AI in church communications: useful assistant, poor final authority

AI can get a church past a blank page. Give it source material the church has already approved, and it can prepare a draft.

That is where its authority ends. It should not decide what the church teaches or how the church responds to a person. It should not decide whether sensitive information may be used or what gets published. Put a named person over every one of those decisions.

A short policy can make that boundary clear:

AI may prepare candidate material. People retain responsibility for doctrine, pastoral care, safeguarding, privacy, consent, dignity, rights, facts, accessibility, approval, and publishing.

This leaves room for practical help without giving ministry authority to software.

The tool can start a draft. It cannot own the result

Picture an ordinary Monday. A communications volunteer has an approved sermon transcript, a blank caption, and several other responsibilities. A drafting tool may suggest a summary, a discussion question, or a few caption options. That can remove the blank page.

The next decisions are still human decisions. Did the summary preserve the preacher's point? Is the question pastorally appropriate? Does the draft expose a person? Does the church have the right to use the source? Can everyone understand the finished post? Who approves it? Who publishes it?

Plausible wording is not the same as a sound decision. OpenAI's own terms say output may be inaccurate. They require users to evaluate it before use or sharing, with human review where appropriate.

That is a vendor's term for its service, not a complete church policy. It still makes one limit plain. A tool cannot review itself into authority.

Several church bodies have reached a similar responsibility principle from different traditions. The United States Conference of Catholic Bishops recommends local guidelines for church personnel who use AI on social media. The Vatican's Antiqua et nova says ultimate responsibility for AI-assisted decisions rests with human decision-makers. The Presbyterian Church U.S.A. offers a discernment worksheet instead of treating AI use as a simple efficiency choice.

These sources do not create one rule for every church. Your denomination, country, insurer, employment agreements, and safeguarding policy may set different or stricter requirements. A local policy should name the people who carry them.

Use a responsibility matrix

A small team does not need a committee for every caption. It does need an owner for every kind of risk.

One person may fill several roles in a small church. That is fine. Write the name beside the role so the team knows who decides and who can stop the work.

Keep red-line material out of ordinary drafting tools

Some material should not enter an AI drafting workflow at all. The default red-line list should include:

• identifiable counselling notes or pastoral conversations
• prayer requests with personal or sensitive details
• safeguarding concerns, allegations, reports, or case notes
• crisis disclosures involving self-harm, abuse, violence, health, housing, or legal risk
• information about children or vulnerable people
• confessions, discipline, employment concerns, or donor records
• testimonies, photos, voices, or stories without specific permission for the planned use
• confidential leadership discussion or unpublished personal information

Removing a name does not always make a story anonymous. A rare event, a family relationship, a photo, a ministry role, or a date may still identify the person.

Canadian privacy regulators tell organizations using generative AI to prefer anonymized or de-identified information when reasonable and to enter personal or sensitive information only when authorized. Their guidance for organizations using generative AI also says obligations vary by jurisdiction, sector, organization, and activity. The UK Information Commissioner's Office has separate guidance for AI systems that process personal data.

Those are Canadian and UK sources, not worldwide legal advice. The practical church rule travels better: do not paste sensitive ministry material into a tool because the prompt box is convenient. If a legitimate process may require such use, the church's privacy and safeguarding owners should assess the tool, contract, authority, retention, access, disclosure, and incident response first.

Safeguarding never becomes a content task first

A disclosure may arrive through a direct message, comment, email, transcript, form, or conversation. Once the material raises a safeguarding concern, stop the content workflow.

Do not ask AI to decide whether the concern is serious. Do not use it to draft the first pastoral response, investigate the report, summarize case notes, or decide what the church should disclose. Follow the church's approved safeguarding process and contact the responsible person.

The Church of England safeguarding guidance includes online harm and directs its churches and officers to established reporting and response processes. Its legal status is specific to that church and jurisdiction. The broader editorial lesson is still useful: a public post comes after safeguarding decisions, if a public post is appropriate at all.

Give AI approved source material, not authority to invent the source

AI-assisted drafting is easiest to review when the source is narrow and visible. Start with an approved sermon transcript, event brief, policy, announcement, or set of facts. Tell the reviewer what source the draft should follow.

Then check the draft against that source.

For sermon-based content, confirm the main claim, qualifications, Scripture references, quotations, speaker attribution, and intended application. A clean sentence can still distort a message by dropping a "not," removing uncertainty, or joining two separate ideas.

For event content, verify the date, time, location, audience, registration details, cost, access needs, and contact route. For a story, verify permission, dignity, attribution, and whether the person understood the planned channels and duration.

If the tool introduces a fact that is absent from the approved source, the tool has not discovered supporting evidence. The editor must find a reliable source or remove the claim.

A reviewer should make real editorial decisions

Human review is not a ceremonial click. The reviewer needs time, authority, the source, and a clear pass condition.

Ask these questions on the final draft:

1. What source supports each teaching point, quotation, name, date, and factual claim?
2. What did the draft omit, combine, soften, or make more certain?
3. Could a person recognize themselves or someone else in this material?
4. Do we have the needed permission and rights for the words, image, audio, and story?
5. Does the copy respect the church's doctrine, pastoral tone, and safeguarding process?
6. Can a person using assistive technology get the same useful meaning?
7. Who approved this exact version, and who will publish it?

If the reviewer cannot answer one of those questions, pause. A deadline does not supply evidence or permission.

Copyright still needs a separate check

AI assistance does not settle ownership or permission.

The U.S. Copyright Office says AI-assisted work may qualify for U.S. copyright protection when a human author contributes sufficient expressive elements. A prompt alone is not enough. That finding concerns U.S. copyrightability. It does not grant permission to use someone else's sermon, photograph, song lyric, book excerpt, Bible translation, testimony, or recording.

Keep two questions separate:

1. Did a person make the creative and editorial decisions in the finished work?
2. Does the church have the right to use every source included in it?

The answer to the first question does not answer the second. Rights also differ by country, contract, employment relationship, licence, and intended use. Ask the rights holder or a qualified adviser when the answer is unclear.

Accessibility needs a person looking at the final asset

An AI tool may suggest alt text or simplify a sentence. It cannot certify the finished post as accessible.

The W3C image guidance says informative images need a text alternative that conveys their essential information. It also recommends avoiding images of text when the text is not a logo. The right alt text depends on why the image appears in that post.

Check the final asset, not the prompt that requested it. Confirm that the alt text describes what is actually visible and relevant. Keep essential wording as real text. Review captions, contrast, reading order, link meaning, and plain-language instructions on the channel where the post will appear.

Accessibility is a publication condition. It is not a style request that software can mark complete on its own.

Will AI-assisted content hurt search visibility?

Google does not say that using AI automatically helps or harms a page. Its guidance for generative AI features in Search says the established SEO fundamentals still apply. It calls for original, useful, reliable, people-first content and warns against making large numbers of pages to manipulate rankings or AI responses.

That means there is no responsible shortcut called "write for AEO" or "write for GEO." Answer a real question. Add knowledge the church can stand behind. Name sources near consequential claims. Use descriptive headings. Make the page accessible. Remove filler.

AI may help organize a draft. The church's experience, judgment, sources, and accountability are what make the finished page worth reading. No workflow can promise rankings, recommendations, citations, or reach.

A small church can adopt this ten-point policy

Use this as a starting point, then align it with local law, denominational policy, safeguarding procedures, and the tools your church has approved.

1. Use AI only for drafts based on approved, non-sensitive sources.
2. Name the person who owns each draft before work begins.
3. Keep red-line material out of prompts and uploads.
4. Check source, doctrine, facts, rights, consent, dignity, and accessibility.
5. Route pastoral and safeguarding material to people through approved processes.
6. Require approval of the exact final copy and asset.
7. Let a named person publish. Do not make automatic posting the default.
8. Record the source, reviewer, approval, and material changes.
9. Recheck the tool when its contract, data use, features, or your local rules change.
10. Stop when the team cannot name the accountable person or show the supporting source.

This policy is intentionally plain. A volunteer should be able to use it on a busy week without guessing what "responsible AI" means.

Frequently asked questions

Can AI write a church social caption?

It can prepare candidate wording from an approved, non-sensitive source. A person still checks the source, doctrine, facts, rights, consent, dignity, accessibility, and final wording. Another church may choose a stricter boundary.

Can a pastor use AI to write teaching?

The tool can generate text, but that fact gives it no teaching authority. Each church must decide where drafting ends and pastoral or doctrinal work begins. If the tool adds interpretation, application, counsel, or a theological claim, an authorized person owns the decision to revise, reject, or use it.

Can we paste a prayer request into AI after deleting the name?

Do not assume that deleting a name removes the risk. Details may still identify the person, and the request may contain sensitive pastoral information. Keep it out unless an approved church process has assessed the tool, authority, privacy duties, consent, access, retention, and response.

Does human review make AI output safe or accurate?

No. Review is a control, not a guarantee. It works only when the reviewer has the source, time, skill, authority, and a clear reason to stop. Some material should never enter the drafting workflow.

Should we tell readers that AI helped?

There is no single disclosure sentence that fits every tool, country, denomination, and use. Ask whether a reasonable person would need the information to understand how the content was made, assess its trustworthiness, exercise a right, or avoid being misled. Check the tool's terms and applicable rules. Do not present generated material as a personal testimony, pastoral response, or human quotation when it is not one.

Can AI create alt text?

It can suggest a draft. A person must inspect the final image and its purpose in context. The alt text should describe the useful visible information, not the prompt, the intended image, or objects that are missing from the final crop.

Can an AI tool approve or publish the post after review?

The policy in this article says no by default. Approval and publishing belong to named people. If a church later considers an automated step, it needs a separate risk decision, clear authority, access controls, a recovery plan, and proof that the step cannot bypass the final approved version.

What should we record?

Keep the approved source, the final copy and asset, the responsible owner, the reviewers, the approval, and any decision about rights, consent, privacy, or safeguarding. Record material changes after approval. The note should let another person explain why the church published that exact version.

AI is useful when it stays in a small, honest role. It may help a church get from an approved source to a candidate draft. People decide whether that draft is faithful to its source, respectful of people, supported by facts, accessible, approved, and ready to publish.