Skip to main content

Understanding the Organization Prompt in Fullview AI

Give the Fullview AI assistant the right starting point, every time.

A
Written by Alex
Updated this week

1. What the Organization Prompt is

The Organization Prompt (shown in Settings → AI Agent → “General description of your product”) is the very first piece of context Fullview AI reads before it processes a customer’s question. Internally, we inject this text as a system‑level instruction to the large‑language model (LLM). It tells the model:

Purpose

Why it matters to the AI

Product context

Explains what the application does, its core objects (e.g., invoices, clients) and the general purpose of the application you are running Fullview AI on.

Scope guardrails

Lets the AI know what is in‑scope (feature questions, onboarding steps) and what is out‑of‑scope (billing inquiries, roadmap speculation, legal advice).

Tone and persona (optional)

Sets expectations for voice, formality and brand style.

If you think of every AI response as a conversation, the Organization Prompt acts as the opening statement, it frames everything that comes next.


2. How Fullview AI uses the prompt in the reasoning pipeline

  1. Read the Organization Prompt

  2. Receive the customer’s message

  3. Fetch matching Help‑center articles / human training (if applied)

  4. Analyze what is seen in the DOM of the application

  5. Check confidence score (compare the model’s answer confidence to your Minimum confidence score slider)

  6. Generate or defer

    • If confidence ≥ threshold → answer the customer and guide them in the application.

    • If confidence < threshold → ticket is flagged to the AI Library for human review and escalated to human according to the logic you have set in your helpdesk software integration with Fullview AI.

Because step 1 sets the scene, a well‑written prompt directly boosts answer accuracy and reduces false deflections.


3. Writing an effective Organization Prompt

Do

Why

Keep it concise (≤ 1,500 characters)

Long prompts dilute key facts and increase token usage.

State primary user jobs‑to‑be‑done

E.g., “Users track cash flow, create and filter invoices, and manage client profiles.”

List major entities & synonyms

invoice ↔ bill, client ↔ customer, etc.—helps the model map vocabulary.

Define supported actions

“The AI may guide clicks, highlight UI elements, or fill forms on the user’s behalf.”

Call out exclusions

“Do not answer contract, pricing or legal questions - escalate instead.”

Mention tone if important

“Respond in a friendly, professional voice; keep answers <150 words.”

Don’t

Impact

Bury the lead in marketing slogans

Increases hallucination risk.

Reveal confidential roadmap or PII

Introduces security risks


4. Example prompts

Quality

Sample text

Notes

✔ Good

“You are assisting users of Fintech App, a SaaS where SMBs track cashflow, view & filter client invoices, and create new invoices. Users can… Do not discuss pricing or roadmap.”

Clear, scoped, brief.

✖ Poor

Fintech App is the best end‑to‑end visionary solution changing the future of finance… Ask me anything!”

Marketing fluff, no scope boundaries.


5. Maintaining the prompt over time

When to update

Trigger examples

Major new feature ships

Added new "Payroll" product in a new product tab

Support policy changes

AI may now handle billing tier questions.


6. Frequently asked questions

Question

Answer

How long can the prompt be?

1500 characters

Does the confidence slider interact with the prompt?

Indirectly. A clearer prompt boosts the model’s confidence, so you can often raise the slider and still serve most questions.


7. Quick checklist before hitting Save changes

  • Covers what the product is and who it serves

  • Lists key entities & user actions

  • States in‑scope vs out‑of‑scope topics (e.g "do not answer legal questions"

  • Defines voice/tone (if needed)

  • ≤ 1,500 characters, plain language

Craft a solid Organization Prompt once, keep it fresh, and help Fullview AI consistently deliver accurate, on‑brand answers, leaving your team free to tackle the edge cases that truly need a human touch.

Did this answer your question?