A one-page policy you can copy, edit and share with your team and clients. It covers when AI tools are allowed, who reviews the output, what you tell clients, and the records to keep.
Not legal advice. Laws and platform rules change; check this against your contracts and the places you operate. Links to the US FTC review rule, EU AI Act Article 50 and Google's guidance are included so you can read the sources yourself.
1. Why we have this policy
Our clients pay us for work they can trust. We use AI tools where they help, and we stay responsible for every word we deliver. This policy says when AI may be used, who checks the output, and what we tell clients.
2. Scope
This policy covers everyone who produces work for clients: staff, freelancers and subcontractors. It applies to blog posts, web pages, emails, social posts, ads, scripts, reports and anything else we write or edit for a client.
3. Allowed uses (with human review)
Research help, outlines and topic ideas.
First drafts, if a named editor rewrites and approves the final version.
Editing help: grammar, shorter sentences, alternative headlines.
Summaries of material the client gave us, if checked against the source.
Translation drafts, if reviewed by a fluent speaker.
4. Not allowed
Fake reviews, testimonials or endorsements. We never write reviews or testimonials that pretend to come from a real customer, or from someone who doesn't exist. In the US the FTC's rule on consumer reviews and testimonials (16 CFR Part 465) bans this, including AI-generated ones. https://www.ecfr.gov/current/title-16/chapter-I/subchapter-D/part-465
Invented facts, quotes, statistics or sources. Every number, quote and citation must be traced to a real source before delivery.
Pretending to be a person. No AI-written content published under a real person's name without that person reviewing and approving it.
Client confidential data in unapproved tools. Only use AI tools on the approved list (section 9) with client material, and never paste in anything a client has marked confidential unless their contract allows it.
AI use a client has ruled out. If a contract says "no AI-generated deliverables", that wins over this policy.
5. Human review
Every deliverable that used AI at any stage has a named reviewer who:
reads the whole piece, not a sample;
checks facts, numbers, names, links and quotes against sources;
rewrites generic or padded passages so the piece says something specific;
runs our style checks (see section 6); and
records the review (section 8).
The reviewer is responsible for the final text. "The tool wrote it" is never an excuse.
6. Quality checks
Before delivery we check for:
accuracy (section 5);
brand voice and the client's style guide;
plagiarism or close copying of other sources;
AI-writing style patterns that make copy read as generic: stock phrases ("in today's fast-paced world", "delve"), "it's not X, it's Y" contrasts, dramatic one-line closers, heavy em-dash use. A pattern checker such as SlopScore can point these out. A pattern check is an editing aid only. It does not prove who or what wrote a text, and we never use it to accuse a writer.
7. Telling clients
Our contracts and statements of work say whether and how we use AI tools.
If a client asks how a deliverable was made, we answer truthfully.
We follow any disclosure rules that apply to the published content. Example: from 2 August 2026, Article 50 of the EU AI Act requires deployers who publish AI-generated text to inform the public on matters of public interest to disclose it, unless the text went through human review or editorial control and someone holds editorial responsibility. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50
For each deliverable we keep, for at least [12] months: the AI tools used, the reviewer's name, the date of review, and the checks run (with any report files). This lets us answer client questions and show the human review that disclosure rules may require.
9. Approved tools
Tool
Approved for
Client data allowed?
Owner
[tool]
[drafting / editing / research]
[yes / no / only with consent]
[name]
New tools need approval from [role] before use with client work.
10. Training and breaches
New team members read this policy in their first week. Breaches go to [role]. If an error reaches a client, we tell them promptly and fix it.
11. Review
[Owner] reviews this policy every six months, or sooner if laws, platform rules or client contracts change.
How to use it
Fill in the brackets: owner, review dates, approved tools, how long you keep records.
Check section 4 against each client contract. A "no AI" clause in a contract overrides the policy.
Decide what your review record looks like. A shared sheet with the piece, tools used, reviewer and date is enough to start.
Tell clients you have it. Many now ask.
For the quality-check step, you can paste drafts into the free SlopScore checker. The Agency plan adds batch checks and a report you can send with each delivery.
# AI content policy for [Agency name]
Version 1.0 · Effective [date] · Owner: [name, role] · Next review: [date + 6 months]
This is a starting template from SlopScore (https://slopscore-nine.vercel.app/templates/agency-ai-content-policy/). It is not legal advice. Edit it to match how your team works, and have a lawyer check it against your client contracts and the places you operate.
## 1. Why we have this policy
Our clients pay us for work they can trust. We use AI tools where they help, and we stay responsible for every word we deliver. This policy says when AI may be used, who checks the output, and what we tell clients.
## 2. Scope
This policy covers everyone who produces work for clients: staff, freelancers and subcontractors. It applies to blog posts, web pages, emails, social posts, ads, scripts, reports and anything else we write or edit for a client.
## 3. Allowed uses (with human review)
- Research help, outlines and topic ideas.
- First drafts, if a named editor rewrites and approves the final version.
- Editing help: grammar, shorter sentences, alternative headlines.
- Summaries of material the client gave us, if checked against the source.
- Translation drafts, if reviewed by a fluent speaker.
## 4. Not allowed
- **Fake reviews, testimonials or endorsements.** We never write reviews or testimonials that pretend to come from a real customer, or from someone who doesn't exist. In the US the FTC's rule on consumer reviews and testimonials (16 CFR Part 465) bans this, including AI-generated ones. https://www.ecfr.gov/current/title-16/chapter-I/subchapter-D/part-465
- **Invented facts, quotes, statistics or sources.** Every number, quote and citation must be traced to a real source before delivery.
- **Pretending to be a person.** No AI-written content published under a real person's name without that person reviewing and approving it.
- **Client confidential data in unapproved tools.** Only use AI tools on the approved list (section 9) with client material, and never paste in anything a client has marked confidential unless their contract allows it.
- **AI use a client has ruled out.** If a contract says "no AI-generated deliverables", that wins over this policy.
## 5. Human review
Every deliverable that used AI at any stage has a named reviewer who:
1. reads the whole piece, not a sample;
2. checks facts, numbers, names, links and quotes against sources;
3. rewrites generic or padded passages so the piece says something specific;
4. runs our style checks (see section 6); and
5. records the review (section 8).
The reviewer is responsible for the final text. "The tool wrote it" is never an excuse.
## 6. Quality checks
Before delivery we check for:
- accuracy (section 5);
- brand voice and the client's style guide;
- plagiarism or close copying of other sources;
- AI-writing style patterns that make copy read as generic: stock phrases ("in today's fast-paced world", "delve"), "it's not X, it's Y" contrasts, dramatic one-line closers, heavy em-dash use. A pattern checker such as SlopScore can point these out. A pattern check is an editing aid only. It does not prove who or what wrote a text, and we never use it to accuse a writer.
## 7. Telling clients
- Our contracts and statements of work say whether and how we use AI tools.
- If a client asks how a deliverable was made, we answer truthfully.
- We follow any disclosure rules that apply to the published content. Example: from 2 August 2026, Article 50 of the EU AI Act requires deployers who publish AI-generated text to inform the public on matters of public interest to disclose it, unless the text went through human review or editorial control and someone holds editorial responsibility. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50
- Search engines: Google's guidance does not ban AI-assisted content, but mass-produced pages with little value can break its spam policies. We publish for readers first. https://developers.google.com/search/docs/fundamentals/using-gen-ai-content
## 8. Records
For each deliverable we keep, for at least [12] months: the AI tools used, the reviewer's name, the date of review, and the checks run (with any report files). This lets us answer client questions and show the human review that disclosure rules may require.
## 9. Approved tools
| Tool | Approved for | Client data allowed? | Owner |
|---|---|---|---|
| [tool] | [drafting / editing / research] | [yes / no / only with consent] | [name] |
New tools need approval from [role] before use with client work.
## 10. Training and breaches
New team members read this policy in their first week. Breaches go to [role]. If an error reaches a client, we tell them promptly and fix it.
## 11. Review
[Owner] reviews this policy every six months, or sooner if laws, platform rules or client contracts change.