Dartzo
How it worksTrackingAnalyticsPricingFAQ
Log inStart free
AI replies

Set your AI voice and write good FAQs

Choose tone, reply length, emojis and a persona for your AI replies, and write FAQs it can answer from accurately.

Updated

Two things decide whether AI replies feel like you: the voice (how it talks) and the FAQs (what it knows). They do different jobs, and mixing them up is the most common mistake. The voice shapes the *wording*. The FAQs supply the *facts*. A friendly tone can’t make up for a missing answer, and the agent will never fill a gap with something it invented.

Step 2: Define the voice

On the AI Agent page, Step 2 is called Define the voice. Its note says it plainly: “This shapes the wording; the facts still come from your FAQs only.” Everything you set here is saved with the one Save changes button in the status bar.

ControlWhat it doesOptions
PersonaWho the agent is, in a sentence or two. Example in the field: “The friendly founder of a small skincare brand. Warm, direct, first names only.”Free text
RulesDos and don’ts for how it behaves. The hint under the box says “Rules, not facts.” Example: “Never quote a price that isn’t in the FAQs. Point to the link when relevant.”Free text
Tone of voiceThe overall register of every reply.Upbeat (warm, with energy), Neutral (clear and friendly, the default), Formal (polite, no slang), Casual (like texting a friend)
Typical reply lengthHow much it says.Concise (one or two short sentences), Balanced (two to four sentences, the default), Comprehensive (as complete as your FAQs allow, in one short paragraph)
Use emojisWhether it may add “an emoji or two where it feels natural”. When off, it’s told not to use any.On / off (off by default)
Persona and Rules are not a place for facts
Put “we ship in 3 to 5 days” in an FAQ, not in Persona. The agent answers facts from your FAQs and facts list. If you hide a fact inside a persona line, you’re asking it to treat the voice as a source of truth, and you’ll get inconsistent results.

The same question, four voices

These are illustrations of how the Tone setting shifts wording for one FAQ (“Do you ship outside India? Yes, to the UAE, UK and US, in 7 to 10 days.”). Real replies will vary with your persona and your fan’s message.

ToneA reply might read
UpbeatYes, we do! UAE, UK and US, usually in 7 to 10 days.
NeutralYes, we ship to the UAE, UK and US. Delivery takes 7 to 10 days.
FormalYes, we deliver to the UAE, the UK and the US. Delivery takes seven to ten days.
CasualYep, UAE, UK and US. Takes about 7 to 10 days.

Step 3 controls that affect how it sounds

  • Answer in their language — on by default, so a fan who writes in Hindi, Tamil or Hinglish gets a reply in that language and script. Your FAQs can stay in one language; the agent translates the answer, and the facts still come only from them. Turn it off to pin every reply to one language: English, Hindi, Hinglish, Tamil, Telugu, Marathi, Bengali, Gujarati or Kannada.
  • Mark replies as automated — adds “Automated reply” under every AI-written message, after a blank line. Off by default. It’s a plain disclosure. Use it if your audience would care that a machine wrote the answer. It doesn’t change the wording of the reply itself.
  • When the FAQs can’t answer — the message that goes out *instead of a guess*. Up to 500 characters. If you leave it empty, fans get “Thanks for reaching out! I’ll have someone follow up with you shortly.” That’s a promise, so only keep it if someone will follow up, or write your own, such as “I’ll check and get back to you today.”
  • Reply to greetings — a fixed reply to a bare “hi” or “good morning”, sent with no AI call and no credits. Empty means “Hey! Thanks for reaching out. What can I help you with?” Write your own in your voice.
Write fan-facing messages like a person
The fallback and greeting go to your fans as DMs. Keep them plain: no apologies about how the system works, no technical words. “I’ll check and get back to you” reads like you; “our automated system couldn’t find an answer” doesn’t.

Step 1: Add your FAQs

Step 1 is where the agent’s knowledge lives. The note says: “The agent answers from these only — nothing here, nothing invented.” Click Add a FAQ and choose a kind:

  • FAQ — a Question and an Answer. You only need to write the question once; a fan who words it differently is still matched, so you don’t have to list every variation.
  • Fact — a Title and its Content, with no particular question. It’s something the agent may state whenever it fits, such as “Return policy” or “Studio hours”.

Each entry has an on/off switch, an Edit button and a Delete button. An entry that’s switched off is ignored. If every entry is off, or there are none, the status badge reads Needs FAQs and every question gets your fallback message.

What makes a FAQ work

  1. One question, one answer
    “Do you ship abroad?” and “What does shipping cost?” are two FAQs. A single entry that tries to cover both gives the agent a muddy answer to either.
  2. Make the answer stand on its own
    The agent reads each answer without the rest of your page. Say “Yes, we ship to the UAE, UK and US in 7 to 10 days” rather than “Yes, see above.” Include the specific number, date or link a fan would need.
  3. Put real details in
    Prices, sizes, timings, places and your link belong in the answer. If a price changes, edit the FAQ. If the number isn’t in your FAQs, the agent has no way to give it.
  4. Say what you do not do
    “We don’t offer refunds on sale items” is as useful as a yes. Without it, a fan asking about refunds gets your fallback message, not a clear no.
  5. Use Facts for things nobody asks directly
    Opening hours, your return policy, your brand story: write each as a Fact and the agent can use it whenever it’s relevant.
  6. Test the odd phrasings
    Open Test your agent and ask in the words your fans actually use, including Hinglish or spelling mistakes. Fix gaps by adding or sharpening an FAQ.

A small example set

For a home bakery, a good starting set might look like this. Written as they’d appear in the editor (illustrative):

KindQuestion or titleAnswer or content
FAQDo you deliver outside Jaipur?Yes, anywhere in Rajasthan by courier, 2 to 3 days. Outside Rajasthan we don’t deliver cakes, only cookie boxes.
FAQHow far ahead should I order a custom cake?At least 3 days ahead. For weddings, 2 weeks.
FAQIs anything eggless?Yes. The chocolate, vanilla and red velvet cakes can all be made eggless at no extra cost.
FactOpening hoursOrders are taken Monday to Saturday, 10am to 7pm. Closed Sundays.

Notice there are no prices. If you don’t want the agent to quote them, leave them out and add a Rule such as “For prices, ask the person to check the menu link.” It can’t answer what isn’t written down.

Keeping it within limits

Your FAQs, facts, persona and rules together are sent with every question. In Step 1 the Prompt size meter shows how many tokens that is against the 6,000 allowed. When it says “getting full”, tighten wording or switch off entries you don’t need. If it goes over, the oldest entries are left out of every reply, so trim before that happens. Shorter prompts also use fewer credits, which matters on small allowances. See AI credits.

Common questions

What’s the difference between Persona and Rules?
Persona describes who is talking, such as “the friendly founder of a small skincare brand”. Rules are instructions on what to do or avoid, such as “never quote a price that isn’t in the FAQs”. Neither is a place for facts. Facts go in your FAQs.
Can I make the AI reply in Hinglish?
Yes. With Answer in their language on, a fan who writes in Hinglish gets Hinglish back. You can also pin the reply language to Hinglish, or any one language on the list, by turning that switch off and choosing it.
Do I need a separate FAQ for every way of asking?
No. Write the question once. A fan who words it differently is still matched to the FAQ. Write a second entry only when the *answer* is different.
Why does it reply with my fallback message when the answer is in my FAQs?
Usually the FAQ is too vague or covers two things at once, or the entry is switched off or pushed out by the prompt-size limit. Test the exact question in Test your agent, then make that entry a single clear question with a full answer.
Should I turn on Mark replies as automated?
It’s your call. It adds “Automated reply” under each AI-written message, which is the honest option if your fans would want to know. It’s off by default and doesn’t change what the reply says.
WhatsApp