What is the difference between a chatbot and an agent?
A chatbot produces a reply and waits for the next question. An agent works towards a goal: it plans steps, uses tools and can change something in another system.
Google Cloud (last updated 2 April 2026) describes an AI agent as software that uses AI to pursue goals and complete tasks, with reasoning, planning, memory and some autonomy. It describes a bot as following pre-defined rules, and an assistant as responding to prompts while the user decides. IBM draws the same line from the other side: a chatbot with no tools, no memory and no plan can only chase a short goal and needs the user at every step.
The table applies that definition to a dental practice. It is a way of sorting tools, not a study of practices.
| Chatbot | Handoff agent | Booking or account agent | |
|---|---|---|---|
| What it produces | A reply in the window | A message a person acts on | A change in another system |
| A typical dental job | Hours, suburb, parking, “are you taking new patients?” from an approved list | Takes the name and number, or drafts an ad change, then stops | Writes the appointment, or edits the live ads account |
| Better tool when | A wrong sentence costs little and a person is available | The practice will not give the tool the diary or the spend | The book or account is connected and tested, and a person can undo the change |
| Wrong tool when | The caller needs a clinical answer, a fee the practice has not approved, or help in an emergency | The practice expected it to book, and nobody picks up the handoff | It can suggest treatment, invent a fee, or spend past the agreed budget |
Swipe the table sideways to see every column.
Sources: Google Cloud, last updated 2 April 2026; IBM, read 12 October 2026; Ryz Labs, updated October 2026. The sorting is a definition, not a measurement.
Handoff agent is software that collects what a person needs, passes it to your team and stops. It changes nothing in your appointment book or your accounts.
A model that only answers is still a chatbot, however fluent it sounds. Ryz Labs (updated October 2026) puts the failure in one line each: a chatbot’s failure is a wrong sentence, and an agent’s failure is a wrong action.
When is a chatbot the better tool for a dental practice?
A chatbot is the better tool when the job is an approved fact and a person will finish the task. Hours, address, parking and a yes-or-no on new patients fit that job.
Ryz Labs (updated October 2026) gives the same test: build a grounded chatbot when a person will take the action. The answers come from a short list the practice has signed off, and anything off the list goes to your front desk.
When is an agent worth it?
An agent is worth it when the same job repeats across more than one system, and every action can be checked and undone. Ryz Labs (updated October 2026) says to build an agent when the system must look across systems, decide the next step, call an API and update a record, inside set limits.
Four things make that safe for a practice:
- A written limit on what the agent may change.
- A budget cap, if it touches advertising spend.
- A log of every change that a person can read.
- A way to undo a wrong change the same day.
Enterprise Monkey (27 August 2026) describes three rungs for practices: a chatbot answers and waits, a scribe drafts a note, and an agent books, sends and files. The risk moves with the rung. A bad chatbot reply confuses a patient. A bad agent action has already left the building.
Can it book the appointment?
There are three answers, and a vendor’s page often hides which one it means.
- No, it only replies. A chatbot gives your booking link and a person or the patient does the rest.
- It takes a request. A handoff agent collects the name, the number and what the person needs, and passes it to the front desk. A booking happens only when the appointment system is connected and tested. Otherwise the tool captures a request.
- It writes to the book. This needs a connected appointment book, identity checks for changes and cancellations, and a person who can undo a wrong booking the same day.
More autonomy is not a better product here. A write to the appointment book is the risky step, and a handoff is often the safer one. Shoutout Agents does not take access to practice-management software.
What are you still responsible for?
You are. A tool does not take over the practice’s duties, and a vendor cannot certify your clinic. Four rules apply to a chat or an agent on a dental website.
- Ahpra, 22 August 2024. Practitioners stay under their code of conduct when they use AI. The principles named are accountability, understanding the tool, transparency, informed consent, and the ethical and legal issues.
- OAIC, updated 17 January 2025. A public chatbot must be clearly identified as a chatbot. Personal information that goes into an AI system, and personal information it generates, is still personal information. Your privacy policy and collection notice need a plain line about the AI.
- TGA, last updated 5 February 2026. The TGA regulates by intended purpose, not by the word “AI”. A chatbot that suggests, delivers or monitors treatment is one of the TGA’s own examples of a medical device. A tool that later adds treatment suggestions has changed its purpose.
- Ahpra advertising rules, section 133. If your practice publishes the words, they are advertising. They must not be false or misleading, use a testimonial about care, or create an unreasonable expectation of benefit. The guidelines page was reviewed on 2 September 2025.
The ADA’s Policy 6.34 (version November 2023, read 12 October 2026) adds that clinical AI is supervised by a dental practitioner and is not actioned by the patient alone. Generative AI stays on low-risk drafting that a person checks. For how our agents treat patient information, read AI agents and patient data.
What does this look like in marketing, rather than on the phone?
The same split applies to advertising. An agent that reads the search terms and drafts a change is a handoff. An agent that edits the live account is the riskier kind, so it needs a limit on spend and a log.
At Shoutout Agents, we agree the budget with the practice. Our team and agents then run budget, bids and campaigns 24/7, and the practice manages nothing. Agents can raise ad spend by up to 10% a day when the data shows it will help; a bigger rise waits for your OK in the dashboard, and every rise shows there with its reason. Every change is logged for the practice to see. See what our agents do.
If you are weighing a hired team against doing it yourself, read managed Google Ads or doing it yourself. For the wider picture of AI in a dental practice, read AI in dentistry.
Questions
Is an AI chatbot the same as an AI agent?
No. A chatbot replies. An agent pursues a goal and can act in other systems.
Google Cloud (last updated 2 April 2026) describes an agent as software that uses AI to pursue goals and complete tasks, with reasoning, planning, memory and some autonomy. A chatbot with no tools and no plan needs the user at every step, as IBM puts it.
The two words are often used for the same product, which is why it helps to ask one question of any tool: what can it change without a person saying yes?
Can a dental chatbot book an appointment in Australia?
Some can, but only when the appointment system is connected and tested. Otherwise the tool captures a request and a person books.
Booking is the risky step. It needs a connected appointment book, identity checks for changes and cancellations, and a person who can undo a wrong booking the same day.
Without those, a handoff is often the safer design: the tool collects the name, the number and what the person needs, passes it to your team and stops. Shoutout Agents’ agents do not take access to practice-management software, so they never book. A person at the practice writes every appointment.
Does the chat have to tell the patient it is a chat?
Yes. The OAIC (updated 17 January 2025) says a public chatbot must be clearly identified as a chatbot.
The same guidance says personal information put into an AI system, and personal information the system generates, is still personal information. Your privacy policy and collection notice need a plain line about the AI.
A chat that opens by saying what it is, what it can answer and where it passes the enquiry meets the first point without extra work. Your own privacy advisor can check the notice.
Can we paste patient notes into ChatGPT?
Not into a public tool. The OAIC’s best-practice line (updated 17 January 2025) is not to enter personal information, and particularly sensitive information, into a publicly available generative AI tool.
Patient notes are health information, which is sensitive. If the practice wants AI for drafting, the ADA’s Policy 6.34 (version November 2023) keeps generative AI on low-risk drafting that a person checks.
Our agents work in advertising and website accounts. They do not read clinical notes, the appointment book or My Health Record. Read what they can and cannot see.
Who is responsible if the tool gives treatment advice?
The practice is, and a practitioner may be. Ahpra (22 August 2024) says practitioners stay under their code of conduct when they use AI, and Meridian Lawyers (18 October 2024) reads the guidance as requiring human judgment on any AI output.
There is a second problem. The TGA (last updated 5 February 2026) lists a chatbot that suggests, delivers or monitors treatment as one of its own examples of a medical device.
The simplest safe design is a tool that does not give treatment advice at all, and sends those questions to your team.
Is a website chat widget an agent?
Not by itself. A widget is only a place on the page. What matters is what sits behind it.
A widget that answers from an approved list is a chatbot. A widget that collects a name and number and passes them to your team is a handoff agent. A widget wired to your appointment book or ads account is a booking or account agent, and carries that risk.
Ask your supplier which of the three you are buying, and what it can change without a person.
What should the tool refuse to answer?
It should refuse clinical questions, fees the practice has not approved, and anything that sounds like an emergency, and send them to a person.
Those are the cases where a wrong sentence does the most harm. The ADA’s Policy 6.34 (version November 2023) says the practitioner decides, using history, examination and tests, and the patient does not action clinical AI alone.
Shoutout Agents’ agents follow the same line. They never give clinical advice, answer the phone or read clinical notes, and they never message patients or reply to people who enquired. Those questions stay with your team.
Does an agent remove the need for the front desk?
No. A handoff only works if someone picks it up.
The tool captures the enquiry after hours or while the phone is busy, and the front desk calls back. The practice still decides what to book, what to quote and how to speak to a patient.
The ADA’s Policy 6.34 (version November 2023) says the practitioner decides and the patient does not action clinical AI alone.
If nobody picks up the handoff, the handoff agent is the wrong tool for that practice. Fix that first, before adding more software.
Sources
How this page was made. Drafted with AI assistance from the sources below, all read on 12 October 2026 unless a date says otherwise.
- Google Cloud: what are AI agents?Last updated 2 April 2026
- IBM: what are AI agents?Undated page, read 12 October 2026
- Ryz Labs: AI agents vs chatbotsUpdated October 2026
- Enterprise Monkey: AI agents in healthcare, Australia27 August 2026
- OAIC: privacy guidance on commercially available AI productsPublished 21 October 2024, updated 17 January 2025
- TGA: AI and medical device software regulationLast updated 5 February 2026
- Ahpra: AI in healthcare and the code of conductNews item, 22 August 2024
- Meridian Lawyers: Ahpra issues guidance on using artificial intelligence18 October 2024
- Ahpra: guidelines for advertising a regulated health servicePage reviewed 2 September 2025
- ADA Policy 6.34: artificial intelligence in dentistryVersion November 2023, read 12 October 2026
Related reading. See what our agents do and how they treat patient data.
Published . Last updated .