At 7:58 on a Monday morning, the phones at a small veterinary clinic start ringing. A dog ate something off the counter. A cat hasn’t eaten since Saturday. Someone needs a refill. Someone wants to move Thursday’s appointment. Two front-desk staff answer what they can, and the rest go to voicemail. By 9:30, there are fourteen messages waiting, and one of them is from a woman whose elderly dog died over the weekend. She’s calling to ask what happens next.

That last call is the reason this piece exists. A clinic can use AI to help with the Monday-morning pile: answering routine questions, booking appointments, returning missed calls with a text. Those are real improvements. But the same tools, used carelessly, can send a cheerful “We miss Biscuit! Time for his annual checkup” reminder to a family that buried him on Saturday. Kind and careful AI means getting the first part right without ever letting the second happen.

This piece follows an invented four-veterinarian clinic in a suburban American town, managed by a practice manager we’ll call Elena. The clinic, its team, and its clients are illustrative, not a client story. The approach applies to any service business where routine requests and sensitive moments arrive through the same phone line.

Where AI helps a clinic

Elena started by listing what actually happens at the front desk, hour by hour. Most of it was routine:

• Questions about hours, directions, and parking.
• Requests to book, move, or cancel appointments.
• Refill requests that need a veterinarian’s approval.
• Questions about what to bring to a first visit.
• Calls after hours, when nobody answers at all.

Those are the places AI can help. With the WeKinnect Marketing Platform, depending on the edition and how the account is set up, a clinic can use an AI receptionist that answers chat and, in higher editions, phone calls, using the clinic’s own information: its hours, address, services, and a written list of common questions and answers. It can collect a caller’s name and contact details, book into a connected calendar, and, when it doesn’t have enough information to answer, say so and take a message so a team member can follow up. Missed calls can get a text back, so a client who hung up during the morning rush hears from the clinic instead of trying the clinic across town.

For Elena’s team, the goal wasn’t fewer people. It was fewer dropped conversations, and more time at the desk for the calls that need a person.

Where AI should stop

Then Elena made a second list: the moments where AI should not be the one talking.

• Emergencies. A pet that ate something toxic, can’t breathe, or was hit by a car needs a person immediately, or a clear instruction to go to an emergency hospital. The AI’s only job is to get the caller to a human or to emergency care, fast.
• Medical questions. “Is this lump serious?” “Can I give my dog the same pain medicine I take?” Those belong to a veterinarian or a trained technician, not to an assistant answering from a frequently-asked-questions list.
• End of life and loss. Calls about euthanasia, a pet that has died, or aftercare arrangements are some of the most important calls a clinic takes. They should always reach a person.
• Complaints and billing disputes. An upset client needs to be heard by someone who can act.
• Anything the AI isn’t sure about. When in doubt, the handoff is the right answer.

The platform’s AI receptionist can be set up to hand complex inquiries to staff and transfer calls based on what the caller needs and conditions such as time of day. Elena’s team wrote those handoff rules first, before turning anything on. The list of where AI stops was more important than the list of where it helps. It was also shorter to write than to agree on. The veterinarians, technicians, and front-desk staff each had cases the others hadn’t thought of, which is exactly why the whole team wrote it together.

Writing the assistant’s instructions with care

An AI assistant is only as kind as the instructions and information it’s given. Elena’s team spent more time on this than on any setting.

The platform lets a clinic give its AI a written role: its tone, its greeting, and its key tasks, plus instructions that can differ by channel, such as short plain text for text messages and fuller replies by email. Those instructions change how the AI communicates, not what it’s allowed to do, so the clinic wrote both carefully.

The tone section was short: warm, plain, and calm. No exclamation points in replies about a sick pet. No jokes. No pretending to have feelings. When a client sounds worried, acknowledge it simply and get them to a person.

The knowledge section was longer: hours, holiday closures, the services the clinic offers and doesn’t, what to bring to a first visit, how refills work, where the nearest emergency hospital is and when to go there. Each answer was written by someone on the team and reviewed by a veterinarian before it went in.

Then the team tested it. The platform allows configurations to be tested with simulated conversations before they go live, so staff played difficult callers: a panicked owner, a confused elderly client, someone asking for a diagnosis, someone calling about a pet that had just died. Every test that went wrong led to a rewritten instruction or a new handoff rule.

Seeing the busy hour coming

Kind AI isn’t only about tone. It’s also about giving the team enough room to be kind, and that starts with knowing when the pile will form.

After a few weeks, the conversation logs and summaries showed Elena a pattern she had felt but never measured: most routine calls arrived in the first ninety minutes after opening and again just after lunch, and a large share of them were appointment changes and refill questions. With that pattern in view, she moved one technician’s refill work to mid-morning and had the second front-desk person start a little earlier on Mondays.

Nothing about that required the AI to make a decision. It simply made a pattern visible so a person could staff for it. That’s the kind of help a clinic should expect from AI: a clearer view of demand, so the people who still make the call can make it earlier.

Why the sensitive moments matter so much

There’s research on how people react when they learn a heartfelt message came from AI. In a pair of studies published in Frontiers in Psychology, researchers examined how people judged the same supportive note, written to a woman in the hospital, depending on who they were told had written it: a close friend, a florist, or ChatGPT. When participants learned the note was AI-generated, their ratings of its authenticity and respectfulness dropped sharply, and the effect was most pronounced in the most serious scenario, a terminal illness. The authors suggest that organizations be cautious about using AI for messages about grief, celebration, or caregiving, and that human oversight and clear disclosure can help.

The studies were small and scenario-based, and the authors say their findings need replication. But the lesson fits what any veterinary team already knows. When a family loses a pet, the message that matters is the one a person wrote. A condolence card signed by the veterinarian who cared for the animal means something. The same words generated by software and sent automatically mean much less, and if the family ever finds out, they may mean the opposite.

The AVMA’s guidance on coping with the loss of a pet notes that grief doesn’t follow a prescribed pattern or series of stages, and that it can take weeks or months to acknowledge the full reality of a loss. A clinic’s communication should respect that pace. No automated message should arrive on a schedule that ignores it.

The human approval step

The heart of Elena’s plan is simple: AI can draft, but a person approves anything that’s personal.

Here’s how the clinic divided the work:

AI sends on its own: answers to factual questions from the clinic’s written information (hours, directions, what to bring), appointment confirmations the client requested, and missed-call texts that say only that the clinic saw the call and will follow up.

AI drafts, a person approves: follow-up messages after a visit, responses to online reviews, and any message that mentions a pet’s health.

People only: anything about end of life, loss, emergencies, medical advice, complaints, and condolences.

The approval step isn’t a formality. Each morning, a technician or front-desk lead reviews the drafted messages, edits what needs editing, and sends or discards each one. If a draft would be wrong for a family because of something the AI couldn’t know, the person catches it.

The platform supports this kind of review. Depending on setup, each AI response can show an explanation of how it was produced, including which of the clinic’s knowledge sources it used, and conversation logs can be reviewed to find questions the AI handled poorly. For review responses, the reputation tools can be configured to require manual approval, with negative and sensitive reviews held for a person.

Stopping the wrong reminder

The most important single safeguard in Elena’s plan addresses the scenario in the opening: the reminder sent to a family whose pet has died.

Reminder messages come from the clinic’s records. If a pet’s record isn’t updated after it dies, the reminders keep coming. That’s not an AI problem, exactly, but automation makes it worse, because reminders go out faster and more reliably than they used to.

So the clinic made a rule: when a pet dies, the first task, before anything else, is to update the record so every reminder, campaign, and automated message for that pet stops. The person who takes the call does it, or assigns it with a due date of that day. A second person checks it the next morning.

Then the clinic sends what a person writes. A handwritten card from the veterinarian, signed by the team, goes out within a few days. Some clinics also make a small donation in the pet’s name or send a paw print. Elena’s clinic keeps a short list of local grief resources, including the AVMA’s page, to share if the family asks.

Being honest about AI

Clients should know when they’re talking to software. Elena’s clinic decided that its AI receptionist would introduce itself plainly as the clinic’s virtual assistant, not with a human name and a backstory, and would say how to reach a person.

That decision matters for trust. A system that sounds human can borrow trust it hasn’t earned, and people often feel misled when they find out what they were talking to. A clinic that’s honest about AI from the first sentence never has that problem.

Honesty also extends to the clinic’s own team. Staff should know which messages the AI sends, which it drafts, and which it never touches, so no one is surprised by something a client received.

And honesty means not overstating what the AI does. The clinic doesn’t describe its assistant as “knowing” its pets or “caring” about them. It describes what it does: answers common questions, books appointments, and makes sure messages reach the team.

The human in the loop has to be confident

Putting a person in the approval step isn’t enough on its own. That person has to feel able to say no.

At a 2025 AVMA forum on AI, panelists discussed “automation bias,” the tendency of people to defer to automated systems even when their own judgment says otherwise. One panelist put it this way: it’s not enough to have a human in the loop; it has to be an educated, confident human in the loop. The same panel urged caution about software that records and transcribes exam-room conversations, and about how that data might be used.

For Elena’s team, that meant training. The people who review drafts were told explicitly that rejecting or rewriting a draft is part of the job, not a failure. If something feels off, it doesn’t go out. And the clinic decided not to record exam-room conversations for now, until it understood exactly where that data would go and who could use it.

What to audit before turning anything on

Before switching on any AI, Elena’s clinic worked through an audit:

The knowledge sources. Is the clinic’s written information current: hours, services, policies, what to bring? The AI answers from what it’s given, so outdated information means outdated answers. The platform’s own guidance is to keep knowledge sources current.

The handoff rules. Which kinds of calls and messages go straight to a person? Are emergencies routed instantly? What happens after hours?

The records. Are deceased pets marked correctly, so reminders stop? Are client contact preferences recorded?

The approval step. Who reviews drafts, when, and what’s their authority to change or stop them?

The disclosure. How does the AI introduce itself, and how does a client reach a person?

The data. What does the AI record, who can see it, and how long is it kept?

The review schedule. Who reads the conversation logs, and how often? The platform’s guidance suggests regular monitoring, with weekly reviews as a starting point, changing one thing at a time.

A chatbot versus a workforce

There’s a difference between a chatbot and an AI workforce, and it matters for a clinic.

A chatbot captures a lead: someone asks a question, the bot answers, and maybe it collects a phone number. That’s useful, but limited.

A workforce routes, reminds, and reports. It sends the emergency to a person, the refill request to the right technician, and the appointment to the calendar. It sends the missed-call text. It drafts the follow-up for a person to approve. And it shows the team, through logs and summaries, where questions are piling up, so the clinic can fix the cause. If twenty people a week ask whether the clinic sees rabbits, the answer belongs on the website.

The difference is the handoff. A chatbot tries to finish the conversation. A good workforce knows when to pass it on. For a clinic, passing it on quickly is often the kindest thing the technology can do.

How a nonprofit path differs

Compare Elena’s clinic with an invented animal shelter in the same town, a nonprofit that relies on adopters, donors, and volunteers.

The clinic’s AI work is mostly about scheduling and response: getting the right call to the right person and keeping the calendar accurate. The shelter’s is about stewardship. Its donors need to be thanked personally, and its adopters need follow-up that’s warm and specific: how is the dog settling in, does the family need help with training?

For the shelter, the human step is even more central. AI might help sort the donor list, flag first-time donors who haven’t been thanked, or draft an update for a staff member to personalize. But the thank-you itself stays personal. A scored donor list is a sort order, not a verdict on a person. And an adoption follow-up written by software and signed by a staff member who never saw it would undercut exactly the trust the shelter depends on.

The clinic’s scheduling path needs speed and accuracy. The shelter’s donor path needs warmth and memory. Both need people at the moments that matter.

What changed at the clinic

Six months in, picture the Monday morning again. At 7:58, the phones start ringing. The AI receptionist answers the routine questions, books two appointments, and takes refill requests for the technicians. The dog that ate something toxic is transferred straight to a technician. Missed calls get a text within minutes.

The woman whose elderly dog died over the weekend reaches a person. The person who answers is ready: she updates the record so nothing automated goes out, tells the woman what her options are for aftercare, and makes a note for the veterinarian to call her that afternoon. A card goes out by Wednesday, signed by everyone who knew the dog.

That’s a hypothetical morning, not a report, but it describes what Elena’s plan was built to do. The AI didn’t make the clinic warmer. It made room for the clinic’s people to be warm when it counted.

Getting started

If you’d like to bring AI into your clinic or service business without losing what makes it personal:

List where AI helps. Write down the routine requests that fill your phones and inbox.

List where AI stops. Write down the moments that must always reach a person. Make this list first.

Define the approval step. Decide what AI can send, what it can draft for approval, and what stays human only.

Fix the records. Make sure the data behind every reminder is accurate, especially for pets or clients who’ve passed away.

Be honest. Tell people when they’re talking to AI, and how to reach a person.

Review regularly. Read the logs, fix the gaps, and change one thing at a time.

If you’d like help setting up AI that handles the routine and protects the moments that matter, start a conversation with WeKinnect. We’ll start with your two lists: where AI helps, and where it should stop.

Photo by Jonatan Bustos on Unsplash