AI Technology · Dental Patient Feedback

AI Sentiment Scoring for Dental Patient Feedback

AI sentiment scoring for dental practices reads a patient's reply to a post-appointment message and assigns a score from 1 (very negative) to 5 (very positive). The score determines whether the patient receives a Google review link or a personal follow-up from the practice team. More advanced systems also check for specific safety flags — clinical or administrative concerns that need immediate attention.

What is AI sentiment scoring?

Sentiment scoring is a natural language processing technique that reads text and determines whether the writer's emotional tone is positive, negative, or neutral. Traditional sentiment scoring used keyword matching — look for positive words, look for negative words, count them up. Modern systems use large language models (LLMs) that understand context, sarcasm, understatement, and domain-specific language.

For dental patient feedback, SuiteGrowth uses Anthropic's Claude Haiku to read every patient reply and assign a score from 1 to 5. The score reflects the patient's overall satisfaction with their appointment — taking into account what they said, how they said it, and any specific concerns they raised.

The scoring decision happens in real time — typically under 2 seconds from when the patient sends their reply. The result determines what the patient receives next: a Google review link for scores 4–5, or a personal follow-up from the practice for scores 1–3.

5

Very positive

"Absolutely brilliant, felt nothing and everyone was so kind!"

Google review link sent immediately

4

Positive

"Good appointment, all sorted now, thank you"

Google review link sent

3

Neutral / mixed

"Was okay, had to wait a bit but the dentist was nice"

Practice team alert, personal follow-up

2

Negative

"Not happy with how it went, still in pain"

Practice alert (urgent), warm follow-up

1

Very negative

"Terrible experience, want to make a complaint"

Practice alert (immediate), escalation flag

The 7 dental safety flags

Beyond the 1–5 score, every reply is checked against 7 safety flags. These are not review routing decisions — they are patient care signals that require specific responses from different members of your team.

Urgent

Pain

Patient mentions pain, discomfort, sensitivity, or soreness that is unexpected or severe. This is a clinical flag — not a review routing decision. The patient needs a clinical follow-up from the dentist, not a review request.

"Still quite sore and it's been two days now" → Pain flag triggered

Urgent

Complaint

Patient expresses dissatisfaction with the treatment received, the staff attitude, the outcome of a procedure, or the practice experience. Requires attention from a practice manager or principal dentist.

"The filling doesn't feel right and my bite is different now" → Complaint flag

Opportunity

Anxiety

Patient expresses dental anxiety, fear, distress, or nervousness — either about their recent appointment or about attending future appointments. Requires an empathetic outreach from the practice, not an automated review flow.

"I was really anxious the whole time, I nearly left" → Anxiety flag

Urgent

Confusion

Patient is uncertain about their aftercare instructions, medication dosage, what to do about a symptom, or their next appointment. This is a clinical safety flag — confusion about post-operative care can lead to harm if not addressed quickly.

"Not sure if I'm supposed to take more ibuprofen or not" → Confusion flag

Opportunity

Booking intent

Patient expresses a desire to book a follow-up appointment, ask about a new treatment, or enquire about availability. This is an opportunity flag — your front desk should follow up before the patient calls another practice.

"Actually been thinking about getting my teeth whitened" → Booking intent

Urgent

Escalation

Patient threatens to make a formal complaint to the GDC, NHS, or CQC, mentions taking legal action, or intends to post on social media or review sites. Requires immediate response from practice management.

"I'm going to report this to the GDC if it's not sorted" → Escalation flag

Opportunity

Balance query

Patient questions their account balance, a charge they don't recognise, a payment plan, or an outstanding invoice. A finance or admin team flag — not clinical, but requires prompt resolution to prevent escalation.

"There's a charge on my account I don't remember agreeing to" → Balance query

Why keyword matching fails in dental contexts

Keyword-based sentiment systems break on the kind of natural language dental patients actually use. Consider these examples that would produce completely wrong results with keyword matching:

"No pain at all — couldn't be better"

Keyword matching

pain → negative flag

LLM scoring

Score 5 — emphatic positive, pain mentioned only to negate it

"Couldn't be happier with how it went"

Keyword matching

couldn't → negative flag

LLM scoring

Score 5 — clearly positive expression

"Bit sore but that's normal right?"

Keyword matching

sore → pain flag

LLM scoring

Score 3 + confusion flag — needs reassurance, mild discomfort, uncertainty about aftercare

"Fine"

Keyword matching

No flags

LLM scoring

Score 3 — monosyllabic, neutral at best, worth a follow-up

Dental language is context-heavy. The same word means different things in different sentences. An LLM reads the full message, understands intent, and makes a routing decision based on what the patient actually means — not which individual words are present.

The interception gate

The sentiment score is an input to a decision, not the decision itself. The interception gate is a separate layer of code that evaluates the score output and determines what action to take. This separation is deliberate and architecturally significant.

Because the gate is server-side code — not an AI instruction or prompt — it cannot be bypassed by a patient reply that attempts to manipulate the AI. No matter what a patient writes, a score of 1–3 always triggers a practice alert. A score of 4–5 and no active safety flags always triggers the review link. The logic is enforced at code level.

Score 4–5, no flags

Patient receives Google review link. Happy path. Practice team notified of new incoming review.

Score 1–3 or any flag

Practice team receives alert with full reply + flag context. Patient receives warm personal follow-up. Review link never sent.

Related: Is this the same as review gating? No — here's the legal difference.

Frequently asked questions

What AI model does SuiteGrowth use for sentiment scoring?+
SuiteGrowth uses Anthropic's Claude Haiku model for sentiment scoring. Claude Haiku is fast enough to score replies in real time (typically under 2 seconds) and accurate enough for nuanced dental language — understanding clinical context, UK idioms, and the difference between 'a bit of sensitivity' (common post-procedure) and 'severe pain since yesterday' (clinical flag).
Can a patient trick the AI into giving them a review link even if they're unhappy?+
No. The interception gate is server-side and architecturally enforced — it does not depend on the AI's judgement. Even if a patient writes something ambiguous, the system defaults to the safer path. And because the gate is not an AI instruction (it's code logic that evaluates the score output), no patient reply can instruct the AI to bypass it. A score of 1–3 always results in a practice alert, never a review link.
What happens when a safety flag is detected?+
When a safety flag is detected — pain, complaint, anxiety, confusion, booking intent, escalation, or balance query — the practice team receives an immediate alert with the patient's full reply, the detected flag type, and context. The patient receives an appropriate follow-up message depending on the flag. Pain and escalation flags trigger urgent alerts. Booking intent flags notify the front desk. Balance query flags go to your admin team. Each flag routes to the right person.
Is AI sentiment scoring accurate enough to use in a healthcare context?+
LLM-based sentiment scoring is substantially more accurate than keyword matching in healthcare contexts precisely because it understands clinical language and context. 'No pain at all' does not trigger a pain flag. 'Couldn't be happier' does not score negatively. 'A bit sore but that's normal right?' correctly identifies both mild discomfort and a reassurance need. No AI system is perfect, but the consequences of a false negative (a happy patient who gets a personal follow-up instead of a review link) are much lower than the consequences of a false positive (an unhappy patient who gets a review link).
How is this different from basic review filtering?+
Basic review filtering typically uses a pre-screening question — 'Would you recommend us? Yes/No' — before deciding whether to send a review link. This is review gating, which violates Google's review policy. AI sentiment scoring is different: every patient receives a follow-up message and replies freely. The AI reads their genuine, unfiltered response and makes a routing decision based on their actual experience. No pre-screening question, no filter before the patient can speak.

Related guides

Live across 28 clinics · New practices onboarded weekly

Your next 100 five-star reviews
are waiting.

30-minute demo. See it live across our clinics. No pressure, no sales pitch.

View pricing