Suggestions

Have the AI read through the answers your agent got wrong, group them into patterns, and tell you what knowledge is missing — and where a live-data API tool would help instead.

Overview

The Suggestions tab analyzes the conversations your AI agent failed to answer, finds the patterns in them, and turns those into a short list of things worth adding. Rather than reading failed chats one by one, you get the recurring problems ranked by how often they came up.

Generating suggestions regularly is the cheapest way to keep the agent effective: it tells you which gaps are actually costing you answers, instead of leaving you to guess.

How It Works

The system reads the recent turns where the agent could not answer, or escalated to a person because it had nothing to say, and uses them to:

  • Identify patterns — find recurring issues and unanswered questions
  • Prioritize improvements — put the most frequent problems first
  • Recommend knowledge — specific questions and topics to add to the brand
  • Recommend API tools — where the question needs live data that no stored article could ever hold

Each run looks at the failed answers since your previous analysis, so you get fresh findings rather than the same list again. Requests where the visitor simply asked for a human are left out — that is routing, not a knowledge gap.

Generating Suggestions

Generating a report is paid like other AI work. It is billed first to your free 14 days' credits, then to credit you bought, then to your paid plan's usage. If none of them can pay, nothing runs and the payment window opens. Reading reports you already generated stays free.

Open the Suggestions tab

Go to Brand → (your brand) → AI agent → Suggestions.

Click Generate

Click Generate suggestions to start the analysis (once a report exists, the button reads Generate new report). If your plan lets you choose the AI model, a model selector sits beside the button.

Wait for Analysis

The system reads the recent failed answers, works out the patterns, and streams the result onto the page as it is written.

Review Results

The finished report opens with a summary of the main patterns found, then lists the prioritized questions and topics to add under Context Ideas, and — when the failures clearly need live data — a few API tool suggestions under Context Tool Ideas.

Analysis needs at least 10 new failed answers since your last report. With fewer than that, wait until more conversations have come in.

Acting on a report

A report tells you what is missing; you add it in one of three places.

  • Knowledge gaps — add the answer at Brand → (your brand) → Knowledge. A Q&A pair is usually the quickest fix for a single recurring question; a longer topic belongs in a help article or an uploaded file.
  • Product gaps — if the questions are about products you sell, the fix is a product feed at Brand → (your brand) → Products, not a written article.
  • Live-data gaps — anything about this customer's order, booking or account can never be answered from stored text. Those are the ones the report lists as tool ideas, and they belong in an API tool, which you add with Add knowledge on Brand → (your brand) → Knowledge.

Viewing Previous Suggestions

All previously generated reports are listed on the page in date order, newest first, each as a card titled Suggestions report with its date, showing:

  • When it was generated
  • The analysis summary — an overview of the patterns identified
  • Context Ideas — specific questions or topics to add
  • Tool ideas — API suggestions for questions needing live data

Reports belong to the AI agent. Deleting Yaplet removes them along with its conversation starters and its answer reports; your knowledge is untouched.

Best Practices

  • Generate regularly — run the analysis weekly, or after a significant change to your knowledge
  • Review patterns — start with the most frequent issues for the biggest improvement
  • Add what it asks for — put the suggested questions and answers into Brand → Knowledge and re-check next week
  • Take the tool ideas seriously — a question that needs live data will keep failing no matter how much you write
The AI prioritizes quality over quantity, focusing on the most impactful improvements that will resolve the largest number of failed conversations.

© 2026 Yaplet. All rights reserved.