Introduction
The Conversations page in Rep AI provides a comprehensive view of all customer interactions with your AI assistant. This central hub allows you to monitor, review, and manage conversations between your customers and your AI, helping you ensure quality customer service and gain valuable insights into customer needs.
Features of the Conversations Page
Conversation List Panel (Left Column)
The left side of the screen displays a chronological list of customer conversations, including:
Location & Timestamp: Shows city, state, country, and the exact date and time of each conversation (e.g., "Fairfax Station, Virginia, United States - 6 Jun 2025 at 03:36 am")
Customer Information: Displays customer email and device type (Mobile/Desktop)
Subscription Status: Indicates whether the customer is subscribed to a mailing list
Conversation Duration: Shows the length of the interaction in minutes and seconds
Order Information: When applicable, displays order ID and total amount
Save Option: Star icon allows you to save important conversations for quick access later
Share Option: Share icon enables you to export or share conversation details with team members
You can click on any conversation to view its full details in the right panel.
Filtering and Sorting Options
At the top of the page, you'll find tools to organize your conversations:
Date Range Filter: Select a specific time period (e.g., "1 Jun 2025 - 30 Jun 2025")
Sort Options: Arrange conversations chronologically or by other criteria
View Toggles: Switch between All, Saved, and Unread tabs to focus on the conversations you want to review
Search Bar: Find specific conversations by keywords or content
Conversation Counts on Each Tab
Each of the three tabs at the top of the Conversations page — All, Unread, and Saved — shows a small count bubble next to its name. At a glance, you can see how many conversations sit in each tab without having to open it and scroll.
How it works:
The number on each tab reflects how many conversations are currently in that tab.
Counts respect your active view. When you set a date range or apply filters (platform, topic, shopper emotion, and so on), each tab's count updates to match what you'd actually see inside that tab.
The counts stay current as you work. Open an unread conversation and the Unread count goes down; save or unsave a conversation and the Saved count adjusts right away.
Why it's useful:
See your unread backlog instantly. The Unread count works like the unread badge on an email folder — it tells you how many conversations are still waiting for review before you click in.
Gauge volume before you dig in. Knowing whether a tab holds a handful or hundreds of conversations helps you plan your review session.
Confirm your filters are working. Because the counts follow your filters and date range, they're a quick sanity check that you're looking at the right slice of conversations.
Advanced Filtering Modal
The "Filters" button opens a comprehensive filtering panel that allows you to narrow conversations by:
Platform: All, Web (Desktop + Mobile), Desktop, Mobile or Helpdesk
Shopper Emotion: Search for specific customer sentiments
Customer Problem: Filter by customer issue types
Customers: All, New, or Returning customers
Traffic Source: All, Direct, Referral, Search, or Social
Tag: Add multiple custom tags for organization
Topic: Filter by conversation topics
Messaging Channels: Instagram DM, Facebook DM, or WhatsApp DM
Email: Find conversations by shopper email address — type a full address or just part of one
Email Filter
The Email filter lets you pull up conversations tied to a specific shopper by their email address — ideal when a customer reaches out and you want to see their full history with your AI.
How to use it:
Navigate to the Conversations page.
Click the Filters button to open the filtering panel.
In the Email field, type the shopper's email address. You can enter a full address (for example, jane@example.com) or just part of one (for example, "jane" or "example.com").
The conversation list narrows to shoppers whose email matches what you typed.
Full or partial match: You don't need the exact address. Typing part of an email returns every conversation whose shopper email contains that text — handy when you only remember a fragment, or want to group shoppers from the same email domain.
Works alongside your other filters: Like every filter on this page, Email combines with your date range, platform, topic, and the All / Unread / Saved tabs — so you can, for example, see one shopper's unread conversations from the last month in a couple of clicks.
Email filter vs. the Search Bar: The Search Bar looks across conversation content and keywords. The Email filter is narrower on purpose — it matches only the shopper's email address, so you get that customer's conversations without unrelated keyword hits.
AI-Generated Order Filter
The AI-Generated Order filter helps you quickly find conversations where your AI successfully completed an order-related customer inquiry from start to finish.
What qualifies as an "AI-Generated Order" conversation:
This special filter shows conversations that meet both of these criteria:
Order Completed: The conversation involved a complete order-related interaction (order status check, tracking update, order modification, etc.)
Resolved by AI: Your AI assistant handled the entire inquiry without requiring human intervention or handoff
How to use this filter:
Navigate to the Conversations page
Click the "Filters" button to open the Advanced Filtering Modal
Look for the AI-Generated Order tag option in the filter list
Combine with other filters (date range, platform, customer type) for more specific analysis
Note: This filter excludes conversations where customers requested human support or where the AI created a support ticket for your team to handle. It shows only fully AI-resolved order interactions.
Using the Unread Tab
The Unread tab helps you systematically review conversations you haven't viewed yet, making quality assurance much easier — especially for high-volume stores.
You'll find it at the top of the Conversations page, alongside All and Saved. The Unread tab also displays a count bubble showing how many conversations are still waiting for review, so you can see the size of your backlog before you open the tab.
How it works:
Click the Unread tab at the top of the Conversations page.
The list filters to show only conversations you (or your team) haven't reviewed yet.
All your other filters — date range, platform, topic, shopper emotion, and more — continue to work alongside the Unread filter.
Once you open a conversation to view it, it's automatically marked as read and removed from the Unread view.
Switch between All, Saved, and Unread at any time without losing your filters.
When to use the Unread tab:
Daily quality reviews — start each day by checking unread conversations to confirm your AI is performing the way you expect.
Batch reviews — work through accumulated conversations systematically without losing your place between sessions.
Team workflows — read status is shared at the account level, so you and your teammates won't duplicate review effort. Once anyone on your team opens a conversation, it's marked read for everyone.
Focused QA — combine Unread with filters like Topic or Shopper Emotion to zero in on the specific conversation types you want to audit.
Empty state: When you see "You're all caught up! All conversations have been reviewed," it means every conversation matching your current filters has been viewed.
Conversation Viewer (Right Panel)
The right side of the screen shows the full conversation transcript:
Customer Messages: Displayed with customer icon and timestamp
AI Responses: Shown in branded bubbles with a ChatGPT indicator
Product Images: When relevant, product images appear in the conversation
Page Visit Information: Shows which product pages the customer viewed during the conversation
Added Products: Lists items the customer added to their cart
Suggestion Chips: Interactive buttons offered to customers during conversation (e.g., "I have a question", "What's recommended?")
Ticket Resolved by AI: For conversations handled through Omnichannel AI, a "Ticket resolved by AI" marker appears in the transcript at the exact moment your AI resolved the conversation.
Conversation Rating Requests: When a shopper taps the rating icon at the top of the chat widget, a "Customer clicked 'Rate this conversation'" marker appears in the transcript at the moment they tapped it.
Ticket Resolved by AI Indicator
When your AI resolves a conversation through Omnichannel AI, the transcript shows a "Ticket resolved by AI" marker placed at the exact point in the timeline where the resolution happened — right alongside the messages and other activity in that chat.
Omnichannel AI is what lets Rep AI answer customers on channels beyond your website — email, Instagram, Facebook, and WhatsApp — where you're charged only when the AI resolves the conversation. This marker is how you see each of those resolutions on the conversation itself. To learn how to enable it, see the article Let Rep AI Answer Customers on Every Channel with Omnichannel AI.
What it shows: Confirmation that the AI resolved this specific conversation, and the time it did so.
Where it appears: In the conversation transcript on the right panel, positioned chronologically among the messages — at the moment the resolution occurred, not at the top or bottom.
What the timestamp means: It's the actual time the AI resolved the conversation, shown in the same time format as the surrounding entries — not the time you opened the page.
When you'll see it: Only on conversations resolved by Omnichannel AI. If you don't have Omnichannel AI enabled, you generally won't see this marker. Conversations handed off to a human, or that the AI didn't resolve, won't show it either.
The marker is informational — there's nothing to click and nothing to set up. It appears automatically wherever it applies, so you can confirm a resolution conversation-by-conversation and reconcile individual chats against the AI-resolution totals shown on your dashboards and reports.
Seeing When a Shopper Asked to Rate the Conversation
Shoppers can rate their chat at any time using the rating icon in the top row of your chat widget. Tapping it opens the "Was this conversation helpful?" prompt.
Previously, that prompt appeared in the transcript with nothing to explain it — leaving you to guess whether the shopper triggered it or your AI did. Now the transcript shows the trigger.
What you'll see: A marker reading Customer clicked "Rate this conversation" with a star icon and the time of the tap, set apart from the messages as a neutral system entry — the same treatment as other activity markers in the transcript.
Where it appears: In the conversation transcript on the right panel, in chronological order — at the moment of the tap, directly before the "Was this conversation helpful?" prompt that followed. You can now read the full chain in sequence: the shopper tapped the rating icon, the prompt appeared, and the shopper answered.
What the timestamp means: The actual moment the shopper tapped the icon, shown in your local time — not when the prompt appeared and not when you opened the page.
When you'll see it:
Only on conversations where the shopper actually tapped the rating icon. If the satisfaction prompt appeared for another reason — for example because it was configured to follow an FAQ answer — the marker does not appear. That difference is exactly what the marker is for.
Once per tap. If a shopper tapped the icon more than once, you'll see a marker for each tap.
On past conversations as well as new ones — the marker appears across your conversation history, not only on chats from today onward.
On shared conversation preview links, so a teammate you've shared a conversation with sees the same marker.
The marker is informational — there's nothing to click and nothing to turn on. It appears automatically wherever it applies.
Why it's useful: An unexpected "Was this conversation helpful?" prompt is one of the most common things merchants ask us about. In most cases the shopper simply tapped the rating icon — sometimes by accident. This marker lets you confirm that yourself, on the conversation, in seconds.
Conversation Management Tools
There are several action buttons that help you optimize your AI's performance and gain insights into how it interacts with customers:
Instruct/Correct the AI:
-
Instruct the AI:
Purpose: Appears only for proactive approach messages, allowing you to adjust how the AI initiates conversations with customers.
When to use: When you want to modify how your AI reaches out to customers or suggests products/services before the customer has asked a specific question.
Benefits: Helps refine your AI's conversation starters and proactive customer engagement approaches.
-
Correct the AI:
Purpose: Available for regular messages that occur during customer exchanges, letting you improve responses after customer interaction has occurred.
When to use: After reviewing a conversation where the AI has responded to a customer query in a way that could be improved.
-
Benefits:
Directly improves your AI's responses for similar future questions
Helps train the AI to better understand your brand voice and product information
Creates a continuous feedback loop that enhances AI performance over time
Best practice: For both options, provide specific, clear guidance that explains not just what should be changed but how the AI should approach similar situations in the future.
Report a Problem:
Purpose: Flag a conversation that needs a closer look from Rep's team — for example a harmful or off-tone AI response, a technical glitch in the conversation flow, a misunderstanding of customer intent, or inaccurate product information.
-
Where you'll find it: Report a problem now follows the AI's reasoning instead of competing with it.
When an explanation is available for the AI response you're reviewing, open the Explanation panel first (see below). The Report a problem option appears at the bottom of that panel, after the reasoning and sources — so you decide whether to report only once you've seen why the AI answered the way it did.
When no explanation is available for that response, Report a problem stays in the action row exactly as before, so you never lose the ability to flag an issue.
How it works: Clicking Report a problem opens a short form where you describe what went wrong during the conversation. Submitting sends the report to Rep's team for review. What you submit — and who reviews it — is unchanged; only the button's placement has moved.
Why this matters: Reviewing the AI's reasoning first means many concerns resolve themselves, so reporting becomes an optional second step you take only when something still looks wrong. This reduces premature reports and helps Rep's team focus on the conversations that genuinely need attention.
Explanation:
Purpose: Reveals the actual reasoning behind an AI response — captured while your AI works, not guessed after the fact — so you can see how and why it produced a specific answer.
-
How it works: Clicking Explanation (also shown as "Show reasoning") expands a panel directly beneath the response. Instead of an after-the-fact estimate, the panel shows the real basis for the answer, organized into clear sections:
Summary — a short, plain-language explanation of why the AI answered the way it did.
Knowledge used — the specific sources the AI drew on (such as a FAQ, a policy page, or a product detail), with links you can open to verify them. If the AI didn't find a relevant source and gave a general response, the panel says so.
Written for people, not engineers: you won't see system prompts, model names, or raw technical details — the summary and sources are always shown in plain language.
When an explanation isn't available: For older conversations where the original reasoning is no longer stored, the panel shows a reconstructed explanation and labels it as such, so you know it's an approximation. Some messages handled by an older response method can't be explained, and the panel will tell you so.
Reporting from the panel: Once you've reviewed the reasoning and sources, you'll find a Report a problem option at the bottom of the panel — making "understand first, report only if needed" the natural order of the review.
Feedback option: The panel includes a "Was this helpful?" option to further improve the explanation feature.
-
Benefits:
Transparency into AI decision-making, grounded in what the AI actually used
Helps identify gaps in your AI's knowledge or reasoning
Provides insight into why certain responses might need improvement
Sources:
Purpose: This tool reveals which knowledge base content, product information, or catalog data was used to generate the AI's response.
-
How it works: When you click "Sources," you'll see a list of all the information sources that informed the AI's answer, including:
Links to specific product pages
Knowledge base articles or FAQs that were referenced
Custom instructions that guided the response
Website content that was used
-
Benefits:
Helps you verify that the AI is using the most relevant and accurate information
Identifies which knowledge sources are most frequently used
Shows when the AI might be missing important information
Optimization opportunity: Review frequently used sources to ensure they contain comprehensive and up-to-date information.
Common Use Cases
Quality Monitoring: Review conversations to ensure your AI is providing accurate and helpful information
Customer Insight Collection: Identify common questions, concerns, or product interests
AI Performance Improvement: Use the correction tools to help train your AI to provide better responses
Product Page Optimization: See which products customers are viewing and discussing most frequently
Custom FAQ Creation: Turn common customer questions into custom FAQs for future reference
Best Practices
Regularly review conversations to identify opportunities for improving your AI's knowledge base
Use the filtering options to focus on specific customer segments or time periods
Create custom FAQs for frequently asked questions to improve response accuracy
Review the explanation feature to understand how your AI is responding to customers
Use the "Correct the AI" feature to provide guidance when responses need improvement
Now that you understand the Conversations page, you can effectively monitor and improve customer interactions with your AI assistant, ensuring optimal customer service and gathering valuable insights for your business.