What your bot may tell leads

On this page

Leads ask more than your intake questions do. "Do you take HDFC Ergo?" "Where are you?" "What's included?" Your bot can answer them, but only from facts you have written down. It never guesses a price, a date or a policy, because a wrong answer from your bot is a wrong answer from you.

You write those facts under Business setup, What your bot may tell leads. Owners and admins can change them; everyone else cannot.

The kinds of fact

Kind	What it is for	Example
Common questions	Questions leads ask, with the answer you would give	"Do you work on weekends?" "Yes, by appointment. Message us the day before."
Prices	What things usually cost, as you would say it to a customer	"A first consultation is 800."
Policies	Cancellations, payments and other rules you stick to	"You can cancel free of charge up to 24 hours before."
Places	Where you are and how to find you	"We're on the second floor, next to the pharmacy."
Hours	Holidays and changes to your usual hours	"Closed 20 to 22 December."
Terms	What your own words mean, so your bot uses them the same way	"A 'session' is 45 minutes."

Your weekly opening hours are not written here. They come from your business profile, under When you're open, and your bot uses them from there. The Hours kind is only for the exceptions.

Selling material is not written here. What makes your business a good choice, your answers to objections, testimonials and upsells live on their own page, Business setup, Selling, so this list stays a list of facts a lead can ask for (How your bot sells). The same rule applies to both: the bot says only what is written, and a figure in your selling material is one it may use.

Each fact has a short title (up to 120 characters) and the text your bot may say (up to 600 characters). Write it the way you would tell a customer; your bot puts it in its own words and says only what is written. A fact can be about one option only, such as a price for one region, and then your bot uses it for leads who chose that option. You can keep up to 200 facts.

Where the first facts come from

A new workspace arrives with a few facts from your template, marked From your template. They are deliberately plain:

A fact that arrives as a prompt, "[Add your answer: …]", is switched off. Your bot will not say a prompt out loud. Press Add your answer, write yours, and it goes live.
A price or a policy that came from a template, or that is read from somewhere else, waits for you. It has a Confirm button, and your bot does not use it until you press it. A template's price is someone else's price. When you edit a fact, or write your own, it counts as yours and confirmed.

Facts read from your website or your files

If you gave a website at signup, we read it once and turn what it says into facts here, each marked From your website and showing the sentence it came from. If you set up with AI from a description and files, the facts your files and your own words contain arrive the same way, marked From your file. Two rules decide whether your bot may use one:

Common questions, hours and places are ready at once. Your bot can answer from them as soon as they arrive. You can edit or delete any of them, and Setup history can undo the whole set (Setup history).
Prices and policies wait for you. A price or a policy read from a website or a file is marked Read from your website — confirm (or Read from your file — confirm), and your bot does not use it until you confirm it. Whatever the page or file says, and however sure it looks, only you can make a price or a policy true for your bot.

A figure is only kept if it is in the sentence it came from. A price written as a common question is still treated as a price and waits. And a number a draft puts on a rule waits too: your bot does not say it until you confirm it on the rule's card.

Confirming prices and policies

A card on your bot's to-do, Check facts we read from your website, appears as soon as there is something to check: it does not wait for the one-new-card-a-day pace, because you asked for it by giving a website or a file. It opens this page.

Read each waiting price or policy next to the sentence it came from, and compare it with your business. Edit it if it is slightly wrong; saving an edit makes it yours and confirmed.
Press Confirm on the ones that are right, or Confirm all at the top.
Delete the ones that are not.

Until you confirm one, your bot says it will check with the team, exactly as it does for any question it has no fact for (When your bot does not know). Try it in the playground: ask for a price before and after you confirm it.

What we keep of a website or a file

We keep the facts and the sentence each came from, never a copy of your pages. A file's bytes are never stored; its text is deleted once your draft is applied or declined, and after 7 days at the latest (Uploading your price list or roster). The facts that became knowledge stay here, as yours, until you delete them.

What your bot says, and what it will not

Your bot answers a lead's question from these facts, your opening hours, and the limits you allowed it to state on a rule. It also says back what the lead told it.

A rule's limit counts as a source only when you allowed it. The bot says a limit out loud ("the least we usually work with is 400") only when all three hold: the rule's Tell the lead this limit switch is on, you have confirmed the number by choosing Hard or Negotiable on the rule, and Safe mode is off. Otherwise the bot knows a limit exists and says only that the answer is a bit outside what you usually do.

Before every reply is sent, the product reads it and checks it against those sources. A reply is not sent as written if it contains:

a price or other figure that is not in a fact, your selling material, a rule limit you allowed, or the lead's own words, or a judgement of the lead's budget ("that's a healthy budget");
a discount, special price or free extra that none of your offers or facts contains. A lead's claim that someone promised them a discount is not a source. Concessions are meant to come from your offers (What your bot may offer);
invented urgency or scarcity ("only 2 left", "offer ends today", "prices go up soon") that no fact says;
a claim about seats, stock or availability ("we have space in December") that no fact says;
a promise with a time on it ("someone will call you within two hours", "I'll ring you today");
a link, email address or phone number that is not written in a fact, a link you chose to send (such as your booking page), or something the lead wrote themselves;
a statement of what your business offers, accepts, covers or bills ("we work with your insurer", "home visits are something we offer", "they bill your insurer directly") that no fact says. The answers you listed for a question count while the bot asks it ("we offer in-person or video visits, which suits you?"), but a day is checked on its own: listing home visits does not let the bot say they happen on Sundays. Who is billed counts the same whoever the sentence is about ("we", "some of our physios", "a practitioner who takes it") and however it is put ("can bill them directly", "if your insurer is X we can ..."): a fact has to say that you accept the insurer, and one that says you bill it directly (or handle its claims) has to say that too. A definition of a word such as "co-payment" or "self-pay" does not;
the name of one of your own records that the lead was not shown. With options on, the records the bot has just shown are fine, and so is what you made visible about them and nothing more;
something a lead told your bot in private, other than the one brief acknowledgement, or what the lead asks to hear again (below);
a promise that someone will contact, check or confirm, to a lead you have turned away, or to a lead who missed only a Hard limit, because nobody reviews those;
more than 450 characters, which is about three short sentences. A long reply is asked for again, then cut at the last full sentence.

Two kinds of friendly colour are kept out by the instructions your bot is given, because plain code cannot tell a true one from a false one. The bot says nothing about the weather, the climate or what a season is like ("the dry season settles in", "quieter and greener in May"). And it says nothing about what your business covers, serves or will not do (a town, a range of trips or jobs, a kind of case, a discount, a policy) unless one of your facts or limits says so; a lead naming a place does not make it covered. If you want the bot to say "we don't match other agencies' discounts" or "we cover Chakan", write it as a fact and it will.

A reply that fails is asked for once more. If the second try fails too, the lead gets the "let me check that with the team" line below, a sentence about a private answer or invented urgency is simply left out, and a reply that is too long is cut. For a lead you turned away, the closing message is sent instead, with no promise of a check.

The check is plain code, not another AI, so it is the same every time and it adds nothing to your usage. It reads every reply the AI wrote, also on the rare turn when the first step fails and one call writes the whole reply. Booking times are the one thing it holds to a stricter rule: the bot may only name a time that the product offered the lead.

When a lead asks what they told your bot

A lead who asks "what did I tell you?", "what details do you have for me?" or "can you remind me what I said about my knee?" is answered. On that one reply your bot says back what the lead told it, plainly and in their words where it kept them, even for a question you marked Sensitive: it is their own answer, they asked for it, and they are the one reading it. Refusing would only look like the bot is hiding something.

Only what they asked for. A general request ("what details do you have?") lists everything they answered. A request about one thing ("what did I say about my budget?") says that one.
Only the lead's own answers. Never what the bot worked out for itself, the notes your team wrote, the score or the verdict, an answer to a question you have switched off, or anything about another lead. It is said on the lead's own conversation and nowhere else.
Only on the reply to the asking message. On every other reply a sensitive answer is still never repeated. It is also still kept out of the running summary, the notes and every alert: the alert to your team says some details are private and shown only on the lead's page.

The bot recognises the request from the AI's own reading of the message and from a short list of English and Hindi phrasings.

When your bot does not know

If a lead asks something no fact covers, your bot says so honestly and carries on. For example:

Good question — let me check that with the team and come back to you. Could you also tell me about your budget?

It keeps the conversation moving, like a receptionist who says "I'll find out, and meanwhile…". It adds no question for a lead who is being passed to a person (a complaint, a support request, an emergency), nor once the intake is finished. A lead whom your rules turned away is not told anyone will check: they get the closing message instead. The sentence is worded differently if the lead has already heard it.

The same words go out if the check refused a reply twice, for example when the bot was about to state a price you never wrote. What your team sees depends on the question:

A price question is flagged on the lead as Asked about price.
An availability question is flagged as Asked about availability.
Anything else is noted on the lead as Couldn't answer, quietly, with no alert.

The lead has been told someone will check, so your team follows up. Your bot does not go back to the lead by itself.

Questions your bot couldn't answer

Every question your bot could not answer is kept, once, on the same page, under Questions your bot couldn't answer. Each line shows the question, what it is about, how many leads asked it, and when it was last asked.

Answer opens the form with the question as the title. Write the answer once, press save, and it becomes a fact. Your bot knows it from the next conversation on.
Not now sets the question aside. It comes back if another lead asks it again.

Questions asked by several leads are shown once. The list also appears as a card on your bot's to-do, "Answer what your bot couldn't", so you will see it without looking for it (Your bot's to-do).

A question counts as covered when your bot finds a fact that answers it, in whatever language the lead asked, or when one of your facts shares a meaningful word with it. A question your bot answered from a fact is not kept here, so the list can show fewer questions than leads asked.

Keeping the facts true

A fact you delete stops being used straight away. A lead who asks is told you will check.
Check prices after a change. Your bot repeats a price exactly as written, so an old fact is an old price.
Write it short and plain. The reply is limited to about three sentences, and a fact that is a paragraph of small print will not fit in one.
A long list is fine. When your facts do not all fit in one reply, the bot is given the ones that match the lead's question (and always your hours or places when asked about them).
Safe mode turns facts off altogether, your weekly hours included, for a bot you have decided to rein in (how it works).

How a reply is made, step by step, and what Safe mode does: How the bot builds its conversation.
Telling your team about price and availability questions: When the bot hands a lead to a person.
Tone and length: How your bot sounds.
Selling points, objection answers, testimonials and upsells: How your bot sells.