> ## Documentation Index
> Fetch the complete documentation index at: https://aitextura.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Persona, Knowledge Base & Training

> Setting up instructions, tone, scripts, knowledge base, self-learning and quick answers

Setting up instructions, tone, scripts, knowledge base, self-learning and quick answers

***

## PERSONA: INSTRUCTIONS, TONE, SCRIPTS

<AccordionGroup>
  <Accordion title="How is the Persona structured?">
    The Persona is a block editor. Each block controls a separate aspect of the AI's behaviour:

    * **Role** — who the AI is and what business it works in.
    * **Tone** — communication style, formality, response length.
    * **Instructions** — behaviour scenarios, scripts.
    * **Prohibited actions** — what the AI must never do under any circumstances.
    * **Additional instructions** — any clarifications and exceptions.
    * **Script** — ready-made conversation scripts.
      Blocks can be nested inside one another to create a hierarchy of rules.
  </Accordion>

  <Accordion title="How do I describe the AI employee's role correctly?">
    Fill in the "Role" block using the following structure:

    1. State who the AI is (1–2 sentences).
    2. State what business/property it works in.
    3. Define its main task.
       Example: "You are a booking manager at the Revelton Suites hotel in Riga. Your task is to advise guests and help with room bookings."
  </Accordion>

  <Accordion title="How do I configure the communication tone (formality, formal address)?">
    Go to the "Tone" block and specify the following:

    1. Address style: "Use formal address (You/Sir/Ma'am)".
    2. Register: "Formal / informal".
    3. Length: "Short responses, up to 40 words".
    4. Emoji: "Do not use emoji unless the guest uses them".
       You can copy this example and adapt it to your brand.
  </Accordion>

  <Accordion title="How do I make responses shorter?">
    Follow these steps:

    1. In the "Tone" block, add: "Reply as briefly as possible, 1–2 sentences, to the point only. No extra details or closing filler phrases."
    2. In the employee's "Additional settings", set a word limit if that option is available in your version of the account.
  </Accordion>

  <Accordion title="How do I remove filler phrases like “If you have any more questions, feel free to ask”?">
    Go to the "Prohibited actions" block and add the line:
    "Do not end every message with an invitation to ask more questions."
    After saving, click "Retrain" — the change will take effect.
  </Accordion>

  <Accordion title="What language should I write instructions in?">
    It is recommended to write instructions in **English** — this improves the accuracy with which rules are followed. Use the following practice:

    1. Start the instruction with the line: "Always respond in the language the guest writes in".
    2. Describe scenarios in English.
       If the instructions are written in Russian, the AI may occasionally switch to Russian when communicating in other languages — this is a quirk of the language model that English instructions eliminate.
  </Accordion>

  <Accordion title="Are there any limits on instruction length?">
    There are no technical limits, but follow this recommendation:

    * Keep instructions within **500–1000 words**.
    * Put everything else (price lists, room descriptions, FAQ) in the **Knowledge Base**.
      The longer the instruction, the more tokens are consumed per request, which affects cost. Moving data to the Knowledge Base optimises expenses.
  </Accordion>

  <Accordion title="Can I store instructions in Google Docs?">
    Yes. Here's how:

    1. Create a document in Google Docs and write your instructions.
    2. In the Persona editor, use the function to import blocks from Google Docs.
    3. Every time you update the document in Google Docs — return to the employee card and click "Retrain".
  </Accordion>

  <Accordion title="How do I prevent the AI from answering off-topic questions?">
    Use both methods together for a reliable result:

    1. In the "Instructions" block, add: "Only answer questions related to bookings and property services. For other topics, politely decline: say you cannot help with that and suggest returning to the subject."
    2. In the employee's settings, enable the **"Protection"** option — it restricts topics at the system level.
  </Accordion>

  <Accordion title="How do I prohibit certain actions (confirming cancellations, financial operations)?">
    Go to the "Prohibited actions" block and specify the restrictions explicitly. Example:

    * "Never confirm a booking cancellation on your own."
    * "Never promise a refund."
    * "For such requests — tell the guest that a manager needs to handle this, and transfer the chat."
      The more specific the prohibition, the more accurately the AI will follow it.
  </Accordion>

  <Accordion title="The AI gets the gender wrong (writes in the feminine form instead of masculine).">
    This is a configuration detail — the role needs to be specified explicitly. Follow these steps:

    1. In the "Role" block, add: "You are a male hotel administrator."
    2. If the mismatch persists — also add to "Prohibited actions": "Never use feminine forms when referring to yourself."
    3. Click "Retrain".
  </Accordion>

  <Accordion title="How do I configure the AI to first clarify needs rather than immediately listing all rates?">
    In the "Instructions" block, add a sales script:
    "Before offering options, ask clarifying questions: check-in/check-out dates, number of guests, preferences. Only then suggest 2–3 of the most suitable options. Do not list all rates at once."
  </Accordion>

  <Accordion title="How do I configure a transfer to a manager for group bookings?">
    In the "Instructions" block, add the rule:
    "If the group is more than 8 people or the request involves a corporate event — do not calculate the cost yourself, transfer the request to a manager."
    By default, booking two or more rooms simultaneously is also redirected to a manager — check whether this setting is enabled in "Talents" → PMS integration settings.
  </Accordion>

  <Accordion title="How do I configure discounts and promotions?">
    In the "Instructions" block, create a separate section for each promotion with the following structure:

    1. Promotion name.
    2. Conditions of application (dates, guest categories).
    3. Exclusions.
    4. What to do when requested.
       Example: "15% birthday discount. Does not apply to additional services. When requested — ask for the guest's details and pass to a manager for processing."
  </Accordion>

  <Accordion title="How do I configure information about children / surcharges?">
    In the "Instructions" block, create a separate section by age category:

    1. Specify each age group and its conditions.
    2. Define the information-gathering rule.
       Example: "Under 5 — free, no extra bed. 5–12 — surcharge of 1,000 RUB/night. Always ask for children's ages when processing a booking request."
  </Accordion>

  <Accordion title="How do I configure a “weekend minimum 2 nights” rule?">
    Add a separate block to the "Instructions" section:
    "On Fridays and Saturdays the minimum stay is 2 nights. If a guest requests one night on these days — do not say 'no availability', explain the minimum stay rule and offer to find suitable dates."
  </Accordion>

  <Accordion title="Where can I find templates / examples of good instructions?">
    Ready-made templates are available directly in the Persona editor:

    1. Open the Persona editor in the employee card.
    2. Click the "/" button — the **Rules Hub** will open, with ready-made blocks from AI TEXTURA for various scenarios.
    3. Select the appropriate block and insert it into your instructions.
       If you need an example for a specific niche — request one from an AI TEXTURA methodologist.
  </Accordion>

  <Accordion title="Can I import blocks from the Rules Hub?">
    Yes. Follow these steps:

    1. In the Persona editor, click "/" and open the Rules Hub.
    2. Select the desired block and insert it into your instructions — it will be added as an independent copy.
    3. Edit the inserted block for your needs without any restrictions.
       Please note: if a block in the Rules Hub is updated, the inserted copy will not be updated automatically. Keep track of relevance manually.
  </Accordion>
</AccordionGroup>

***

## KNOWLEDGE BASE AND DATA SOURCES

<AccordionGroup>
  <Accordion title="What is the Knowledge Base and why do I need it?">
    The Knowledge Base is the central information repository that AI employees use when generating responses. Key benefits:

    * A single source can be connected to multiple AI employees at once.
    * Data is stored separately from instructions, keeping the Persona cleaner and saving tokens.
      Add everything the AI should know to the Knowledge Base: price lists, property descriptions, policies, FAQs.
  </Accordion>

  <Accordion title="Which file formats are supported?">
    The following formats and sources are supported:

    * Files: PDF (text layer only — not scans), DOCX, TXT, CSV.
    * Manual text input.
    * Links: web pages, YouTube videos (the system transcribes the audio).
    * Cloud: Google Docs, Google Sheets.
      **Important:** Word files (.docx) should be saved as a Google Doc first for more reliable recognition.
  </Accordion>

  <Accordion title="Is data from files updated automatically?">
    No. Data in the Knowledge Base is static. When information changes:

    1. Go to "Knowledge Base" → find the relevant source.
    2. Delete the outdated file.
    3. Upload the updated version.
    4. Click "Retrain" in the employee card.
       Exception: **Google Sheets connected via "Talents"** — they update dynamically, with no manual action required.
  </Accordion>

  <Accordion title="How do I connect a Google Spreadsheet?">
    Follow these steps:

    1. In the employee card, go to "Talents" → select "Google Sheets".
    2. Enter the SpreadsheetID of your spreadsheet (see the next FAQ entry).
    3. Make sure the spreadsheet has a "flat" structure: clear column headers in the first row, no merged cells, no nested tables.
    4. In the "Instructions" block in the Persona, specify when and for what purpose the AI should refer to this spreadsheet.
  </Accordion>

  <Accordion title="How do I find the SpreadsheetID of a Google spreadsheet?">
    The SpreadsheetID is in the spreadsheet URL between `/d/` and `/edit`. Follow these steps:

    1. Open the Google Spreadsheet in your browser.
    2. Copy the page address from the address bar.
    3. Find the fragment in this format: `https://docs.google.com/spreadsheets/d/**[YOUR_ID]**/edit`
    4. Copy only the ID — the string of characters between `/d/` and `/edit`.
       Example: in the link `https://docs.google.com/spreadsheets/d/1X4YNQ10dgk.../edit` — the ID is `1X4YNQ10dgk...`.
  </Accordion>

  <Accordion title="Can the AI read information from a website?">
    Yes. To connect a website as a source:

    1. In the "Knowledge Base", click "Add source" → select "URL".
    2. Paste the link to the relevant page.
       Keep the following in mind:

    * The AI only reads text content — images are not recognised.
    * If the website is protected against automated data collection, the information will not be retrieved. In that case, copy the text manually and add it as a text source.
  </Accordion>

  <Accordion title="The AI is pulling information from the wrong source (mixing up data from different sources).">
    This is a data conflict that can be resolved in the settings. Follow these steps:

    1. Check whether there are any contradictions between different sources in the Knowledge Base (e.g. the same price listed differently in two places).
    2. Check whether the "Instructions" block contains an explicit instruction: when and from which source the AI should take data.
    3. In the chat with the bot, use the **"Response analysis"** button — it will show which source was used when generating the response.
       If the behaviour persists after these steps — contact support with an example question, the response you received, and the source you expected the answer to come from.
  </Accordion>

  <Accordion title="Deleted documents are still showing in the account.">
    This is a known interface display quirk that does not affect how the AI works — deleted sources are no longer used when generating responses. To verify:

    1. Use "Response analysis" in the chat to check whether the AI is referencing that source.
    2. If the visual "ghost" of the document is bothering you — refresh the page or log out and back in to your account.
       If the document is still showing after these steps — contact support with the document name and the section of the account where it appears.
  </Accordion>

  <Accordion title="How many files can I add to the Knowledge Base?">
    There is no limit on the number of files in the Knowledge Base. Add as many sources as needed for accurate AI employee responses.
  </Accordion>

  <Accordion title="Which is better — PDF or Google Docs?">
    **Google Docs** is recommended as the primary format. Here's why:

    * Google Docs is reliably recognised by the system.
    * PDF works correctly only if it contains a text layer (not a scan or photo of a document).
      If your data is currently in PDF:

    1. Open the PDF.
    2. Copy the text.
    3. Create a Google Doc and paste the text there.
    4. Add the Google Doc as a source in the Knowledge Base.
  </Accordion>
</AccordionGroup>

***

## SELF-LEARNING (FAQ)

<AccordionGroup>
  <Accordion title="How does self-learning work?">
    The system automatically analyses real conversations and extracts question–answer pairs from them. Here's how to work with it:

    1. Go to the **"FAQ"** section in your account.
    2. You will see accumulated records with the status "New".
    3. For each record, choose an action: approve, edit, or delete.
    4. After finishing moderation, click the **"Update knowledge base"** button.
       Without clicking this button, approved answers will not be applied.
  </Accordion>

  <Accordion title="Why aren't approved answers being used by the AI yet?">
    Approving a record in the interface does not apply it automatically. Complete this step:

    1. Make sure you have clicked the **"Update knowledge base"** button after moderation.
       Only after that will the AI start using approved answers.
  </Accordion>

  <Accordion title="Why does self-learning collect ALL questions, not just those without an answer?">
    This is a known characteristic of the self-learning algorithm at its current stage — the system does not yet filter questions based on whether an answer exists in the base; it collects all interactions. The platform team is working on optimising this behaviour. When moderating the "FAQ" section, simply delete records whose answers are already in the Knowledge Base.
  </Accordion>

  <Accordion title="When updating the knowledge base, unusual behaviour occurs (a notification about a quirk is shown) but questions are still added — is this normal?">
    Yes, this is a confirmed technical quirk of the update interface. Despite the unusual notification, questions are correctly added to the Knowledge Base. To verify:

    1. Go to the "Knowledge Base" and confirm that the new records are visible.
    2. Test the AI employee by asking one of the approved questions.
  </Accordion>

  <Accordion title="How do I enable strict FAQ mode?">
    Follow these steps:

    1. Go to the **"FAQ"** section in the employee's settings.
    2. Find the "Behaviour when no answer is found" parameter.
    3. Select one of the modes:

    * **"Answer independently"** — the AI generates a response based on the Knowledge Base.
    * **"Stay silent"** — the AI does not respond if there is no matching FAQ answer.
    * **"Send a template response"** — the AI sends a pre-written text (up to 450 characters).
  </Accordion>
</AccordionGroup>

***

## QUICK ANSWERS

<AccordionGroup>
  <Accordion title="What are Quick Answers and how are they different from FAQ?">
    Quick Answers are pre-configured reactions to specific phrases. They are checked before FAQ and have higher priority. Use them when you need to guarantee an exact response to a specific trigger — without processing through the knowledge base.
  </Accordion>

  <Accordion title="What action types are available?">
    You have 5 action types available for each quick answer:

    1. **Reply exactly** — the AI reproduces your text word for word.
    2. **Based on the answer** — the AI rephrases the given text in its own words.
    3. **Transfer to another AI** — the conversation is passed to another AI employee.
    4. **Get an answer from another AI** — the request is sent to another AI and the response is returned to the current conversation.
    5. **Expert council** — 2 to 10 AI employees are engaged simultaneously.
       Select the required type when creating or editing a quick answer in the **AI Employee → Quick Answers** section.
  </Accordion>

  <Accordion title="Are there any limitations?">
    Yes, keep two things in mind:

    1. The trigger question must be no more than **50 words**. If the phrase is longer — shorten it to the key part.
    2. The number of quick answers is **unlimited** — create as many as you need.
  </Accordion>
</AccordionGroup>