Tutorial · 5 min read
A Dental Practice Build Sheet: What to Switch On First
On this page
- Step one: connect, and check what you connected
- Step two: write the refusal list before the first trigger
- Step three: build exactly one automation
- Step four: Test on myself
- Step five: go live on one automation, and stop
- Step six: add the first-visit automation
- Step seven: export a CSV once, on purpose
- Step eight: set quiet hours against your patients' clock
- Step nine: make the review queue your default posture
- What not to build in month one
This is the build sheet. It assumes you have already accepted the argument — that a dental account can automate the front desk and must not automate the clinic — and it just tells you what to do, in order, on a Tuesday afternoon between patients.
If you have not accepted the argument yet, the case is here and the clinical line is here. Come back after.
The order below is not arbitrary. Almost every practice that gets into trouble did the steps in a different sequence — usually by building a nice automation first and thinking about refusals afterwards.
Step one: connect, and check what you connected
Connect the Instagram account through the official Graph API flow. It will ask for permissions and you should read what it asks for.
Two things to confirm before moving on. First, that it is the professional account for the practice and not somebody's personal handle that got linked by muscle memory. Second, that whoever owns the login is somebody who still works there. Clinics change front-desk staff, and an automation connected through a departed receptionist's session is a support ticket waiting for a bad week.
Step two: write the refusal list before the first trigger
Nothing else in this document matters as much as this one.
Open the negative keywords field and fill it before you build a single automation. Pain, dard, swelling, sujan, bleeding, khoon, abscess, broken, toota, chipped, sensitivity, wisdom, emergency, urgent, infection, pus, medicine, tablet, dose, antibiotic, X-ray, report, plus the words for toothache in whatever languages your patients actually type.
Add price words too — cost, kitna, charge, rate, fees, EMI — unless what you are quoting is the consultation fee and nothing else. Treatment prices depend on an examination, and a number sent before the examination is a figure you will be arguing about later.
Step three: build exactly one automation
Pick your most-commented evergreen post. Not a reel that went unusually well — an ordinary clinic post that keeps attracting the same questions.
- Trigger. That post, plus a small any-of keyword set: timing, hours, open, address, location, parking, Sunday, Saturday.
- Negative keywords. The list from step two. It runs regardless.
- Public reply. Generic and short. "Sent you the details." It is visible to everyone, so it says nothing that is true of this person in particular.
- Private reply. The actual front-desk facts — hours, address with a landmark, parking, lift, reception number and when it is staffed.
- Capture to Leads. On. You want to know who is asking, even in week one.
- Advanced. Leave quiet hours alone for now and set a rate budget you are comfortable with. Skip first-time-senders-only for this one.

Step four: Test on myself
The button exists so you never learn what your automation says from a patient.
Run it. Read the public reply as a stranger scrolling the post would read it, not as the person who wrote it. Ask whether it reveals anything about the commenter. Then read the private reply and check every fact in it against reality — the phone number, the hours, the landmark. Practices get this wrong more often than they expect, usually because the address in the template came from the website and the website is two years old.

Step five: go live on one automation, and stop
Switch it on. Do not build the second one today.
For the next week the job is reading Activity. You are looking for three things: replies that went out and should not have, replies that were blocked and the reason given, and messages that arrived and matched nothing. That third category is your real keyword list, written for you by your patients.
Step six: add the first-visit automation
Once the front-desk one has run clean for a week, the second automation is procedural anxiety — how long the first appointment takes, whether to come fasting, whether anything is done on day one, what to bring, whether a parent can come in, how payment works.
Same shape. Generic public reply, detailed private reply, refusal list intact.
Step seven: export a CSV once, on purpose
Go to Leads and export. Open the file.
Do this before you have accumulated anything worth keeping, so that you learn what the export contains while it does not matter. There is no CRM sync and no one-click integration to anything — the CSV is the handoff, and your practice manager works it like a callback list. Knowing that in week two is better than discovering it in month six.
Step eight: set quiet hours against your patients' clock
Late night off. Evening open, because evening is when people who spent the day at work think about their teeth.
Check the timezone the quiet hours are configured against. This is the setting practices get wrong most often and notice last.
Step nine: make the review queue your default posture
Anything that is not a fixed, reviewed template should wait for a person. On a clinical account that is not a quality preference, it is the boundary between supervised and unsupervised communication.
And leave takeover alone — it is on, and it means the moment anyone at your desk replies by hand, automation is out of that thread for good.
What not to build in month one
No price automation beyond the consultation fee. No triage, ever. No appointment confirmation, because there is no calendar integration and nothing in the system can see your book. No story-reply automation until you have watched comments behave for a month.
Templated replies are free and unlimited; a credit goes only on an AI-written reply. Free tier is 100 credits with no card, packs from ₹499. A practice that follows this sheet spends almost nothing, because almost everything on it is a template. If you want the fuller picture of what stops a send before it happens, the checks are documented here.



