Prostay AIHow it writes

How it writes: Behavior and General Settings

Two screens decide how the assistant sounds: one sets tone, length and the phrases it may not use, the other sets the formats and hours it writes in. Here is every dial, which ones are worth your attention, and the one field that cannot currently be set correctly.

Where to find it
Prostay AIBehavior
Last checked
August 17, 2026

Two screens decide how the assistant sounds. Behavior sets the voice: how formal it is, how long it goes on, what it may not say. General AI System Settings sets the conventions it writes in: how it renders a date, a time and a price, and which languages it may answer in. They are covered together because nobody sets one without the other.

Almost everything here has a sensible default, and one control is worth more than all the dials combined. If you are short of time, skip to the free text box.

Two screens, one job

Behavior has seven cards and a single Save Changes at the top that commits all of them, so you can work down the screen and save once at the end. General AI System Settings works the same way.

Tone and length

The DETAILS PREFERENCES card on Behavior. Tone / Persona reads Friendly-Professional and Verbosity reads Balanced. On the right, Max tokens per reply reads 400 and a ticked checkbox below it reads Hard cap safeguard.
Two words for the voice and one number for the length. The number is a budget, not a target.
FieldWhat it does
Tone / Persona Three choices: Friendly-Professional, Luxury-Warm and Playful. Pick the one your own marketing sounds like. Friendly-Professional is the safe default for almost every property.
Verbosity Concise, Balanced or Detailed. Concise suits chat, where a guest is reading on a phone. Detailed suits email.
Max tokens per reply A ceiling on the length of one answer. Tokens are roughly three quarters of a word each, so 400 is a long paragraph or two short ones.
Hard cap safeguard Ticked, the ceiling is absolute. Unticked, it is a guideline the reply may exceed.

The token ceiling is a budget rather than a target, so a low number does not make answers terse, it makes long ones stop. If you want short answers, set Verbosity to Concise and leave the ceiling generous enough that a genuinely complicated answer can finish its sentence.

Formality and temperature

Two sliders side by side on Behavior. The left is labelled Formality (formality) with end labels Casual and Formal, and its handle sits a little past the middle towards Formal. The right is labelled Response Temperature (temperature) with end labels 0 and 1, and its handle sits about a third of the way along. Neither slider shows a number.
Formality is how it addresses a guest. Temperature is how much it varies from one answer to the next. Both labels carry the field name in brackets, which is the defect noted below.

Formality runs from Casual to Formal and decides how it addresses a guest, from a first name and a contraction at one end to Dear Mr Hart at the other.

Response Temperature runs from 0 to 1 and decides how much the wording varies between two answers to the same question. Low is repetitive and predictable; high is more natural and less controllable. Somewhere around 0.3 is a reasonable place to sit: recognisably human, without inventing a new way to describe your cancellation policy every afternoon.

Phrases it may not use

The BANNED WORDS / PHRASES card on Behavior. Default Language reads English. Below an input reading Type the word and press Enter, three chips read guaranteed upgrade, cheapest and no problem at all.
Phrases you do not want written on your behalf. Each one is a promise or a tone you would not sign off.

Type a word or phrase, press Enter, and it becomes a chip. The useful entries are not rude words, which were never going to come up. They are the phrases that commit you to something:

  • guaranteed upgrade, because nobody can guarantee one.
  • cheapest, because it is a claim about your competitors.
  • no problem at all, if that is not how your property speaks.

A good way to build the list is to think of the sentence you would not want to read in a complaint, and ban the phrase at the centre of it.

Domains it may not correspond with

The EMAIL DOMAIN NAME RESTRICTIONS card on Behavior, beside the banned words card. An input reads Type the word and press Enter, and two chips below it read competitor-travel.example and unverified-agency.example.
Domains the assistant will not correspond with. This is the one control on the screen that is about who it writes to rather than how.

The card beside banned words takes email domains rather than phrases, and it is the only control here about who rather than how. The obvious candidates are agencies and intermediaries you would rather a person handled, and any domain whose mail should always reach a human.

House vocabulary and closings

The PREFERRED TERMS / REPLACEMENTS table on Behavior, with an Add New button above it. Columns read Name, Preferred Term and Actions. Three rows: hotel becomes property, cheap becomes great value, problem becomes matter.
House vocabulary. The left column is what it might have written; the right is what you would rather it wrote.

Preferred Terms / Replacements is a two column list: the word it might reach for, and the word you would rather it used. This is where a house style actually lives, and it is worth more than the tone picker for a property with any kind of voice of its own. hotel becoming property, problem becoming matter: small substitutions, and between them they do most of the work of sounding like you.

The MANDATORY CLOSINGS card on Behavior. An input reads Type the word and press Enter, and one chip below it reads We look forward to welcoming you to The Marlowe Hotel.
A line you want on the end of everything. Worth keeping to one, because every reply will carry all of them.

Mandatory Closings takes lines you want at the end of every reply. Keep it to one. Every entry is added to every answer, so three closings produce a three line sign-off on a message whose body was one sentence, and a guest reads that as a form letter.

The box that does the most work

The EXTRA SYSTEM INSTRUCTIONS card on Behavior. A text area holds four sentences beginning Never confirm a rate that is not in the current price list, and ending Quote check-in as 3pm and check-out as 11am unless the reservation says otherwise. A counter below the box reads 267 / 2000.
The free text box, and the most useful control on the screen. Everything above it is a dial; this is a sentence.

Extra System Instructions is two thousand characters of plain English, and it is the most useful control on either screen. Everything above it is a dial with three positions. This is where you can express the things your property actually cares about, in the words you would use to brief a new member of staff.

The instructions that earn their place are the ones about your risks:

  • Never confirm a rate that is not in the current price list.
  • If a guest asks to change or cancel a booking, tell them the front desk will confirm it and leave a note on the reservation.
  • Quote check-in as 3pm and check-out as 11am unless the reservation says otherwise.

Dates, times and money

The Display Preferences card on General AI System Settings. AI Name reads Marlowe Concierge, AI Time Format reads 11:59 PM (12 HOURS TIME), AI Date Format reads MM/DD/YYYY, AI Currency Format reads 1,000.00, AI Time Zone reads (UTC-7:00) America/Los Angeles and AI Primary Currency reads USD.
How the assistant writes a date, a time and a price. Set these to match the property, not the head office.

These are the conventions the assistant writes in, and the reason to care is that a guest reading 06/07 needs to know whether that is June or July.

FieldWhat to set it to
AI Name What the assistant calls itself. Something that sounds like your property rather than like software.
AI Date Format The one your guests read, which usually means the country the hotel is in rather than the country your office is in.
AI Time Format 12 or 24 hour. Match how your own signage and confirmations are written.
AI Time Zone The property's own zone. This is what makes tomorrow mean the right day.
AI Currency Format and AI Primary Currency The separator style and the currency prices are quoted in.

Hours and languages

The AI Behavior card on General AI System Settings. Cut Off Time reads 07:00AM to 10:00PM. Response Time reads 11:59 PM (12 HOURS TIME). Primary Language reads English. Multi-Select Language shows English, Espanol and Francais. Tone Of Voice AI reads Professional - Friendly.
The hours it answers in and the languages it will answer in. Response Time offers clock times, which is the defect noted below.

Cut Off Time is a pair of times bounding the hours the assistant works in, read in the time zone above. Primary Language is the language it answers in by default, and Multi-Select Language is the set of others it may answer in, as removable chips. Add the languages your guests actually arrive speaking; there is no cost to a language nobody uses, and a guest who writes in Spanish and is answered in English usually stops writing.

Tone Of Voice AI here overlaps with Tone / Persona on Behavior. Set them consistently rather than trying to work out which one wins.

Where to go next

The voice is set. The remaining screen is the one your guests actually see.

Common questions

  • What is a token, and is 400 a lot?

    A token is roughly three quarters of a word, so 400 tokens is about 300 words: a long paragraph or two short ones. It is a ceiling rather than a target, so raising it does not make answers longer. If you want shorter answers, set Verbosity to Concise instead.

  • What should I set Response Temperature to?

    Around 0.3 is a reasonable starting point: the wording varies enough to read naturally without reinventing your policies. Move it down if you want near-identical answers to the same question, and up if replies read stiffly. It is not a quality dial, so higher is not better.

  • Why does the Formality slider not show a number?

    Because the value only appears while you are touching the handle. Nudge it and a label appears above it, showing Casual, Semi-Formal or Formal rather than a number, then put it back where it was.

  • Response Time only offers clock times. What am I missing?

    Nothing. The field is populated from the wrong list, so every option is a clock format rather than a duration and none of them is a valid response time. Leave it alone and use Cut Off Time for the hours you want the assistant working. We have reported the defect.

  • Tone appears on both screens. Which one wins?

    Set them to the same thing rather than relying on precedence. Tone / Persona on Behavior and Tone Of Voice AI on General AI System Settings describe the same intention with different option lists, so the reliable approach is not to make them disagree.

  • Can I add more than one mandatory closing?

    You can, and it is usually a mistake. Every entry goes on the end of every reply, so two closings produce a two line sign-off on a one line answer. Keep one, and put anything else in the signature on Email Account Settings if it belongs on email only.

Was this article helpful?

Related articles

Still stuck?

Support answers from inside the app as well, so if you are already signed in you will get a faster reply there. Otherwise send us the property name and what you were trying to do.