AI Use Policy
For every Narqix team member. After reading this page you will know when you may use AI tools on client work, what you must never put into them, and how to use them so your work gets better, not riskier.
Track: All tracks · Role: Everyone · Time to complete: 15 min
Our position on AI
Narqix uses AI. It helps us draft faster, summarise long threads, check our writing and research more efficiently. Used well, it makes a good operator excellent. Used carelessly, it leaks data, invents facts and produces replies that sound like a robot.
The rule is simple: AI can help you do the work. It never replaces your judgment, and you own everything it produces.
The eight rules
- Approved accounts only. Use only AI accounts provided by Narqix or approved by the client. Never use a personal or free account for client work. Business accounts are configured so your inputs are not used to train public models; personal accounts often are.
- Treat every prompt as public. Assume anything you type could be seen by others. Never enter customer personal data, passwords, payment details, health or financial information, contracts or anything the client marked confidential, unless the client has approved that specific tool for that specific data in writing.
- Anonymise first. Replace real details with placeholders before you paste anything: (CUSTOMER), (AMOUNT), (COMPANY), (ADDRESS).
- You own the output. Check every fact, number, name, date, link and policy statement against the source before it goes anywhere.
- No unedited customer replies. Any AI-assisted message to a customer is read fully and personalised by you before sending.
- No AI decisions on money, law or compliance. AI must not decide refunds, credits, contract wording, legal answers, compliance questions or offers.
- Be honest about AI. Follow the client's disclosure rules. Never let an AI tool claim to be a human. Never claim a human wrote something you know an AI agent sent on its own.
- Stay within approved uses. Each Client Playbook lists the AI uses the client approves (for example drafting, summarising, research, data cleanup). If a use is not listed, ask your Campaign Lead first.
Never do this: Never paste a customer's name with their phone number, address, account details or story into an AI tool. Never paste passwords, card numbers or contract text. Never send an AI-written reply to a customer without reading and editing it.
AI voice and calling
US law treats AI-generated or cloned voices as "artificial voices". Calling consumers with them requires prior express consent, and written consent for marketing. Narqix does not use AI voices or voice clones to call or leave voicemails for consumers unless the client's compliance team has confirmed the required consent exists and Narqix has approved it. See Compliance Basics for Calling and Texting.
Good uses and risky uses
| Good uses (when approved for your campaign) | Risky or banned uses |
|---|---|
| Improving the grammar and clarity of your own draft | Writing customer replies you send without editing |
| Summarising a long, anonymised email thread | Pasting a real customer conversation with names and numbers |
| Turning your rough notes into a clean SOP draft | Asking AI what the law says and repeating it to a customer |
| Brainstorming questions to ask a new client | Letting AI decide whether to approve a refund |
| Creating a practice role-play for yourself | Generating fake reviews, testimonials or data |
| Checking a formula or a spreadsheet approach | Uploading client files or spreadsheets with personal data |
| Explaining an unfamiliar US term or idiom | Using AI to research a specific private person |
Example: anonymise before you prompt
Wrong:
"Rewrite this to be more polite: (Real customer name), your refund of 612 dollars for order 48213 to card ending 4421 is denied because..."
Right:
"Rewrite this to be more polite and clear for a US customer: Hi (CUSTOMER), your refund of (AMOUNT) for order (ORDER NUMBER) cannot be processed because the item was delivered (NUMBER) days ago, outside the 30-day window. I can offer (ALTERNATIVE)."
Then you put the real details back in, inside the client's system, and check every word.
Checking AI output
AI tools can produce confident, well-written text that is wrong. Before using any output:
- Every fact checked against the source (CRM, policy, order, contract)
- Every number recalculated or checked
- Every name and spelling correct
- No invented links, laws, dates or quotes
- Tone matches the client's style and the Playbook
- It sounds like you, a real person, not a template
- All placeholders replaced
Coach's note: The best way to use AI is as a second pair of eyes, not a first draft machine. Write your reply first, then ask AI "how could this be clearer for a US reader?" You stay in control of the facts, and your English improves every time.
If something goes wrong
If you put restricted data into an AI tool by mistake, it is a security incident. Report it to your Campaign Lead within 15 minutes. See Data Security and Privacy.
Check yourself
- May you use your personal AI account to rewrite a client email if you remove names first?
- What do you replace real details with before prompting?
- A customer asks if their state allows them to cancel a contract. Can you ask AI and repeat the answer?
- Name two uses of AI that are banned.
- Who decides which AI uses are allowed on your campaign?
Answers
- No. Only Narqix-provided or client-approved accounts may be used for client work.
- Placeholders such as (CUSTOMER), (AMOUNT), (COMPANY), (ADDRESS).
- No. Legal and compliance answers are never decided by AI. Escalate and use the client's approved answer.
- Any two: sending unedited AI replies, AI decisions on refunds or legal matters, pasting personal data, generating fake reviews or data, AI voice calls without verified consent.
- The client, recorded in the Client Playbook, with Narqix approval. Ask your Campaign Lead if unsure.
© Narqix Academy