.
- Day 3: Freeze banned paste. File the page in the substantiation folder. EmpCo still applies: EU green claims.
- Day 4: Open Team seats or lock founder-only. Memory off. Share the facts pack.
- Day 5: Point ads and email at their gates. CS stays on macros.
- Day 6: Kill leftover consumer accounts. Do not leave a shared Plus password in 1Password titled "AI".
- Day 7: Run the 10-check log. Score leftover invented-claim drafts in the green claims calculator.
Frequently asked questions
Should eco founders let the team use ChatGPT before they freeze an AI usage policy?
No. Freeze a one-page AI usage policy first: named tools, banned paste, claims kit, and a human sign-off. A shared consumer ChatGPT is not a company tool. Do not invent a productivity percentage. Kill greener prompts.
Is this the same as the ad-copy or email gate?
No. Those gates decide whether a draft may ship. This page decides who may open the chat and what they may paste. Run the policy first, then the gate for that desk.
Can we keep using free ChatGPT?
Founder-only research on public URLs, with memory off and the kit pasted, can wait a week for Team. Staff brand work on a personal Plus fails. Do not share one password.
What if the team already used ChatGPT for months?
That is the leftover consumer-account row. Freeze the policy this week. Move seats. Log invented-claim drafts. Do not brief hours already saved.
What do we measure if not productivity %?
Signed dated policy, leftover consumer accounts, invented-claim incidents in drafts, whether the kit was pasted, and whether a public line shipped without a signer.