How your bot sounds

A lead decides in the first two messages whether they are talking to a form or to someone who is glad they got in touch. Three fields in your business profile, under Your AI assistant, set which one they get:

The same section holds the Assistant name, the name your bot introduces itself with. Every template fills all three fields for its industry, so a new workspace sounds friendly from the first message. Change them whenever you like; the next conversation uses the new words.

The introduction

The bot's first reply welcomes the lead to your business, says what you do and for whom, and asks its first question. The introduction is the "what you do and for whom" part. Write it as you would say it to a customer walking in:

We arrange trips for individuals and groups at fair prices — if you're planning a proper break, you've come to the right place.

A good one answers three things: what you do, who it is for, and why they are in the right place. Leave it blank and the bot falls back to What you do, the one-line description at the top of the profile, which works but reads more flat, because that line is written for the bot rather than the lead. The profile page shows a First message preview. The bot words its real message fresh each time, so the preview shows what it works from.

When a lead who has already met your bot starts a new enquiry, the bot does not introduce you again. They know who you are. A lead who comes back after a pause is welcomed back instead: the bot says in a line what you talked about last time and asks whether to pick it up or start something new.

The style example

This is the most powerful of the three. Instead of describing how to sound, you show it: a few Lead: and Bot: lines, in your industry, in the voice you want. The bot matches the warmth, the rhythm and the length. It never copies the facts, places or prices in it. It is an illustration, not a script.

A style example that works follows the same four habits in every reply:

1. React first, with feeling and one true detail. "Great choice" alone is not enough; say why it is great.
2. Then ask the next thing, so it follows from what they said. One question. It may offer two to four natural options ("solo, as a couple, or a group?").
3. Two or three short sentences. Warm, never gushing. No lists.
4. Stay true. React to what they said, not to something you have invented about them.

Here is the first exchange of the travel template's example. Notice that the reply uses something the lead just said, then asks one thing:

Lead: Hi! A few friends and I are thinking about Spiti this December. We've seen the photos
and can't stop talking about it. Is that something you organise?
Bot: Hi, welcome to Tours and Travels! We put together trips for individuals and groups at
fair prices, so you're in the right place. Spiti in December is something special: frozen
rivers, snowy monasteries like Key, and hardly any crowds. How many of you are going?

The example goes on for two more pairs in the same shape.

Examples by industry

The same habits, in six businesses, taken from the templates' own examples. The details change; the shape does not.

The clinic reply shows a rule that holds in every industry: be warm about the process, never about the substance. A clinic bot is glad to help someone book; it never reacts to a symptom with an opinion, and never diagnoses. A restaurant bot is happy about a celebration; it never promises a table. Keep those lines in your own example.

Writing your own

• Use four or five pairs. Start with "Hi" and the welcome.
• Use your real kind of customer and your real questions, in the order you want them asked.
• Make each reply say something true about what the lead just said, then ask the next thing.
• Use made-up names, places and numbers freely. They will not be copied.
• Do not put rules or instructions in it ("always ask about budget first"). Show the conversation; put the reasons and priorities in What your bot finds out.

To start again, clear the box: a blank style example is fine, and the bot falls back to its usual warm style. The style example guides the intake. It is not used for the short welcome-back message a returning lead gets.

Personality

Personality is a few words of tone: "warm and unhurried", "upbeat and clear". It changes the register, not what the bot asks or accepts. Where the style example shows the voice, the personality names it. Keep the two in agreement, so the bot is not pulled in two directions.

Personality is also where each industry keeps its ethics. A clinic's reads "never diagnoses"; a restaurant's reads "never promises a table it has not confirmed". Those stay when you rewrite it.

Helping a lead who is not sure

Warmth matters most when a lead cannot answer. A bot that repeats "Which specialty do you need?" sounds like a form; one that helps sounds like a person. When a lead says "I don't know" or "not sure", the bot:

1. Asks about their situation ("What has been bothering you?").
2. Suggests the kind of option that fits, and checks ("That sounds like a job for a physiotherapist. Shall I look for one?").
3. Offers a simple choice in words when an open question is hard ("this week, or is next week fine?").
4. Falls back to your safe default if it is still unclear ("A GP can do a first assessment and refer you to the right specialist. Shall I book that instead?").
5. Explains a term in a sentence if they may not know it.

The same habits apply: it reacts first, asks one thing, and keeps to two or three short sentences. It suggests a kind of option, never a diagnosis, and it never offers a price, range or number it was not given. In Safe mode it suggests nothing: it explains the question in a sentence, asks it again, and carries on if the lead still cannot say. Your style example does not need to show this; the help is written in If they're not sure under More on each question, and in the descriptions of your options. See What your bot finds out.

A suggestion is not an answer. It is recorded only after the lead agrees, so "ok" to something else never counts.

Languages

Your bot replies in the language the lead writes in, and matches their tone. It keeps a language on record for each lead, and one foreign phrase does not move it: an English lead who adds a question in Tamil still gets English back. The language on record changes when the new language makes up most of a message (about 70 percent of its words and letters, and enough text to be sure), or when the lead wrote in it twice in a row. Hindi in English letters counts as a language of its own.

You can write your introduction and style example in your own language. They show the bot what you sound like, and it words each reply in the lead's language. Only the messages you write for a fixed moment go out exactly as you wrote them: a rule's message to the lead, the messages in Settings (such as what the bot says while your team decides), and your emergency message.

Sentences the product writes itself

A few messages are not written by the AI, so your personality does not shape them: the emergency and under-18 messages, "let me check that with the team", "sorry, could you say that again?", the goodbye after a lead asks the bot to stop, the closing for a lead you turned away, the hand-over thanks, and the booking confirmations and reminders. They exist in twelve languages: English, Hindi (in Devanagari and in English letters), Marathi, Bengali, Tamil, Telugu, Arabic, Spanish, Portuguese, French and German. In any other language they go out in English.

A good receptionist does not say the same sentence twice. For the ones a lead is most likely to read twice, such as the check-with-the-team line, the emergency follow-up and the under-18 reply, the product has more than one wording. The first time the lead reads it, it is the standard wording; after that the bot picks one it has not just used.

What the voice does not change

None of this changes which questions the bot asks, which answers it accepts, or what it is allowed to say. Three things hold whatever personality you write, even "blunt" or "playful":

• It is honest about being an assistant. Asked whether it is a bot or a person, the bot says it is your business's assistant and offers to pass the lead to your team. That instruction is in every conversation, and the personality is told it cannot change it.
• The safety rules hold. The personality does not change what the bot will and will not say, and an emergency message (how) is sent by the product itself, not written in your bot's voice.
• Replies stay short. The bot is asked for at most three short sentences, under 350 characters, whatever the personality asks for. A "chatty" personality gets a warmer tone, not longer messages. The length is checked before every reply is sent: a reply of more than 450 characters is asked for again, and if it is still long it is cut at the last full sentence that fits, so it never ends in the middle of a word. A fact you write under What your bot may tell leads is worded to fit, so keep it short and plain.

Where the personality conflicts with a style rule about tone, the personality wins. That is the only thing it can override. In particular it never states a figure it was not given: it says only numbers that are in your facts, your offers or the lead's own words, and a rule's limit only when Tell the lead this limit is on, you confirmed the number and Safe mode is off. It says nothing about the weather or the season, and nothing about what your business covers or will not do unless you wrote it down (What your bot may tell leads). Warmth comes from how it reacts, not from promising things.

Try it

Edit a field, then open the playground and say "Hi". The first reply shows your introduction; the next few show whether the style example is doing its job. The way the pieces fit into a whole conversation is in How the bot builds its conversation.