Training is where you save the questions that matter to your business. Each trained question can recognize several kinds of customer wording while still using one consistent answer strategy.
#How Training works
Open Training in the IdeaLogic dashboard. The two everyday views are arranged around a simple workflow:
- Questions organizes trained questions and the Unmatched review queue.
- Test Bot lets you ask customer-style questions and inspect the answer and sources.
Each step below matches the dashboard screen you will see next, beginning with Add question and ending in Test Bot.
#1. Add a customer question
Choose the chatbot you want to train, stay on All questions, and select Add question.
- What services do you offer?
- How quickly can I get started?
- Can I book an appointment?
#2. Enter the customer question and click Search
Type the question exactly as a customer might ask it, then select Search.
Searching first prevents duplicate questions. When two questions are intended to produce the same behavior, keep one trained question and add the other wording as a matching phrase.
#3. Create the question
If none of the suggested questions mean the same thing, select Create new question. Review the primary phrasing, add a few natural variations, and then select Create question.
- How long does setup take? ×
- When can we launch? ×
#4. Add and refine matching phrases
After creation, IdeaLogic opens the full question editor. Its Matching panel uses three tools to decide what should—and should not—match:
- Primary phrasing
- The question title displayed in Training and the clearest example of the customer’s intent.
- Other matching phrases
- Natural variations that should be treated as the same question.
- Should not match
- Similar-looking questions that have a different meaning and need separate behavior.
Short phrases, full questions, and common terminology are more useful than repeating the same sentence with tiny punctuation changes.
#5. Control the answer
By default, IdeaLogic uses the chatbot’s assigned knowledge and instructions to answer. Turn on Specific answer when a question needs more control.
Choose the answer style that fits the situation:
- Use exact response sends the answer as written.
- Use wording closely keeps the same meaning and phrasing while allowing small adjustments.
- Use naturally as guidance treats your answer as the source of truth but responds conversationally.
Most new chatbots can be configured and prepared for review in a few business days. The exact timing depends on the amount of content and the integrations you need.
Leave this setting off when the normal answer already works. Too many rigid answers can make conversations feel less natural and harder to maintain.
#6. Control knowledge sources
Pin one or more sources when they contain the best information for a question. Pinned sources are prioritized during retrieval.
Turn on Restrict to pinned sources only when the answer must come exclusively from those sources. A restricted question with no pinned sources will return no sourced results.
- Implementation timelineWebsite page
- Customer onboarding guidePDF
#7. Save and test your chatbot
Select Save changes, open Test Bot, and try several customer-style variations:
- Ask the primary phrasing once.
- Try at least two matching phrases with different wording.
- Try a nearby question that should not match.
- Inspect the answer and its sources, then adjust the question if needed.
#Review unmatched questions
A question appears in Unmatched whenever it did not match a trained question—even if the AI still produced an answer. Questions from live chats and Test Bot are included and grouped by similar wording.
Open an unmatched group and choose the most appropriate action:
- Attach it to an existing question as additional matching wording.
- Create a new trained question when it represents a distinct customer need.
- Leave it unmatched if it is irrelevant or too rare to justify new behavior.
#Training best practices
Prioritize pricing, availability, services, policies, booking, and other questions that affect customer decisions.
Separate topics that need different answers, sources, or downstream behavior.
Review live conversations instead of guessing every possible phrase in advance.
Retest important questions when your pricing, policies, services, or source content changes.