The Perfect Pastoral Email

The case

Canon Michelle Avery oversees a large network of congregations and receives hundreds of messages each week. Her communications director introduces an AI tool that can sort incoming email, summarize threads, identify sentiment, and draft replies in Michelle’s established style. The system saves hours, and Michelle carefully reviews messages before they are sent from her account. Routine scheduling and acknowledgments become faster and more consistent.

The boundaries begin to blur. The tool drafts birthday notes, thank-you messages, condolences, and responses to clergy describing conflict or exhaustion. Michelle edits names and facts but often leaves the emotional language intact because it sounds warmer than what she could write under pressure. Recipients assume she composed the messages personally. The system also marks some messages “low urgency,” allowing staff to send approved templates without bringing them to her attention.

After the sudden death of a priest’s spouse, the tool drafts a tender condolence message drawing on phrases from Michelle’s previous correspondence. The priest, Thomas, is deeply moved and quotes the email at the funeral. Weeks later, during a staff conversation about innovation, he learns that an AI system drafted it and that Michelle had not read his original message until after the funeral. He feels foolish and says the institution performed care rather than offering it.

Michelle is distressed. She did review the final condolence before it was sent, believed every word, and later prayed for Thomas by name. She also knows that without the tool he might have received only a delayed two-line response. Her communications director argues that the real failure was workflow, not AI. Thomas asks for an honest account of what happened. Other leaders now want a policy, but they disagree about whether disclosure would burden every message with technical detail or whether nondisclosure creates a false relationship.

What is known—and what is not

Known so far

• AI supports sorting, summarizing, drafting, and sometimes routing messages without Michelle seeing the original.

• Recipients reasonably infer that intimate messages sent in Michelle’s name reflect her personal attention.

• Thomas experienced the condolence as relational presence and then as institutional performance.

• Michelle approved the words but did not engage the original disclosure at the time.

• The organization’s workload and scale make purely personal correspondence difficult.

Still uncertain

• Which categories of messages are drafted, reviewed, or sent without Michelle’s direct attention.

• What personal information enters the system and how it is retained.

• What recipients would expect to be disclosed about different levels of assistance.

• Whether the sentiment and urgency classifications contain cultural or other bias.

• What repair Thomas would find meaningful.

Pastoral response questions

Pause before solving. Imagine that you are the leader receiving this person or community. What do you notice, what do you need to learn, and what is the next faithful step?

Locate the pastoral task

1. What makes a communication pastoral: the words, the attention, the relationship, the sender’s intention, or the follow-through?

2. Which messages can appropriately be assisted at scale, and which call for direct human presence?

3. Does approving generated language make it authentically Michelle’s?

Examine the use

1. What risks arise when AI classifies emotion or urgency?

2. What should happen when a message contains grief, conflict, abuse, self-harm, or confidential information?

3. When does writing “in Michelle’s voice” become impersonation rather than assistance?

Consider people, power, and trust

1. What did Thomas reasonably believe he was receiving?

2. Does a senior leader’s institutional power change the obligation to be transparent?

3. Who becomes invisible when efficiency metrics determine attention?

Discern boundaries and accountability

1. What should Michelle say to Thomas now?

2. What uses require disclosure, staff attribution, a different sender, or a clear prohibition?

3. Who should approve the policy and audit whether the workflow follows it?

Plan faithful practice

1. Design an augment / caution / never-delegate framework for correspondence. Where do scheduling, newsletters, condolences, conflict, and pastoral disclosures belong?

2. What handoff ensures that an efficient first response becomes real follow-up?

3. How can the organization address workload rather than asking automation to conceal it?

Designing a tool for when you are stuck

Use these final questions to test what a practical pastoral field guide would need to provide.

1. What questions determine when communication assistance is relationally honest?

2. How should a tool account for sender identity, message sensitivity, consent, review, and follow-up?

3. What language could disclose assistance without turning every email into a technical notice?

Developed from the AI Pastoral Toolkit Design Sprint case corpus and Day One Synthesis and Design Charge (July 2026).