Prompt

Here you will find all tips and tricks to create the perfect prompt for your use case.

In this article, you will learn how to write prompts for the Fonio telephone AI so that conversations sound natural, information is reliably captured, and special cases are handled cleanly.

Summary

• Write clearly and concisely. Imagine you are instructing a new employee exactly how to behave in different situations.

• Use sections with headings instead of long continuous text.

• Ensure each piece of information appears only once in the prompt.

• Think in if-then rules.

• Always provide an escape route if the AI does not understand or cannot answer something.

• Enforce duties and information that the AI must ask for or state, and prohibit sensitive answers and topics.

• Structure your prompt like a decision tree.

• Observe the limit of 100,000 characters. As short as possible, as long as necessary.

Why a good prompt is crucial

A well-structured prompt acts like a script for your AI assistant. It defines the goal, tone, flow, and boundaries. This improves recognition rates, shortens conversation duration, and reduces handovers to staff to cases where they are meaningful.
In short: The better your prompt, the better and more competent your AI assistant appears.

Technical framework conditions at Fonio

• Character limit: maximum 100,000 characters

• Structure: Organize your prompt into clearly visible sections

• Formatting: Use Markdown formatting to make your prompt clear and readable

• Language and region: Adapt address and speech style to your use case and callers

• Task focus: Prefer few, clearly defined tasks per assistant

• No comments: The entire prompt is used by the AI; you cannot comment or add text that the AI should ignore

As short as possible, as long as necessary. Remove everything that does not directly contribute to achieving the goal.

The building blocks for strong prompts

Structure your prompt into clearly separable sections. This makes behavior reproducible and easy to modify.

Example of prompt structure:

# Conversation flow
Describe the single goal of your assistant in one or two sentences.

# Behavioral rules
Specify exactly how the assistant should behave in different situations.

## If-Then rules
- If [Condition A], then [Action A]
- If [Condition B], then [Action B]
- If the request is unclear, you should ask an open follow-up question

# General information
- About you: You are an AI telephone assistant from [Company]
- Your name: Marie
- The company you work for: [Company]

# Frequently asked questions
- Question: Do you have parking?
  - Answer: Yes, directly in front of the building

# Sensitive topics
If the customer talks about the following topics, say that you cannot assist:
- Legal questions
- Medical advice and data
- Political topics
- Discounts
- Detailed offers
- Religion

Why you should use a decision tree when creating your prompt

At the beginning of a call, the AI does not yet know what the request is about. For a good prompt, it can be helpful to imagine the different possibilities like a decision tree with branches.

Define in the prompt how the AI should behave for each category of request. This way, it is always clear how to respond to each request.

There should always be an escape route if no specific request can be identified. Usually, a section called “Others” is created for this.

Checklist before saving

• Goal clearly described

• Sections present

• If-Then rules complete

• Duties and prohibitions defined

• Escape hatch with criteria and text

• Text checked for brevity

Testing prompts

1. Quick test: Three to five test calls with typical requests. Pay attention to interruptibility and data capture.

2. Check performance: Success rate per path, handover rate, completeness of mandatory fields, conversation duration.

3. Fine-tuning: Shorten formulations, optimize order, add missing paths.

Aussprache-Probleme in Prompts lösen

Eine häufige Herausforderung beim Erstellen von Prompts ist die korrekte Aussprache von Namen, Firmennamen oder Fachbegriffen durch die KI. Diese Anleitung zeigt, wie Sie Aussprache‑Probleme durch phonetische Schreibweisen im Prompt beheben können.

Vorgehen bei Aussprache-Problemen

1. Phonetische Schreibweise verwenden:

• Ersetzen Sie die problematische Schreibweise durch eine phonetische Variante, die die gewünschte Aussprache erzeugt.

• Testen Sie die Aussprache mit der "Audio testen"-Funktion im Tool.

• Passen Sie die phonetische Schreibweise an, bis die Aussprache korrekt ist.

2. Konkrete Beispiele für phonetische Anpassungen:

• Firmenname "Reidl" → Schreiben Sie im Prompt "Raidl" für korrekte Aussprache.

• Produktname "subreport" → Schreiben Sie "sub-re-port" oder "suub-re-port".

• E‑Mail‑Endung ".de" → Schreiben Sie "De‑e" oder "D‑E" statt "de".

• "Stangl" = "Schtangl" oder "Müller, gesprochen 'Mü‑ller'".

Erweiterte Lösungsansätze

Wenn phonetische Schreibweise nicht ausreicht:

• Probieren Sie verschiedene Stimmen aus (unter "Technisches" → "Stimme & Sprache").

• Reduzieren Sie die Sprechgeschwindigkeit (unter "Technisches" → "Audio"), damit Wörter nicht zusammenrutschen.

• Platzieren Sie Aussprache‑Regeln am Anfang des Prompts unter "Verhaltensregeln", damit sie durchgängig greifen.

Zusätzliche Strategien:

• Bauen Sie Pflichtfragen für kritische Informationen ein: "Bitte buchstabiere deinen Nachnamen".

• Nutzen Sie die Fachbegriffe‑Funktion für wiederkehrende Namen oder Begriffe.

Integration in bestehende Prompts

Fügen Sie einen eigenen Abschnitt für Aussprache‑Regeln in Ihren Prompt ein:

## Aussprache-Regeln
- Sprich "Firmenname" immer als "Firmen-name"
- Sprich E‑Mail‑Adressen mit ".de" als "Punkt De-e"

Tipp: Ergänzen Sie Ihre Prompt‑Checkliste um den Punkt "Aussprache‑Hinweise für schwierige Namen oder Begriffe hinzugefügt und mit der 'Audio testen'-Funktion überprüft".

Template package

The following templates are intended to support you in creating your prompt. You can also find them directly in the platform, where you can incorporate them into your prompt.

Examples:

• Intelligent answering machine

• Receptionist

• Appointment scheduling

• First level support

Handlebars in Fonio prompts

With Handlebars support, you can make your prompts more flexible and dynamic. In addition to simple placeholders like

{{name}}

object values, loops, and conditions are also supported. This allows you to adapt content contextually – for example, depending on the caller, call direction, or CRM data.

Handlebars makes your prompts dynamic, reusable, and context-sensitive. This lets you integrate logic directly into your text – without additional code.

Accessing values in objects:

Hello {{contact.firstName}}, I see you are calling from {{company.name}}.

Conditional logic:

{{#if hasAppointment}}
I see you already have an appointment scheduled today. Would you like to reschedule it?
{{else}}
I see no appointment for today. Would you like to make a new one?
{{/if}}

Different logic for incoming and outgoing calls:

{{#if isInbound}}
Hello! This is {{assistantName}} from {{company.name}}. Thank you for your call.
{{else if isOutbound}}
Good day! This is {{assistantName}} from {{company.name}}. I am calling because you made an inquiry.
{{else}}
Hello! This is {{assistantName}} from {{company.name}}.
{{/if}}

String comparisons:

{{#if (eq callerType "customer")}}
Welcome back! How can I help you today?
{{else if (eq callerType "prospect")}}
Hello! Nice to hear from you. Is this about a new offer?
{{else}}
Good day! How can I assist you?
{{/if}}

For questions or support, our team is available at support@fonio.ai.