Skip to main content
Lesson 3 of 6 — Create a simple FAQ-style Managed Topic.
A simple topic answers one question in one turn with no actions or follow-ups. overview

What “simple” means

  • Completes in one turn
  • Requires no clarification
  • Triggers no actions
  • Works the same way in Chat and Call
If a topic needs any of those features, it should be implemented as a complex topic instead. These will be covered in more detail in Level 2: Advanced of the PolyAcademy.

Is your idea a simple or complex topic?

Use this decision tree before writing anything:

Check your understanding

Before you start: define your intent

Answer these questions before creating the topic:
  • What is the exact question being answered?
  • Can a human answer it in one short response?
  • Can users reasonably phrase this question in several different ways?
If the topic answers more than one question, split it. Example: single intent
  • “What time do you close?”
Example: multiple intents
  • “What time do you close and do you have holiday hours?”
These should be two separate topics.

Step 1: Choose a precise topic name

The topic name is the strongest signal for retrieval. Good topic names:
  • Describe one intent
  • Are specific, not general
  • Use snake_case
Examples
  • closing_time
  • parking_cost
  • pet_policy
Avoid names that are broad or vague. Avoid
  • info
  • general
  • shop_questions
  • misc

Step 2: Write sample questions

Sample questions teach the retriever how users phrase requests. As a baseline, aim for at least 3 sample questions per topic. If the topic is broad or users phrase it in many different ways, include more — up to 20. Add variations that reflect real speech. Include:
  • Short questions
  • Polite phrasing
  • Informal or incomplete phrasing
  • Call-style filler words
  • Different sentence lengths
Example: closing_time sample questions
  • what time is closing
  • closing time
  • when do we have to leave
  • when do you shut
  • what time do we need to be out by
  • uh what’s the closing time
Focus on variety in language and structure, not slight rewordings. Think about the different ways real users might describe the same issue — whether they are confused, specific, vague, or using non-standard terminology.
Avoid:
  • Repeating the same sentence with small wording changes
  • Writing marketing-style language
  • Including answers

How retrieval works

The retriever compares the user’s utterance to the topic Name, Sample questions, and Content to find the closest semantic matches. The top matches are then sent to the LLM, which makes the final decision on which topic to use. These signals are not weighted equally. The topic name is the strongest retrieval signal, followed by sample questions, then content. This means a precise topic name has more impact on retrieval than a large number of sample questions. Because the Name is shown to the retriever, it should be semantically close to the intent. A vague name like general_behavior-payment may trigger on any payment-related utterance instead of only the specific payment issue you intended.

Step 3: Write the Content response

The Content field is the agent’s full response. It should be:
  • Short
  • Direct
  • Easy to understand when spoken
  • Complete on its own
Use this structure:
  1. State the answer clearly
  2. Optionally mention one next step
  3. Stop
Example: closing_time content
We open at 9am. Closing time is at 6 p.m.

If you'd like to hire the store for a late event, I can help with that.
This works because:
  • The answer comes first
  • The response ends cleanly
  • No action is triggered
Avoid:
  • Explanations or justification
  • Policy language
  • Multiple offers or options
Avoid
Closing time is at 6 p.m., which allows our cleaning team to prepare the store for the next day.
This reads well on a page but performs poorly.

Step 4: For now, leave Actions empty

For a simple Managed Topic:
  • Do not add Actions
  • Do not reference functions
  • Do not trigger handoff or SMS
If an action is required, the topic is no longer simple and should be rewritten as a complex topic.

Common simple Managed Topic patterns

Topic name: opening_hoursContent:
We are open daily from 7 a.m. to 6 p.m.
Topic name: pet_policyContent:
Guide dogs and other service animals are allowed in the store. Other pets are not allowed.
Topic name: parking_locationContent:
Parking is available in the structure beneath the mall entrance.
Topic name: parking_costContent:
Self-parking is free on weekdays and $5 an hour on weekends between 9am and 5pm.

Check your understanding

Verification

Test in Chat

Ask the question using:
  • Exact phrasing
  • Informal phrasing
  • Polite phrasing
  • Abrupt phrasing
Confirm:
  • The same topic triggers every time
  • The response does not change
  • No follow-up question is asked
Live, in the test panel, look for the topic citations to confirm which topic was recalled by the agent. overview

Test in Call

Ask the same question out loud, including hesitation or filler words. Confirm:
  • Speech is transcribed correctly
  • The response sounds natural when spoken
  • The agent does not over-explain
In conversation review, make sure topic citations are enabled: overview

Final checklist

Before moving on, confirm:
  • The topic answers exactly one question
  • Sample questions reflect real user phrasing
  • Content is short and speakable
  • Actions are empty
  • The topic behaves the same in Chat and Call

Try it yourself

1

Challenge: Write a pet_policy topic

Write a complete simple topic for a store that allows only service animals.Include:
  1. Topic name (snake_case)
  2. Five sample questions
  3. Content (one to two sentences)
The topic name should describe exactly one intent. Sample questions should reflect how real users speak — including informal and spoken phrasing. Content should be speakable and complete in one or two sentences.
Topic name: pet_policySample questions:
  • are pets allowed
  • can I bring my dog
  • do you allow animals
  • is my cat allowed inside
  • what’s your policy on pets
Content:
Service animals are welcome in the store. Other pets are not allowed inside.

Check your understanding

← Previous: Edit agent behavior

Lesson 2 of 6

Next: Edit voice settings →

Lesson 4 of 6
Last modified on April 1, 2026