AI Use Policy Template for Small Publishers and Newsletters

An open dark grey paper booklet whose right page holds nine light grey strips in ruled rows, with a plain grey disc at the foot

Each group answers one question a reader or contributor would ask. The template below keeps them in this order.

Why a small site needs one

A small site or newsletter already has a policy: the habits of its editor and contributors. Publishing them turns private judgement into a promise readers can check.

Writing it down helps three ways. Contributors know what is allowed before they file. Readers who suspect a piece was machine-written get a page to read instead of a guess, which matters when someone claims you published AI text, the situation covered in being accused of posting AI content. And when a mistake gets through, a written policy tells you what to do next.

The idea of a public policy is not new. The Partnership on AI's Responsible Practices for Synthetic Media, launched in 2023, asks distributors and publishers to provide a published, accessible policy setting out their approach to synthetic media. A small publisher can meet that with one page.

The sections, and the reasoning for each

What AI may and may not be used for

List uses, not attitudes. "We use AI responsibly" tells a contributor nothing. "We use AI to transcribe interviews and suggest headlines" tells them exactly where the line sits. A useful default for a small publisher is to treat anything a tool produces as unchecked source material: a lead to verify, never copy to publish as it stands. That one rule carries the whole checking duty with it.

Human review and accountability

Name a person, not a process. A policy that says content is "reviewed" leaves nobody responsible. Google Search Central's guidance on generative AI content says AI output may contain inaccuracies and should be fact-checked and reviewed before publishing, and it extends that review to title elements, meta descriptions, structured data and image alt text. Small sites often let tools fill those fields unchecked, so name them.

Disclosure lines

Say when you will disclose, where the line goes and what it says. A useful test: disclose when a reader would think differently about the piece if they knew. The Partnership on AI framework uses a similar test, and distinguishes direct disclosure, such as labels and context notes, from indirect disclosure such as provenance metadata. For a small site the direct line matters most, placed where it survives a share, as set out in AI disclosure lines that survive a share.

Images

Images need their own clause because the risk is different. A wrong sentence misleads; a fake photograph misleads faster and travels further. Two rules cover most cases: never use a generative tool to add or remove elements in a news photograph, and publish an AI-generated illustration only with a caption that says so. A small publisher can adopt both as written. Platform labels are a separate matter, covered in how platforms label AI content.

Corrections

Say that errors introduced by a tool are corrected under your normal corrections practice, and that the correction says a tool was involved. Without this clause, the temptation is a quiet fix, which damages trust more than the error.

Data and privacy

Whatever is pasted into a tool leaves your control. Keep confidential and sensitive material out of AI tools altogether. For a small publisher the list is short and concrete: unpublished sources, personal data, embargoed material, private correspondence and subscriber lists.

Reader questions and a review date

Give readers an address to ask about AI use and say who answers. Then put a date on the policy and a date for its next review. A policy with no review date quietly becomes untrue.

The template

Copy this, replace the bracketed parts and delete any clause you cannot honour.

AI USE POLICY
[Publication name]. Version 1.0, adopted [date]. Next review: [date].
Owner: [name, role]. Questions: [email address].

1. What this covers
Everything we publish under our name: articles, newsletters, images, audio, video and social posts. It applies to staff, freelancers and guest writers.

2. What we use AI tools for
Transcribing interviews, suggesting headlines, checking spelling and grammar, summarising documents we have read in full, rough translation for a human editor to check, and research leads. We treat all AI output as unverified source material.

3. What we do not use AI tools for
We do not publish AI-written text as our own reporting or opinion. We do not use AI to invent quotes, sources, figures or people. We do not use AI to add, remove or alter anything in a news photograph, video or audio clip.

4. Human review and accountability
A named editor checks every fact, quote, figure and link before publication, including headlines, captions, alt text and search descriptions. The editor is accountable for everything we publish, however it was produced.

5. Disclosure
When an AI tool shaped what readers see, beyond spelling or transcription, we say so in one line at the top of the piece, naming what the tool did.

6. Images
AI-generated or AI-edited illustrations are labelled in the caption. We do not use them to depict real people or real events.

7. Corrections
Errors introduced by an AI tool are corrected under our normal corrections policy, and the correction says a tool was involved.

8. Data and privacy
We do not paste unpublished sources, confidential documents, personal data or embargoed material into AI tools.

9. Questions and review
Ask about our use of AI at [email address]. We review this policy every six months, or sooner if our tools change. Earlier versions stay available at this address.

Adopting it: five steps

Draft from what you already do
Write down the tools each contributor uses and for what, then sort each use into permitted or banned.
Align it with standards you already follow
If you have a corrections policy, a style guide or a code of practice, make the AI clauses point to them instead of restating them. Check the wording against the Partnership on AI framework so your terms match what readers may have seen elsewhere.
Publish it at a stable address
Use a plain path such as /ai-policy/ and never move it. Show the version and adoption date at the top, and keep earlier versions linked at the foot so a reader can see what applied when a piece was published.
Link it from the about page and every byline
Add the link to the byline template so it appears on every piece, and keep the byline markup accurate, as covered in byline schema for a human author.
Review it on the date you set
Put the review date in a calendar, not just on the page. At review, check each clause against recent pieces, bump the version and note what changed.

Clause and what it prevents

Clause	What it prevents
Permitted uses listed by task	Contributors guessing, and the same argument on every piece
Banned uses, including invented quotes	Fabricated quotes, sources or figures reaching print
Named editor accountable	Errors that belong to nobody
Review covers headlines, alt text, descriptions	Unchecked tool output in the fields search and shares show
Disclosure line at the top	Readers learning about AI use from someone else first
No altering news images; label illustrations	A generated picture taken as a record of a real event
Tool errors corrected openly	Quiet fixes that look like concealment when found
Nothing confidential pasted into tools	Sources, personal data or embargoes leaking through a vendor
Reader address and review date	A stale policy nobody can question

A worked example

The publication and figures here are invented for the example. A local food newsletter has 2 editors, 4 regular freelancers and about 3,000 subscribers. Before writing a policy, the editors list current tool use and find 6 uses: transcription, headline ideas, grammar checks, summarising council minutes, translating a supplier's French menu, and one freelancer generating a photo of a market stall that does not exist.

The first 5 go into the permitted list, with summaries and translations marked as needing a human check. The sixth goes into the banned list as a depiction of a real place that never happened, and the editors add a caption rule for illustrations. They publish it at a fixed path, link it from bylines and set a review 6 months out. At review they find 2 pieces missing a disclosure line, add a checklist item and publish version 1.1.

Pitfalls

Writing values instead of rules. "We use AI ethically" cannot be checked; "we do not use AI to invent quotes" can.
Forgetting the fields readers do not see on the page: titles, meta descriptions, alt text and structured data.
Putting the policy only in a footer. Link it from the about page and every byline, where readers look.
Editing the policy in place with no version or date, so nobody can tell what applied to an older piece.

A policy is the paper side of provenance. The provenance section covers the signals that travel with the work itself: bylines, head tags, disclosure lines and Content Credentials on shared images.

Sources

Questions

Does a small blog need an AI use policy?

Yes, if anyone other than you writes for it or readers might ask how it is made. One page that lists permitted and banned uses, who checks the work and how you disclose answers most questions before they are asked.

What should an AI use policy include?

At minimum: permitted uses, banned uses, human review and accountability, disclosure, images, corrections, data and privacy, a reader contact and a review date. The template above covers all nine.

Where should the AI policy be published?

At a stable address on your own site, linked from the about page and from every byline. Keep earlier versions available so readers can see which version applied to an older piece.

Do I have to disclose every use of AI?

No, a sensible policy discloses when the tool shaped what readers see, beyond spelling or transcription. State the threshold in the policy.

How often should an AI use policy be reviewed?

On a fixed date you set when you publish it, and sooner if your tools or practices change.

Hannah Voss, Editor. Checks every guide against a working WordPress install and the networks' current documentation. Last reviewed September 2026.

About the editor and our methods