The Eddy AI - Assistive Search analytics page gives you a complete picture of how readers interact with Eddy AI across your knowledge base — tracking conversations, query volumes, answered and unanswered searches, and reader feedback. Use this page to identify content gaps, measure AI search effectiveness, and take targeted actions to improve your documentation coverage.
To know more about the languages supported by Eddy AI, click here.
Why use Eddy AI search analytics
Use the Eddy AI search analytics page when you want to:
- Identify unanswered queries — find topics your readers search for but cannot find answers to, signalling content gaps.
- Monitor AI search performance — track answered vs. unanswered query trends over time to assess how well Eddy AI serves your readers.
- Audit reader feedback — review likes, dislikes, and feedback comments to evaluate the quality and relevance of AI-generated responses.
- Investigate individual conversations — drill into specific user interactions to validate citation accuracy and response quality.
Before you begin
You must have admin, owner, or custom access to the Knowledge base portal.
How to access Eddy AI search analytics
- Navigate to Analytics () in the left navigation bar in the Knowledge base portal.
- In the left navigation pane, click Eddy AI. The Eddy AI - Assistive Search page appears.
- By default, data for the current month is displayed.
.png)
Using filters to customise the data
By default, the page shows metrics for the current month. To adjust the timeframe or scope:
- Click the Month filter dropdown.
- Choose a predefined range (This month, Last month) or select Custom to specify a particular month. When using the Custom filter, you can select only the month from the calendar.
- Click Filter to narrow results by application type: All, Knowledge base site, Knowledge base widgets, Chatbot, Public API, or extensions such as Freshdesk or Zendesk.
Understanding the analytics page
Metric tiles
The top of the page displays the following performance metrics as tiles:
| Field | Description |
|---|---|
| Total conversations | Total number of conversations in Eddy AI, where one or more queries count as a conversation. Available only in KB site 2.0. A single message is also counted as a conversation. |
| Total queries | Total count of individual questions asked to Eddy AI. Queries are individual questions within a conversation. |
| Answered queries | Total number of queries that returned a result. |
| Unanswered queries | Total number of queries that did not produce any results. |
| Total likes | Total number of likes received on search responses, indicating a satisfactory answer. |
| Total dislikes | Total number of dislikes received, indicating an unsatisfactory response. Dislike reasons include: difficult to understand, inaccurate or irrelevant content, incorrect sources/references/citations, missing information, don't like the style, and others (with a free-text comment). |
| Total feedback | Total number of feedback comments received for AI assistive search responses. |
If multiple chatbots are deployed, data from all chatbot instances is merged into a single result set. Similarly, data from all knowledge base widget instances is combined.
Query analysis
The Query analysis graph lets you explore query performance over time. You can interact with it using the tabs along the top:
- The horizontal axis displays dates based on the selected Month filter.
- All — view the comparison of answered queries, unanswered queries, and conversations in a single graph.
- Answered — view queries that returned a result.
- Unanswered — view queries that did not produce any result.
- Likes — view likes received on search responses.
- Dislikes — view dislikes received on search responses.
- Feedback — view feedback comments received for search responses.
- Conversations — view conversation trends. Available only for KB site 2.0 projects.
- Hover over the graph to see metrics for a specific date.
- Export image — download the Query analysis graph as a .png file.
Keep Unanswered searches as low as possible relative to Answered searches. Unanswered searches typically occur when relevant information is missing from your documentation. Keeping content up to date helps reduce these instances.
Conversation depth metrics
The Conversation depth metrics graph is available only for KB site 2.0 projects. It shows the distribution of queries per conversation alongside the total number of conversations.
- Hover over each bar to see the exact conversation count. The graph tracks conversations from 1 to 5 queries, with anything beyond 5 queries grouped as 6+.
- Click the Export image () icon to download the graph as a .png file.
Feedback chart
The Feedback chart displays the breakdown of dislikes by reason, along with counts for each category. Use this chart to identify which aspects of Eddy AI responses readers find most unsatisfactory.
Popular queries
This section shows the top five search themes for the selected month, grouped by keywords. It highlights the topics readers search for most frequently and how often each theme appears.
Popular queries are generated only when at least one full month of data is available.
Unanswered queries
This section shows the top five search themes for the selected month where Eddy AI could not return a result. These themes highlight the most common content gaps in your knowledge base.
Query post-processing runs monthly across 30 days of data. The data is sent to OpenAI, which identifies the most common keywords and groups them into up to 10 primary topics. Queries that do not fit a topic are grouped under "Others."
Most referenced articles
This section lists the five articles most frequently cited by Eddy AI in search results for the selected month, along with the citation count for each.
Hover over an article title and click View in Knowledge base () to open the article in the Knowledge base site.
Conversations
The Conversations section provides a detailed, tabular view of every user interaction — including the full query, response status, and any feedback received.
The table includes the following columns and controls:
a. Search field — enter a keyword or phrase to find a specific conversation or question.
b. Conversations — displays the first question of each conversation; subsequent questions are collapsed by default. Expand a row to view all questions, or click a query to open a side panel showing the full conversation, user information (if available), user group, and the date and time of the interaction.
If a conversation started in a previous month and continues into the current month, it is attributed to the month in which it started.
c. Query topic — topics are assigned from the popular queries list. At least 30 days of data is required to generate a query topic. Only the top 5 popular topics are displayed.
d. Response — indicates whether the query was answered () or unanswered ().
e. Feedback icon — shows whether feedback was a like () or dislike ().
f. Feedback — displays a Feedback tag if the query received feedback. Click to navigate to the Eddy AI feedback manager to view the full feedback details.
g. User info — for Mixed and Private projects, displays the name or email address of the user who submitted the query (when available). For JWT users without a name, the associated email address is shown if available.
h. User group — displays the reader group or user group associated with the user, if applicable.
i. Filter — filter conversations by:
- Response status: All, Answered, or Unanswered.
- Feedback: All, None, Like, or Dislike. This covers feedback received on Eddy AI responses from the Knowledge base site, Widget, Chatbot, and Public API.
- Query topic: Filters queries from previous months (unavailable for the current month). Options include Select all (displays both five popular queries and five unanswered queries), Popular queries, and Unanswered queries.
j. Export CSV — download the filtered conversation data as a .csv file. Click the Download () icon to retrieve the last set of files exported.
When the Anonymize reader information setting is enabled, reader-identifiable information linked to Eddy AI conversations is anonymised. User names, email addresses, and reader IDs will not appear in analytics or exports.
Viewing Eddy AI source articles
When you expand a query in the Conversations section, the response displays numbered citations (e.g., 1 2 3) at the end. Hovering over a citation number reveals the article title; clicking it opens the source article in the Knowledge base site.
This lets you quickly validate the sources Eddy AI used to generate a response and identify content that may need to be updated. Citation links always point to the latest published version of the source article. If the article has since been deleted, the link will display an Error 404 page.
Citation details are not available in OpenAPI or API documentation. They are only accessible from the Knowledge base portal or the Knowledge base site.
Understanding AI credit usage and analytics
In some cases, the query count shown in Eddy AI search analytics may differ from the usage credits consumed under the Eddy AI feature. This can happen for two reasons:
-
Multi-language projects — if your project supports multiple languages (for example, English and German), credits are consumed across all languages. However, the Analytics view for a single language shows only queries made in that language, which may appear lower than total credit usage.
-
Non-chargeable interactions — simple greetings such as "Hi" or "Hello" are recorded in Analytics for completeness but do not consume credits. As a result, the query count in Analytics can be slightly higher than the number of credits used.
If you reach the usage limit, contact support to update or verify your detailed usage data.
Best practices
- Review unanswered queries monthly — use the Unanswered queries section to identify recurring content gaps and update or create articles to address them.
- Act on dislike feedback — check the Feedback chart regularly. Responses marked as "inaccurate" or "missing information" are the highest-signal items to investigate.
- Cross-reference most-referenced articles — these articles carry the highest weight in Eddy AI responses. Ensure they are accurate, up to date, and comprehensive.
- Use the Conversations table to validate citations — expand individual queries to confirm that source articles are correct and current. Deleted source articles result in broken citation links.
FAQ
Why am I unable to filter the data on a weekly basis?
Popular queries and unanswered search keywords are derived from algorithmic grouping methods that require at least 30 days of data to categorise keywords based on response frequency. Weekly filtering is not supported because the dataset is too small for accurate grouping.
How do I resolve queries for which no response is generated?
Review the keywords listed in the Unanswered queries section. Use those keywords to identify relevant articles and update or create content that directly addresses the topic. You can also use the Eddy AI feedback manager to track unanswered query feedback to resolution.
How does Eddy AI conversation analytics work when a conversation spans multiple languages?
The conversation is recorded in the conversations table for each language used. However, specific metrics such as total searches, answered queries, unanswered queries, and feedback are only updated for the language in which each interaction occurred.
How does Eddy AI conversation analytics work when a conversation spans multiple workspaces?
The conversation is recorded in the conversations table for each workspace used. Specific metrics such as total searches, answered queries, unanswered queries, and feedback are only updated for the languages used within the respective workspace.
How are queries different from conversations?
A query is an individual question asked to Eddy AI. A conversation is a session that contains one or more queries. For example, a reader asking three follow-up questions in a single session counts as one conversation with three queries.