Sales agent

The Sales agent section sets the identity and personality of the assistant that replies to your leads. This is who the lead thinks they are talking to, so the name and tone you set here are what the lead sees.

Overview#

Everything in this section feeds the assistant's persona. The assistant introduces itself with the name you choose, replies in the languages you pick, and adopts the tone you describe.

• Fields are free text unless noted.
• They save on their own: shortly after you stop typing, or immediately when you click away. A marker next to each label shows Saving..., then Saved, or "Not saved. Try again" with a Try again link if a save fails. What you typed stays in the field until it saves.

What you can do#

Identity#

• Name: the assistant's name. This is the name it uses with leads, so set it to a real-sounding person, not "Gene".
  • It also signs Gene's emails. Every email Gene writes or sends to a lead (replies, welcome steps, follow-ups, monthly updates, outreach, B2B outreach, project pushes, property-portal welcomes, appointment confirmations, reminders, reschedule and cancellation emails) opens with a greeting such as "Hi Reem," and ends with a sign-off: "Best regards," then this name and your company name (the Common name under Company). Both are in the lead's language, and are only added when the message does not already have them. A welcome email with its own HTML design (see Nurturing) is sent as designed, without them.
  • A few emails sign with your company name alone: cold email sequence steps (they go out from your mailbox personas, so the assistant's name would not match the sender) and the email Gene sends when a WhatsApp message cannot be delivered.
  • An email you ask the Gene sidebar to send is signed with your own first name, then the company name.
  • Listing Newsletters, HTML template steps and emails you type yourself in the inbox go out exactly as written.
• Job title: the assistant's role, for example "Sales consultant". The assistant references it when it introduces itself.
• Location: where the assistant is based. It helps the assistant sound local and answer "where are you" questions.

Languages Gene replies in#

A searchable multi-select. Pick one or more of the offered languages, or type your own to add a custom entry.

• All languages sits at the top and stands alone: picking it clears the others, and picking a language clears it.
• The hint reads "Leads writing in another language still get a reply, in one of these".
• With All languages, the assistant replies in whatever language the lead writes in. It never asks to switch to English or another language, and never says it does not speak theirs.
• With a list, the assistant answers in the lead's language when it is on your list. Otherwise it asks once whether it is OK to continue in your first listed language ("Would it be OK to continue in English as I don't speak German?"), and carries on in it. It never lists the languages it speaks.
• A first message that is only a greeting ("Hi", "Ok", "Ciao") says nothing about the lead's language. The assistant opens in English when English is on your list (or with All languages), otherwise in your first listed language, then follows the rules above once the lead shows their language.
• The whole reply is in that language, the greeting included. Job titles (the assistant's own and your senior team's, set in English on these pages) are translated naturally, while people's names, your company name and project names stay as you wrote them.
• Language is never a reason to hand a lead over.
• The same setting decides the language of the other messages Gene writes for you: the welcome it writes when a form or ad lead comes in, follow-up templates sent past WhatsApp's 24-hour window, and the nudge when a known lead inquires again. A lead whose language is not on your list gets your first listed language.

Tone of voice#

A three-stop slider: Casual, Balanced (default) or Formal.

• A caption explains the selected stop.
• An example box ("How your agent sounds") previews a reply written in your own business context.
  • It is rewritten only when you move the slider or change the agent name, company name, specialisation or office location. Editing other fields, or leaving the page and coming back, keeps the same example.
• If you had a custom free-text tone from the old layout, it is shown underneath. Until you pick a level, the assistant keeps using it. Once a level is picked, the note says your custom tone is no longer used, and Use my custom tone again switches back to it (it saves straight away).
• At every stop, only the assistant's first message in a conversation opens with a greeting, in the lead's language ("Hi Dave," at Balanced, "Hey" at Casual, "Good morning" at Formal in English; "Здравствуйте" or "Hola" for a Russian or Spanish lead). Later replies go straight to the point. If a reply still starts with a greeting less than 12 hours after the last message from your side on that thread, the greeting is cut before it sends. Emails keep their greeting and sign-off.

Sample conversation#

Paste (or upload as a .txt or .csv file up to 100 KB) a real exchange you liked, up to 6,000 characters.

• The assistant imitates its style only.
• It never quotes the sample, or reuses facts, names or prices from it.
• It only shows when it is turned on for your account.

Extra rules#

Structured rules the assistant follows in every conversation, in three shapes:

• Your rules are the last thing the assistant reads before the conversation, and they carry the highest priority: where one conflicts with Gene's built-in guidance (booking pacing, follow-up style, handover habits), your rule wins. Before each reply the assistant checks its draft against every rule that applies. The one thing a rule cannot change is honesty: the assistant still never claims an action that did not happen or a fact it was not given.
• A rule to ask for something is already met when the lead has given it. If a lead filled in a form with their budget, timeline or area, an Always rule like "ask the budget as an open question first" does not make the assistant ask again: it uses their answer, and at most confirms it in passing ("you mentioned 400,000 to 500,000 on the form").
• A lead who asks for a call or a meeting gets times quickly. A rule like "qualify motive, budget, timeline and location before offering any meeting time" is met by one meeting-prep question about something still unknown (or none, when nothing is missing). Then the assistant offers two times, and the rest of the qualifying happens after a time is picked. A rule that says not to book with some leads at all (sellers, partners) still holds. See When a lead asks for a call.
• A rule that names a step your workspace does not have (for example "invite them to WhatsApp" while Move leads to WhatsApp is off) is skipped. The assistant never says that step happened and never hands the lead over because of it. It helps the lead on the channel they wrote on instead.
• Add up to 50 rules. Each rule (or each half of an If-then rule) can be up to 1,000 characters. A character counter appears under a rule once it passes 800, and past 1,000 it turns red and says the rule is too long to save: shorten it or split it into two rules. Each change saves on its own, with a Saving... / Saved marker beside the heading, or "Not saved. Try again" with a Try again link if a save fails.
• There is no separate Spelling field. A spelling convention is an Always rule, for example "Write in British English spelling."
• A spelling you picked in the old Spelling field was turned into an Always rule for you.
• Ask Gene AI in the sidebar to point at one rule ("show me rule 3", or "where is the rule about discounts?") and it scrolls to that rule and rings it, not the whole list. Rules are counted from the top. The same works for one FAQ, routing rule, pipeline stage, field or welcome flow message.

Other instructions#

• Extra instructions: a multi-line field for any extra rules or guidance you want applied on top of everything else.
• If your Extra instructions are very long, put the rules Gene must check on every reply in Extra rules above. Those rules always close the assistant's instructions with top priority, however long the rest is. Keep supporting detail in Extra instructions.
• First-reply guidance: a multi-line field for the first message a new lead gets. It applies to two messages:
  • the welcome Gene writes when a lead comes in through a website form or an ad, before the lead has written anything (this is also where you set that welcome before any lead source is connected). Gene only writes the welcome when the source's first message on its Welcome flow tab is empty; a typed first message goes out exactly as written, and
  • Gene's first reply to a new lead, including its reply to a lead who answers the welcome.

  A lead source's own AI instructions (Flows > Lead sources) are more specific and win over it where the two disagree. The welcome shows in the conversation like any other message from Gene.

Confirm form answers before using them#

A switch on the Sales agent tab, under Relaxed punctuation. It is off by default, and then a lead's form answers count as already answered.

Turn it on when your team wants every form answer checked with the lead first:

• The assistant treats what a lead typed on a form (your website, a Gene form, a Meta lead form, a landing page sending leads to a webhook, or a portal enquiry) as not confirmed yet.
• It confirms each answer briefly, one per message, in the order of your Topics to ask about (AI > Conversion), for example "I see you noted a budget of US$800,000 to 950,000, is that correct?". It uses an answer only once the lead confirms it, and if the lead gives a different answer, it uses theirs.
• Until the lead confirms, a form answer does not make the lead warm or hot, and the form's timeline or budget is not saved as the lead's Timeline, Budget or Timeframe. The answers stay on the contact exactly as the lead typed them.
• The welcome Gene writes for a new form lead follows it too: it never states a form answer as settled.

No dashes in what Gene sends#

Gene never sends a lead a dash used as punctuation: no em dash (—), no en dash (–) and no hyphen with a space on both sides (" - "). Each one becomes the comma, full stop or colon a person would have typed. There is nothing to switch on, and an instruction asking for dashes does not bring them back.

• It covers everything Gene writes or sends for you, on every channel: replies, follow-ups, welcome messages, outreach and other flow messages, the WhatsApp templates Gene submits to Meta, AI drafts and translations in the inbox composer, and messages the sidebar sends on your behalf.
• Hyphens inside words, numbers, dates, phone numbers, links and variables stay as they are ("1-bedroom", "2026-10-03", {{first_name}}).
• A range keeps a plain hyphen: "38–100 sqm" goes out as "38-100 sqm".
• A message you type and send yourself from the inbox composer goes out exactly as you typed it.

Checks before a reply goes out#

Before the assistant sends a reply to a lead, it runs a few checks of its own. There is nothing to switch on.

• No internal sources. The assistant never tells a lead where an answer came from in system terms: no "our FAQ lists...", "according to our records", "the live search didn't return..." or "with the current filters". It says it the way a person would ("Our advisor will confirm the exact handover date", "I don't have that one available right now"). If a phrase like that still slips into a draft, it is rewritten or taken out before sending. The welcome Gene writes for a new form or ad lead gets the same check.
• No tool-call text. The assistant looks things up and acts through internal tools (searching units, saving notes, handing over). The lead never sees that machinery: no JSON like {"area":"Uluwatu"}, no function names, no notes the model writes to itself. If the model types a tool call into its message instead of making it, Gene runs the call for real or takes the text out and asks for the reply again. Every automated send, including flow messages, gets the same check, and a message that would contain nothing but tool-call text is not sent at all.
• No asking what a form already answered. A lead who came in through a form (your website, a Gene form or a Meta lead form) has already given their answers, such as budget, timeline and area. The assistant reads them as the lead's own words and does not ask for them again. If you would rather it checks them first, turn on Confirm form answers before using them.
• No re-asking a skipped question. If the assistant asked something in either of its last two replies and the lead passed over it, it does not ask it again, even reworded. It answers what the lead said, or asks for a different missing detail. A draft that repeats the question is rewritten once, and if it still repeats it, the question is cut from the reply.
• Nearest real options. When nothing matches a lead's request exactly, the assistant shows the closest one or two real options it found, with their price and what each misses, before asking what the lead could change. It says how each differs in plain words ("this one is furnished", "it's in Seseh, right next to Canggu"), never in terms of the search or its filters. When the lead gave a budget, it only offers units that have a price, so an unpriced unit is never put forward as a fit for that budget.
• Your rules first. The draft is checked against your Extra rules.

Conversation basics in every workspace#

A few habits are built into every assistant, so you do not need to write them into your Extra rules.

• "Lease" can mean selling. In markets like Bali, an owner who wants to "lease" a property, mentions a leasehold or gives the years left on it often means selling the leasehold, not renting it out. When that is unclear, the assistant asks once: "Would you like to sell the leasehold, or rent the property out?" It never turns down a leasehold sale because it read it as a rental.
• Budgets in another currency. When a lead gives a budget in another currency ("AUD 150,000", "S$500k"), the assistant gives the converted figure once ("AUD 150,000 is about US$97,000"), then keeps talking about the budget in the lead's own currency. Property prices are never converted: they are always stated exactly as listed, in their own currency.
• Rupiah by default in Indonesian. Bare figures from a lead writing in Indonesian ("800", "1 M", "800 juta", "2 miliar") are read as rupiah unless the lead says otherwise. Once a lead has said which currency they mean, the assistant does not ask again.
• A warm close on a goodbye. When a lead signs off ("thanks", "terima kasih", "ok bye"), even right after saying something is over their budget, the assistant replies with a short, warm close in their language that leaves the door open. No question, no new offer, no new unit.

Tips#

• Use a real first name. The assistant must never present itself as "Gene" to a lead.
• Pick the tone stop that matches your market and check the example box; Balanced fits most teams. For hard rules about wording, use Extra rules instead.
• Set Languages Gene replies in to match the markets you actually serve, so the assistant replies in the lead's language. Pick All languages if it should answer every lead in their own language.

---

Details#

Every control in the Sales agent section (/ai#assistant), for when you need to look something up. The sections above explain what each one does; this list adds the control types, the full language list, and the switches not covered there.

• Name, Job title, Location: text fields.
• Languages Gene replies in: searchable multi-select, one or more, custom entries allowed. The options are All languages, then Arabic, Cantonese, Czech, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Malay, Mandarin, Norwegian, Persian, Polish, Portuguese, Romanian, Russian, Spanish, Swedish, Tagalog, Thai, Turkish, Urdu, Vietnamese.
• Extra rules: rule rows in three shapes, Always, Never and If-then (see Extra rules).
• Extra instructions: multi-line text.
• First-reply guidance: multi-line text, with the hint "Used for the welcome Gene writes when a form or ad lead comes in, and for its first reply to a new lead. A lead source's own instructions win over it."
• Tone of voice: three-stop slider, Balanced by default.
• Sample conversation: paste box with a file upload, only when turned on for your account.
• Allow emoji (toggle): lets the assistant use the occasional emoji. Off keeps replies text-only.
• Relaxed punctuation (toggle): a looser texting style, with lowercase starts and skipped end-periods where natural.
• Confirm form answers before using them (toggle, off by default): the assistant checks each answer a lead typed on a form with them before using it. See below.
• Reply timing (two-handle range slider): the window the assistant waits before replying, 30 seconds to 5 minutes. See Message Flow.
• Voice notes: appears here only when it is turned on for your account (off by default). Enable voice notes lets the assistant send some replies as a voice note instead of text. How often sets the mix (every message, or about one in 2, 3, 5 or 10, or a number of your own), and Voice picks the voice, with gender and accent filters and a preview you can play first. You can override it per lead from the conversation in the Inbox.