Overview
Conversations is the central hub for browsing, filtering, reviewing, and scoring all conversation records in Ender Turing. A conversation can be a phone call, chat, email, web interaction, or a manually created record. From this screen you can search across your entire conversation history, listen to call recordings, read transcripts, leave comments, score agents using scorecards, and download records for offline use.
How It Works
Open Conversations from the main navigation. The screen is divided into two panels:
Left panel — a scrollable list of conversations matching your current filters, along with a count of total results and their combined duration.
Right panel — the detail view for the selected conversation, showing its transcript, media player, metadata, and available actions.
Select any conversation in the list to load its details on the right. The selected conversation is highlighted, and the URL updates so you can bookmark or share the link.
Comment and Score Markers in the List
Each row in the conversation list can show two small markers on the right:
Comments (speech-bubble icon, tooltip Conversation has comments) — the conversation has at least one comment. When it has more than one, the number of comments appears next to the icon.
Scored (circle icon, tooltip Conversation is scored) — the conversation has at least one submitted score. When it has more than one, the number of scores appears next to the icon.
The score number counts scores, not people. Every submitted scorecard counts separately, so a conversation scored on two scorecards by Auto QA shows 2, and one reviewer scoring the same conversation twice also shows 2. Scorecard defaults that were never submitted as a review are not counted. A conversation with exactly one score shows the marker without a number, and a conversation with no scores shows no marker.
Filtering Conversations
Click the filter bar at the top to add filters. Each filter can be added as an include or exclude condition, letting you narrow results precisely.
Commonly used filters include:
Filter | What it does |
Search | Full-text search across transcripts; add it more than once to require several keywords or phrases — see Combining Several Searches. Advanced mode can limit a Search to selected summary prompts — see Limiting Search to Summary Prompts |
Has summary | Conversations that have any AI summary, or a summary from selected prompts — see Has Summary Filter |
Record Type | Call, Chat, Email, Web, or Loan |
Agent / Team | Specific agents or entire teams |
Date Range | A single day or custom date range |
Tags | One or more tags, optionally split by speaker |
Topics | Conversation topic or category |
Scorecards | Filter by scorecard used or score status |
Call Direction | Inbound or Outbound |
Call Duration | Preset ranges or a custom value |
Interval to prev call | Time since the same client's previous call — see Repetitive Calls below |
Script Adherence | Compliance rule match percentage |
CRM Status | Status values from an integrated CRM |
Reviewed | Whether the conversation has been scored |
Commented | Whether any comments exist |
Has attachments | Whether the conversation has any attached file |
Silence Period | Low, Medium, or High silence percentage |
Additional filters are available for metrics such as CSAT, NPS, hold time, response time, and more. You can also filter by campaign name, queue name, and termination reason.
Combining Several Searches
On the Conversations screen you can add Search more than once, with one keyword, phrase, or number in each. A conversation is shown only when it matches every Search you added, so you do not need to type AND yourself. For example, add one Search with a client's phone number and another with the word refund to find only that client's conversations that mention a refund.
Click the + button in the filter bar and choose Search.
Enter the first keyword, phrase, or number and add it.
Choose Search again. A new, empty Search opens instead of the one you already added — enter the next value and add it.
Click Apply to update the list (or Discard to drop the changes).
Each Search appears as its own chip. Click a chip to change that value, or click its × to remove it; the other Searches and filters stay as they are. Clear filters removes everything at once.
Things to know:
Each Search matches the same way a single Search always has. Adding more Searches only narrows the results further — there is no "any of these" option.
Search cannot be used as an exclude condition.
All Searches are kept in saved favorites, recent filters, and shared conversation links, and they combine with your other filters as usual.
Repeating Search is available on the Conversations screen. On other screens that use the same filter bar, choosing Search again opens the existing Search for editing.
In the conversation you open, text matching any of your Searches is highlighted in the transcript and in AI summaries. For a structured summary shown in its formatted view, a ✓ Match found marker appears under the summary instead of highlighting individual words.
Limiting Search to Summary Prompts
By default, Search looks everywhere, including transcripts, comments and AI summaries. Use advanced mode to find the text only in summaries made by particular summary prompts. For example, you can find conversations whose Call reason summary mentions refund.
Click the + button in the filter bar and choose Search.
Click the » button in the Search header (labelled Advanced search). A checklist of summary prompts appears under the text field.
Enter your query and tick one or more prompts.
Click Apply.
How the results are matched:
At least one summary must come from a selected prompt and contain the query. If you tick several prompts, a summary from any of them can match.
Text is not combined across summaries. If one word is in one summary and another word is in a second summary, the conversation does not match.
Matches only in the transcript, in comments, or in summaries from other prompts do not count.
Switching modes:
Opening advanced mode does not change your results. While no prompt is ticked, Search still looks everywhere. You can apply a query without ticking any prompt.
Untick all prompts to search everywhere again.
Click the « button (labelled Search everywhere) to return to simple mode. This clears the prompt selection and keeps your query.
The chip shows how many prompts the Search is limited to, for example Search: refund (Selected prompts: 2). Click the chip to reopen the Search in advanced mode with your query and prompts restored.
Things to know:
The checklist shows prompt names, not individual summaries. Ticked prompts are listed first.
It only lists summary prompts whose results your role is allowed to see. Archived prompts stay in the list with an (Archived) label. If a saved filter names a prompt that no longer exists, it appears as Unavailable prompt so you can see it and untick it.
When prompts are ticked, participant prefixes such as
client:,partner:orcustomer:are not allowed. The field shows an error asking you to clear the prompt selection. Use these prefixes in a simple Search instead.A Search limited to prompts cannot be used as an exclude condition.
It combines with your other filters and with other Searches. It is kept in favorites, recent filters and shared conversation links.
A newly generated summary may take a short time to become searchable.
Has Summary Filter
Use Has summary to find conversations that already have an AI summary. You can also require a summary from particular summary prompts. This filter only checks whether a summary exists, not what it says. To search summary text, use Limiting Search to Summary Prompts.
Click the filter bar and choose Has summary.
Leave every prompt unticked to match conversations with any summary. To require specific prompts, tick one or more. Use the search box above the list to find prompts by name.
Apply the filter.
The chip reads Has summary: All when no prompt is ticked. Otherwise it shows how many prompts are ticked, for example Has summary: Selected prompts: 2.
Things to know:
With several prompts ticked, a summary from any of them is enough. Conversations without a qualifying summary are excluded.
Has summary does not change what a separate Search looks at. If you add both, a simple Search still searches everywhere.
The prompt list follows the same rules as advanced Search. It only shows prompts whose results your role can see, archived prompts are labelled (Archived), and removed prompts appear as Unavailable prompt.
With no prompt ticked, summaries you are not allowed to see are ignored. So are summaries from apps that are no longer installed.
If a saved filter names a prompt that you cannot see, or one that belongs to an app that is no longer installed, it returns no conversations.
Has summary cannot be used as an exclude condition.
It is kept in favorites, recent filters and shared conversation links.
Where These Summary Filters Are Available
You can add Has summary and a Search limited to summary prompts in Conversations and in Charts (see Charts). Other screens that use the same filter bar do not offer them:
If such a screen already has one of these filters, for example from a link, it appears as a chip that you can remove but not edit.
Favorites and recent filters that contain them are greyed out on those screens.
Enders can still open a preview in Conversations that uses these filters — see Enders.
Has Attachments Filter
Use Has attachments (paperclip icon) to find conversations that came with attached files, such as documents sent in a chat or email.
Click the filter bar and choose Has attachments.
Select Yes to show only conversations with at least one attachment, or No to show only conversations without attachments.
Apply the filter. The chip reads Has attachments: Yes or Has attachments: No.
Things to know:
The filter checks only whether an attachment exists. It matches any file format, and it does not matter whether the file's text can be read by EnderGPT.
It combines with your other filters and is kept in saved favorites, recent filters, and shared conversation links.
It is also available when you set the scope of a new EnderGPT chat — see Chat with EnderGPT.
It cannot be used in rules that check each conversation while it is being processed, such as Enders filters or topic filter rules. Attachments are added to a conversation after those rules run, so saving such a rule with Has attachments is rejected.
Duration and Time Filters
Filters that accept a time interval let you set a custom value with a number and a unit picker instead of raw seconds. The unit picker offers seconds, minutes, hours, and days. This applies to the following Filters panel entries:
Interval to prev call (Repetitive)
Call Duration
Waiting time
Response time
Longest monologue
Holds — for Overall holds time and Max hold time. The Number of holds field is a plain count.
For each of these filters the custom input has three parts:
An operator button — At least (
>=) or At most (<=). Holds duration fields also offer Exactly (=).A number field for the value.
A unit selector — seconds, minutes, hours, or days.
Where the filter offers preset ranges (for example the Interval to prev call preset 5m–48h, or the Call Duration presets), the presets appear above the custom input and you can pick one instead of entering a value.
Notes:
Fractional numbers are allowed when they resolve to whole seconds — for example 1.5 minutes is accepted and applied as 90 seconds, while 0.5 seconds is rejected with an inline error.
Max hold time requires at least 10 seconds.
Existing saved filters, favorites, and charts continue to work. When you reopen a saved value, it is displayed in the most readable unit — for example a stored value of 300 seconds appears as 5 minutes.
If a saved filter uses a closed range (for example
60–300) that cannot be expressed as a single At least, At most, or Exactly threshold, the value is preserved and still filters results, but the edit dialog shows an inline hint and asks you to replace it with a preset or a supported custom threshold.
Saving and Reusing Filters
You can save filter combinations as favorites and access your filter history directly from the filter bar.
Saving a filter set: Apply your filters on the Conversations page and click the star icon labeled Save to favorites in the filter bar. Enter a name and save.
Quick access panel: Click Favorites & recent in the filter dropdown to open an inline panel with two tabs:
Favorites — lists your saved filter sets. Click any favorite to apply it immediately. If you have saved favorites, this tab opens by default.
Recent — automatically tracked filter history. Switch between Last used (sorted by date) and Often used (sorted by frequency). If a recent filter combination matches a saved favorite, the favorite name is shown.
Click Manage favorites at the bottom of the panel to open the dedicated management page, where you can:
Rename favorites inline
Reorder favorites by drag-and-drop
Delete favorites or history entries
View full filter details for each saved set
Viewing a Conversation
When you select a conversation, the detail panel shows:
Header — conversation type, date and time, duration, direction, agent name, and the client on the other side — see How the Client Is Named.
Transcript — the full conversation text, with speaker labels and timestamps. For calls, you can click a transcript line to jump to that point in the audio.
Media Player — audio or video playback controls (for calls and screen recordings). Supports stereo playback.
AI Summaries — if automated summarization is configured, summaries appear for the conversation. If a summary could not be generated, a notice appears in its place — see When an AI Result Fails below.
Attachments — files sent in a chat or email appear under the message they came with. See Attachments below.
Compliance Matches — if script adherence rules are configured, you can view which rules matched.
Repetitive Calls — for calls from the same client (or to the same destination), the toolbar shows a phone-log icon listing related conversations, and a Prev / Next client request strip appears under the header. See Repetitive Calls below.
How the Client Is Named
The conversation header and each row of the conversation list show the other side of the conversation — the caller for inbound conversations and the destination for outbound ones. Ender Turing picks the most readable label it has, in this order:
Agent (header only) — if the number belongs to a known agent, the agent's name is shown. Users who can view agents can click it to open that agent's page.
Partner — if the number matches one or more entries in your Partners list, the partner names are shown. In the list they appear in a highlighted color.
Customer name from the chat — for inbound or outbound chats that arrived with a customer name, such as Genesys chats, that name is shown.
Original identifier — otherwise, the phone number, address, or messaging ID received from your platform.
The original identifier is always kept, even when a name is shown:
In the header, hover over (or focus) the name, or tap the small arrow next to it on a touch screen, to see the original identifier with a Copy button. Copy copies the identifier, not the name.
Clicking the name in the header (when it is not an agent link) clears your current filters and searches the conversation list for that identifier.
In the list, hovering over the name shows the identifier as a tooltip.
Chat customer names are display labels only. Search, filters, repeat-call matching, and the Client column in exports keep using the original identifier. Internal conversations and voice calls are not affected. Chats imported before customer names were available keep showing their identifier.
When a Recording Cannot Be Played
If something prevents playback, a message appears at the bottom of the screen instead of the player silently doing nothing:
Message | What it means |
This recording could not be played in your browser. | The recording loaded, but your browser refused to play it. The player returns to Play and any accompanying screen recording is paused, so the controls match what you hear. |
Audio file is not available | The audio recording could not be loaded for this conversation. |
Video file is not available | The screen recording could not be loaded. The player switches back to audio only. |
When you see This recording could not be played in your browser., you can click Play again to retry. If it keeps failing, open the conversation in a different browser or on a different device. The transcript, comments, and scoring stay available either way. If Download is available to you, Audiofile saves the recording file so you can try to play it locally.
Attachments
When a message in the transcript came with files, each file is listed by name, with a file icon, directly below that message. What happens when you click a file depends on its type:
PDF files open a small menu with two actions:
See the content — opens a window showing the text extracted from the PDF, so you can read it without leaving the conversation. The window title is the file name; scroll to read long documents, and click Close (or press Escape) to return to the transcript.
Download the attachment — downloads the original PDF file.
Other files download straight away when clicked. They do not have a See the content option.
A file counts as a PDF when it is marked as a PDF or its name ends in .pdf.
See the content shows the text exactly as it was extracted, as plain text. Formatting, images, and layout from the original PDF are not reproduced, so download the file when you need to see the document as it was sent.
If no text is available, the window explains why instead of showing an empty page:
Message | What it means |
No parsed text is available for this attachment. You can download the original file. | No extracted text exists for this file, for example because extraction has not happened for it. |
Text extraction is not supported for this file type. | The file's format cannot be read as text. |
The attachment was too large to extract its text. | The file exceeded the size that can be processed. |
The attachment could not be parsed because its format is invalid. | The file is damaged or not a valid PDF. |
The PDF contains no extractable text. | The PDF holds no text layer — typically a scanned document made of images. |
The original attachment file was unavailable when text extraction was attempted. | The stored file could not be found when its text was being extracted. |
Could not load the attachment content. Please try again. | The text could not be loaded this time. Try again, or download the file instead. |
In all of these cases you can still use Download the attachment to get the original file, unless the stored file itself is missing.
The same extracted text is what EnderGPT reads when it answers questions about PDF attachments — see Chat with EnderGPT. To find conversations that have files attached, use the Has Attachments Filter.
When an AI Result Fails
Summaries, Auto QA scores, and automatic topics are generated in the background by Enders. When the most recent attempt at one of them fails, the conversation shows a short notice in red instead of leaving the result silently missing.
Where the notice appears depends on what failed:
Summary — inside the AI summaries panel, in place of the missing summary, labelled with the prompt name.
Auto QA — under an AI processing heading in the right-hand column, labelled with the scorecard name. In Review mode the same notice moves above the scorecard.
Automatic categorisation — in the same AI processing block, labelled with the prompt name; above the scorecard in Review mode.
You do not need to switch to Review mode to see them — the notices are in the normal conversation view as well.
For Auto QA and categorisation, a score or topics may still be displayed next to the notice. All the notice tells you is that the latest automatic run did not add or replace anything — it says nothing about where the values on screen came from. They may be the output of an earlier successful run, in which case they can be out of date relative to the transcript; but a score can equally be one a reviewer gave by hand (the scoring panel lists Auto QA under the system user and each human reviewer separately), and topics can have been set manually when the conversation was edited. Check the reviewer on a score before treating it as stale.
Every notice has two lines. The first says in plain words what went wrong, for example Input blocked: PROHIBITED_CONTENT or Incomplete AutoQA answer: received 8 of 10 criteria. The second carries a cause code and the time the failure happened, for example: "This summary couldn't be generated. Contact support and mention code "input_blocked" (12 Mar 2026, 14:07)." Quote that code and time when you contact support. The underlying technical message, the prompt, and the provider's own text are deliberately never shown in the interface, so the code is what identifies the cause.
Code | Message you see | What it means |
| Input blocked: … | The AI provider refused to process the prompt it was sent. Besides the conversation, the prompt includes the prompt template, conversation details and, for Auto QA, the scorecard criteria, so any of these can trigger the block. The provider's reason follows the colon, for example |
| Output blocked: … | The provider started answering but stopped its own answer, for example for |
| The provider blocked the response; the exact reason is unknown | The provider reported a block without saying why. |
| The response reached the output token limit | The answer was cut off at the provider's length limit and could not be used. |
| Empty response: reason unknown | The provider returned nothing, and gave no reason. An empty answer on its own is not treated as a safety block. |
| The AI returned a malformed response | The answer was not in the structure the prompt asks for. |
| Provider response failed: … | The provider ended the answer with some other reason, shown after the colon. |
| Incomplete AutoQA answer: received X of Y criteria | Auto QA only. The AI answered fewer scorecard criteria than the scorecard has. |
| AutoQA returned X criteria instead of Y | Auto QA only. The AI answered more criteria than the scorecard has. |
| The AutoQA answer does not match the requested criteria | Auto QA only. The answer included a criterion that is not on the scorecard. |
| The AI returned an invalid response | The provider answered, but the answer could not be used, and no more specific cause is known. |
| The AI provider rejected authentication | Ender Turing could not authenticate with the AI provider. This is usually a key or account problem in the provider configuration. |
| The AI provider was unavailable | The AI provider could not be reached or declined to serve the request. |
| The AI provider rejected the request | The provider rejected the request itself, for example on size limits. |
| An internal error prevented processing | Ender Turing itself failed. For example, a database problem on our side, or an error that does not correspond to any provider response. |
| The cause is unknown | The failure was recorded before cause codes existed, or its cause was not recorded. |
The block reasons, such as PROHIBITED_CONTENT, come from the AI provider. They appear only when the provider reports one. If the reported reason is not one Ender Turing recognizes, it is shown as UNKNOWN. Failures recorded before these detailed causes were introduced keep their original, more general code; they are not reassigned retrospectively.
For some causes, Ender Turing retries automatically before a notice appears:
Auto QA retries an unusable answer (including an incomplete or mismatched set of criteria) and an unreachable provider, up to three more times. When Re-Score if lower then threshold is turned on (see Metrics and Variables), when the regular model's answer has missing, extra, or invalid criteria, the same scoring request is first sent again to the AutoQA re-scoring model within the same attempt. This does not happen for blocked, empty, or malformed answers. On a retry, groups of criteria that were already answered correctly are normally reused rather than scored again. For these causes, a notice appears only when the attempts are used up. Other failures, such as
provider_auth,request_rejected, orinternal_error, are not retried, and the notice appears straight away.Provider blocks count as failures even if the provider returned some text alongside the block, so a partly blocked answer is never saved as a result. Because the same conversation content is sent on a retry, a blocked conversation can be blocked again on a rerun.
A notice is not something you dismiss. It clears on its own when:
A later run succeeds. Re-running the responsible Ender over the affected conversations is the usual retry — Enders → open the Ender → Run Historical (see Enders). For Auto QA, a score for that scorecard clears the notice whichever prompt produced it. Notices are re-checked when you open a conversation, not while it stays open, so reselect or reopen the conversation after the run finishes — otherwise the old notice is still on screen and looks like the retry failed.
Any Ender can do the retry. A failure belongs to the conversation, not to the Ender that ran it, so deleting that Ender does not clear the notice — it deliberately stays, because the result still was never generated. What clears it is a successful result for the same prompt (for Auto QA, the same scorecard), from any Ender. If the original Ender is gone, run another one that uses that prompt or scorecard, or recreate it and run it on history.
The result arrives another way. A summary sent in through the API for that prompt also counts, and clears the notice.
The step can no longer run at all. For a summary or categorisation notice, that means its prompt was archived, deleted, or uninstalled with its app. For an Auto QA notice it is the scorecard that matters — archiving, deleting, or uninstalling the prompt that happened to fail is not enough, because another prompt can still score the same scorecard; the notice is dropped only once the scorecard itself is deleted or archived.
Two points about visibility:
Summary notices follow the summary they stand in for: if a prompt's results are restricted to other roles, neither the summary nor its failure shows up for you. Auto QA and categorisation notices are broader — anyone who can open the conversation sees them, including users who have no access to the scoring interface, so the scorecard or prompt name in the notice is visible to them too.
Recent attempts — successful or failed, with their timing — are in the activity log at the bottom of the Session details dialog. That log requires the Enders / Automation Management permission (under Analytics Configuration in Roles and Permissions), so most reviewers will only see the notices themselves. A failed Auto QA, summary, or categorisation entry in the log shows the same plain-language message as the notice, followed by its code in brackets. The log can also show Categorisation attempt limit reached for this conversation, which means automatic categorisation stopped trying for that conversation after repeated attempts. It is a recent-activity view, not an audit record: it shows the newest entries only (up to 200 per conversation), and routine successful entries are cleared after a retention period (30 days by default). Recorded failures are kept.
Viewed and Unviewed Conversations
The conversation list keeps track of which conversations you have already opened, so you can tell at a glance what is left to go through in a long result set.
A conversation is marked as viewed as soon as you open it in the detail panel. There is no separate "mark as read" button.
Viewed conversations appear with a shaded background and lighter text; conversations you have not opened yet keep the standard styling.
The list is personal to your account — what you have opened does not change how the list looks for your colleagues.
A few things to expect from this indicator:
It is stored in your browser, not on the server. Opening Ender Turing in a different browser or on a different device starts with an empty list, and clearing your browser data resets it.
Each entry records when you first opened that conversation; reopening it later does not refresh that moment. Entries are kept for 180 days from that first view, and only the 5,000 most recently first-opened conversations are retained. Anything older, or beyond that count, appears as unviewed again the next time you see it in the list.
Actions on a Conversation
The detail view toolbar offers several actions depending on your permissions:
Share — copy the conversation link to your clipboard or send it by email.
Copy Transcripts — copy the transcript text to your clipboard.
Download — export the conversation in multiple formats:
Text formats: Word (.docx), Excel (.xls), CSV (.csv), or Plain Text (.txt)
Audio (.opus) — when a recording is available
Video (.mp4) — when screen recording is available
Reactions — add an emoji reaction to the conversation.
Pin to Playlist — add the conversation to a training playlist (requires playlist management permission).
Flag Low Quality — mark the transcription as low quality for review.
View / Edit Session Details — open the Session details dialog. From there, click Edit to switch to the Edit Session form. When the conversation has an average score above zero, the dialog's Score row shows it; if there is more than one score, it adds how many, for example 0.85 from 3 scores. As in the list, each submitted score counts, including several from the same reviewer or from Auto QA.
Editing Session Details
The Edit Session form (opened from Session details → Edit) lets you correct the metadata of an existing conversation. Editable fields are:
Start — the conversation's start date and time. Pick a date from the calendar and a time from the time picker. Click Save to persist your changes.
Agent — the agent associated with the conversation. Start typing to search the agent list.
Agent channel — the channel number representing the agent's side of the recording.
Direction — Inbound or Outbound.
Topics — one or more conversation topics (categories). This field is only available to users who can manage analytics configuration.
After saving, the dialog closes and the conversation reloads with the updated metadata.
Repetitive Calls
When Ender Turing detects that a call belongs to the same client (same caller for inbound calls, same destination for outbound), it links it to the client's earlier call so you can move between repeats and filter the analytics on them.
You will see this in three places on a conversation:
Prev / Next client request strip — appears directly under the conversation header. It shows how much time passed between the current call and the client's previous or next call, with a clickable arrow on each side that opens that linked call. The strip is computed live from the client's phone number and a ±10-day window, so it appears as soon as the matching calls exist.
Repetitive Conversations dropdown — a phone-log icon in the conversation toolbar (visible when more than one related call exists). Opening it lists every related call for that client with date, direction, and agent. Clicking any entry switches the detail panel to that call.
Interval to prev call filter — in the Conversations filter bar. Use it to find calls whose previous call from the same client was within a chosen interval (for example, "5 minutes – 48 hours"), or enter a custom value with At least (
>=) or At most (<=) using the number + unit selector (seconds, minutes, hours, days). Enable Root call only (FCR failed) to keep only the first call of each repeat chain (the call that did not resolve the client's request on the first try).
The Interval to prev call filter is updated by a nightly job, so a call you just ingested may take up to one day to appear in the filter results. The Prev / Next client request strip and the dropdown are not affected by this — they show the current chain immediately.
Leaving Comments
Comments let you leave notes, feedback, or coaching points on any conversation. The comments panel appears in the conversation detail view; type your message in the editor at the bottom and click the send icon to post.
The comment editor supports rich text and a few extras:
Formatting — bold, italic, strikethrough, underline, bulleted list, and ordered list.
Link — click the link icon to attach a URL to the selected text.
Mentions — tag a teammate so they are notified about the comment. Type
@(or click the Add mention button) to open a list of users, then start typing a name or email to filter. Use the arrow keys and Enter to select, or click a name. The agent assigned to the current conversation is included in the suggestion list even if they would not otherwise appear. Each mentioned user receives a notification when you save the comment (and again if you edit the comment to add new mentions).Emoji — click the emoji icon to open a picker and insert an emoji.
Images — attach one or more images with the image icon, or paste an image from your clipboard directly into the editor. Attachments appear beneath the comment after saving.
Attached images appear as small thumbnails, with a loading spinner while each one loads. Click a thumbnail to open a larger preview; from there, use the open-in-new-tab button to view the image in a separate browser tab, or the close button to return. The new tab shows only the image itself, without the rest of Ender Turing. The open-in-new-tab button stays disabled until the image has finished loading.
Previews are shown for JPEG, PNG, GIF, WebP, and SVG images. Any other file attached through the image icon, such as an HTML page, is never previewed or opened in the browser. Its thumbnail is a crossed-out image icon, and you can only download it.
Each attachment has a Download button under its thumbnail. It saves the original file, including files that cannot be previewed. Hover over the button to see the file name. The saved file keeps that name, but the standard extension for its type may be added — for example, photo.jpg downloads as photo.jpg.jpeg.
If an attachment cannot be loaded, its thumbnail is replaced by the crossed-out image icon. When the file could not be fetched, an error notification also appears. That attachment cannot be previewed or opened in a new tab, but you can still try Download. While you edit the comment you can remove it with the × button on its thumbnail.
Keyboard shortcuts inside the editor:
Ctrl + Enter — save the comment.
Escape — cancel editing without saving.
Comments are visible to every user who can access the conversation. You can edit or delete your own comments; other users' comments are read-only.
Review Mode and Scoring
Switch from Default view to Review mode using the toggle in the top bar. Review mode expands the detail panel to full width and reveals the scoring interface on the right side.
In Review mode:
The scorecard assigned to the agent's team appears on the right panel.
Each scorecard item can be rated Yes, No, or N/A.
A progress bar shows your completion percentage as you score.
When finished, click Submit Scores to save.
Use Clean All to reset scores if you need to start over.
The scored agent receives a notification about the review.
You can also leave comments while in Review mode by toggling the comments panel.
Disputes
If disputes are enabled for a scorecard, agents can challenge a submitted score:
The agent initiates a dispute on a scored conversation and provides a suggested score with reasoning.
The reviewer sees a Dispute Pending indicator and can toggle between the original score and the agent's suggested score.
The reviewer can Approve the dispute (accepting the agent's score), Void it (keeping the original score), or Partially approve it.
Sorting
You can sort the conversation list by:
Newest or oldest first
Longest or shortest duration
Relevance (when a text search is active)
The current sort order is shown above the list and can be changed at any time.
Bulk Actions
When filters are applied, bulk operations are available from the conversation list:
Download XLSX — export the filtered set of conversations to an Excel file. There is no separate download permission; the action is allowed for users whose role includes one of the analytics view permissions (
Dashboard: View,Dashboard: View Team,Dashboard: View Agent, orSessions: View Own). The standard Full / Team / Own role bundles each include one of these, so a role configured from the defaults can export, but a custom role with onlySessions: VieworSessions: View Teamwill see the menu and get an insufficient-privileges response. Click the download icon above the list and choose one of two modes:XLSX — one row per conversation with the built-in columns. If AI summary prompts are configured, each prompt's output is appended as an additional column with the full summary text; structured (JSON) summaries are kept as the raw JSON string in this mode.
XLSX Expanded — same rows as XLSX, but when a summary produces a structured (JSON) result, each top-level key is split into its own column named
<Summary name> - <key>. This applies both to your own summary prompts and to the built-in analyses Ender Turing ships with, such as CSAT Dynamics and Interaction Intensity Dynamics. If the export would produce more than 200 expanded summary columns in total, only the keys that repeat across multiple conversations are expanded (most frequent first), and the remaining keys stay in the compact JSON column for that summary.
Both modes include the same built-in columns, with localized headers:
Column | What it contains |
ID | The conversation's internal Ender Turing identifier (always the first column) |
Type | The conversation channel — |
Date / Time | The conversation's start date and time |
Duration | Conversation length |
Direction | Call direction — |
Client | Caller (for inbound) or destination (for outbound) |
Agent / Agent Email / Team | The agent assigned to the conversation, their email, and their team |
ET-Link | A direct link back to the conversation in Ender Turing |
Categories | Topics matched on the conversation |
CRM Statuses / Queue Names | Metadata supplied by the integrated CRM or telephony platform |
Comments | Comments left by users |
Ender Comments | System-generated notes from automations |
Partner | The partner name resolved from the client identifier, when configured |
Type and Direction values are exported as raw lowercase enum strings (Type: call, chat, email, web, loan; Direction: in, out) regardless of the export language. Only the column headers are localized.
In addition, any identifier or link fields your telephony or CRM platform sends with the conversation (field names ending in _id or _url) are added as their own columns, using the field name as the header.
CSAT and Interaction Intensity columns
If your organization runs the built-in CSAT Dynamics or Interaction Intensity Dynamics analysis, their results are structured summaries, so XLSX Expanded gives every metric its own column instead of one cell of JSON you have to take apart by hand:
Analysis | Columns in the export |
CSAT Dynamics |
|
Interaction Intensity Dynamics |
|
Points to keep in mind:
Column headers use the field names exactly as the analysis produces them, so they are not translated with the rest of the header row.
Plain XLSX deliberately keeps the whole result in a single cell as JSON. Choose XLSX Expanded when you want one column per metric.
A column appears only when at least one conversation in the exported set has that result. Conversations that were never analysed leave the cells empty.
Scores are written as text, like every other expanded summary value. Convert the column in your spreadsheet if you want to average or chart the scores.
The columns follow the same filters and date range as the rest of the export, so you can narrow the set first and then correlate satisfaction or intensity with agent, team, topic, or duration.
CSI in the export
If a CSI (Customer Satisfaction Index) value arrives with the conversation, it is exported in a dedicated CSI column. CSI can come from your telephony platform, from a CRM integration (Zendesk, for example, maps its configured CSAT field into CSI), or from conversations uploaded or updated through the API. This lets you correlate CSI with the other exported columns — topics, duration, agent, team — in your own spreadsheet or BI tool.
A few things to expect:
The CSI column only appears when at least one conversation in the exported set actually carries a CSI value. If none of them do, there is no CSI column in the file at all.
Conversations without a CSI value leave the cell empty, the same as any other missing metadata.
CSI is not something you enter in Ender Turing. It arrives with the conversation from the source platform, so if you expect CSI and do not see the column, check that the telephony, CRM, or API integration feeding those conversations is sending it.
How values are written into the sheet
CSI and Duration land as numeric cells, so you can average them, chart them, or reference them in formulas without converting the column first.
Identifier and link fields are always written as text, even when they look like numbers. This is deliberate: long numeric IDs lose their last digits when a spreadsheet parses them as numbers.
Everything inside expanded summary columns is written as text, because summary output is treated as text throughout. Numbers there are not numeric cells, and yes/no values appear as
TrueorFalse.Empty and unavailable values are written as empty cells.
Access and Permissions
Access to Conversations depends on the permissions assigned to your role.
Action | Required permission |
View all conversations | Sessions: View |
View only your team's conversations | Sessions: View Team |
View only your own conversations (agent workspace) | Sessions: View Own |
Score / review conversations | Sessions: Score |
Leave comments on any conversation | Sessions: Comment |
Leave comments on own agent's conversations | Sessions: Comment Own |
Edit transcript text | Sessions: Transcript Edit |
Edit session metadata | Sessions: Meta Edit |
Delete conversations | Sessions: Delete |
Create manual sessions | Sessions: Create |
Initiate a dispute | Sessions: Dispute Initiate |
Approve or partially approve disputes | The original reviewer of the scored conversation |
Void (reject) disputes | Sessions: Dispute Manage, or the original reviewer |
"Your own conversations" means conversations assigned to the agent profile linked to your user account. If your user has no linked Agent, no conversation counts as your own for the Sessions: View Own permission, including conversations that have no agent assigned. An administrator can link your user to an agent in Users & Agents.
With Sessions: View Team, you can open conversations that belong to the teams assigned to your user.
Review mode is available when you are the agent associated with the conversation, or when you have the Score or View Team permission.
Users with view-only access can still read transcripts, play media, copy transcript text, copy the conversation link, and use share options. Anyone who can open a conversation can also see its attachments' extracted text and download the attachments — no extra permission is needed. They cannot score, comment, edit, or delete conversations.
