AI Prompts, called `Tone & Style` when you create them, are reusable instructions that tell the AI how to reply to customer reviews. Instead of accepting generic output, you define the voice, tone, structure, and boundaries the AI should follow.
That is the difference between replies that sound like a machine and replies that sound like the business owner actually read the review and wrote the response.
What AI prompts are
A prompt is effectively a personality brief for the AI. You tell it who it is, how it should sound, what details to pull from the review, and what phrases or habits to avoid.
Without a prompt, the AI tends to fall back into generic thank-you language. With a strong prompt, the replies start to feel specific, calm, and business-appropriate.
- Used when someone clicks Generate with AI on a review
- Used inside Auto Respond rules for fully automated replies
Two levels of prompts
Prompts can live at the agency-wide level or at the organization-specific level. The difference is simply scope.
| Prompt type | Who sees it | Best use case |
|---|---|---|
| Global prompt | Available across the whole platform | Baseline standards that work for many businesses |
| Organization prompt | Only available inside one organization context | Business-specific tone, offers, or industry nuance |
Where to find and create them
Global prompts are created from the main AI Prompts area in settings. Organization-specific prompts are typically created while working inside that organization, either from the review reply modal or from the Auto Respond rule flow.
- Global path: `Settings -> AI Prompts`
- Organization path: Generate AI Reply modal or Auto Respond rule modal inside the organization context
Create a prompt step by step
First give the prompt a clear name so the intent is obvious later. Then write the instructions in plain English, as if you were briefing a new team member on how to respond to reviews for that business.
The better the instructions, the better the output. Vague prompts create vague replies.
Use placeholders properly
| Placeholder | What it becomes |
|---|---|
| {review} | The review text |
| {author} | The reviewer name |
| {rating} | The review rating |
| {organization} | The business name |
| {location} | The location name |
| {language} | The language to respond in |
Using placeholders well makes the replies specific without forcing you to rewrite the prompt for every review.
The formula for human-sounding prompts
The strongest prompts usually do four things. They define the persona, define the tone, tell the AI what to include, and clearly ban the phrases or structures you do not want.
- Who the AI should sound like
- How formal, warm, casual, or restrained the language should be
- What details from the review should be acknowledged
- What phrases, structures, and habits should never appear
Prompt patterns that work well
| Prompt style | Best fit | Why it works |
|---|---|---|
| Warm and personal | Happy customers | Feels like the owner actually read the review |
| Empathetic recovery | Negative reviews | Acknowledges the issue without sounding robotic or defensive |
| Short and punchy | High-volume businesses | Keeps replies brief and natural |
| Professional warmth | Healthcare and similar trust-led services | Balances reassurance with professionalism |
| Hospitality warmth | Restaurants, bars, cafes | Sounds personal and experience-aware |
| Straight-talking trades | Home services and contractor businesses | Feels local, direct, and grounded |
Ban generic AI language explicitly
The easiest way to improve output quality is to tell the AI exactly which phrases it must not use. If you only say “do not sound corporate,” the instruction is too vague.
If you say “never use these exact phrases,” the AI has something concrete to avoid.
- We truly appreciate
- Thank you for taking the time
- It means the world to us
- We strive to
- We pride ourselves
- Valued customer
- We look forward to serving you again
Structural tricks that make replies feel human
- Keep replies short, usually 1 to 3 sentences
- Reference a specific detail from the review
- Vary the opening so every reply does not start the same way
- Use contractions if the business voice is conversational
- Mirror the reviewer tone when it makes sense
Using prompts with Auto Respond
Prompts become even more powerful when paired with Auto Respond rules. The cleanest setup is usually two rules: one prompt for happy reviews and one prompt for unhappy reviews.
That split matters because a five-star response and a one-star response should not sound like the same person copied the same formula.
- Happy review rule: 4 to 5 stars, warmer prompt, no approval, slight delay
- Negative review rule: 1 to 3 stars, empathetic prompt, approval required
Approval flow and delay settings
Approval matters most on negative reviews. Delay matters on all reviews. Instant replies can feel suspicious, while a short delay makes the response feel more believable.
- 1 to 4 hours is a strong default delay for positive reviews
- Negative reviews should usually require approval
- Longer overnight delays can feel more natural than instant posting
Using prompts on older unanswered reviews
If a customer has a backlog of unanswered reviews, prompts can also be used through the bulk reply flow. Spread those replies out over time so the output still feels natural.
Custom plan controls
The agency controls whether customers can use manual AI reply generation, auto-respond, and create custom prompts through the customer plan settings.
| Feature | What it controls |
|---|---|
| Respond with AI | Manual AI-generated replies on reviews |
| Auto-Respond | Automated reply rules |
| Create Custom Prompts | Whether the organization can create its own prompts |
Language support
The AI follows the configured organization language when generating replies. If the business operates in another language, the prompt can also reinforce that using the language placeholder.
Best practices
- Create separate prompts for positive and negative reviews
- Test prompts on real reviews before trusting them widely
- Steal good structural habits from real human business owners
- Keep replies short unless the business voice genuinely calls for more
- Refresh the banned phrase list when you notice repetition