Source Citations

Every suggestion card includes one or more source citations that show exactly which documents informed the AI-generated answer. Sources let you verify accuracy before responding to the customer and provide a trail back to the original content.

Source Card Contents

imageSource citation card showing citation number, document title link, page number, content snippet, relevance score, and feedback buttons

A source citation within an expanded suggestion card

When you expand a suggestion card, the sources section appears below the answer. Each source displays:

Field	Description
Citation Number	A sequential number (1, 2, 3...) matching inline citations in the answer text (e.g., [1], [2]).
Title	The name of the source document. Clicking the title opens the original document in a new tab.
Page Number	The specific page within the document where the relevant content was found. Clicking the page link opens the document at that page. Only shown for paginated documents (e.g., PDFs).
Content Snippet	A preview of the matched text from the document, limited to a few lines. This gives you enough context to judge relevance without opening the full document.
Relevance Score	A percentage indicating how closely this source matched the customer's query. Displayed as a colored badge.
Feedback	Per-source thumbs-up/thumbs-down buttons with the label "Was this helpful?".

How Sources Are Ranked

Sources are ordered by relevance score, highest first. The ranking is determined by the RAG pipeline:

1. The customer query is converted into an embedding vector.
2. The knowledge base is searched for document chunks with the closest vector similarity.
3. Chunks are scored based on cosine similarity to the query embedding.
4. If feedback-based reranking is enabled, chunk scores are adjusted based on historical feedback from agents (sources that have been consistently rated unhelpful are deprioritized).
5. The top-scoring chunks are returned as source citations, ordered by score.

Relevance score guidelines:

Score	Meaning
80%+	Strong match -- the source is highly relevant to the query
60%–80%	Partial match -- the source is related but may not fully address the question
Below 60%	Weak match -- verify the content carefully before using

TIP

If the top source has a low relevance score, the answer may be less reliable. Consider searching manually with different phrasing using the Knowledge Search bar.

Source Feedback

Each source has its own thumbs-up and thumbs-down buttons, separate from the answer feedback. Rating individual sources helps the system learn which documents are useful for specific types of questions.

• Thumbs up -- Marks the source as helpful for this query
• Thumbs down -- Opens a feedback dialog where you can select a reason (Irrelevant, Incorrect, Too Generic, Misleading) and add an optional comment

Sources you have rated are visually highlighted -- green border for positive, red for negative. You can remove your rating by clicking the same button again.

Your source feedback has real impact: consistently downvoted sources are automatically deprioritized in future results. See Source Feedback for details on how this works.

Omni Sources vs. CCAI Sources

Agent Assist supports two types of source citations depending on your deployment mode:

Omni Sources

Omni sources come from the built-in OmniBots RAG pipeline (omni mode). They provide full metadata:

• Document title with clickable link to the original file
• Exact page number with a direct link
• Content snippet from the matched chunk
• Precise relevance score as a percentage
• Per-source feedback controls

CCAI Sources

CCAI sources come from Google Dialogflow knowledge connectors (google_native mode). They provide a more limited set of metadata:

• Document title (link may not be available depending on connector configuration)
• Content snippet
• Confidence score from the Dialogflow knowledge connector

The key difference is that omni sources include richer metadata -- page numbers, direct document links, and per-source feedback -- because the RAG pipeline has full control over the indexing and retrieval process. CCAI sources rely on the metadata that the Dialogflow connector returns.

Using Sources Effectively

• Verify before responding. Click the source title to open the original document and confirm the answer is accurate, especially for policy-sensitive topics.
• Check the snippet. The preview often gives you enough context to judge relevance without opening the full document.
• Use page links. For PDF sources, click the page number to jump directly to the relevant page rather than scrolling through the entire document.
• Watch the scores. A suggestion backed by multiple high-scoring sources is more trustworthy than one with a single low-scoring source.
• Provide feedback. Use the per-source feedback buttons to flag unhelpful or incorrect sources. This directly improves future results for all agents.

Next Steps

• Learn how to search manually: Knowledge Search
• Rate individual sources: Source Feedback