Agent Produces Poor Quality or Irrelevant Output in Eve

Last updated: March 10, 2026

Overview

This article resolves situations where an agent runs successfully but produces output that is off-target, incomplete, or uses the wrong information. This typically occurs when agent instructions are too vague, the agent is reading from the wrong documents, or example files need updating. It affects both Chat Agents and Drafting Agents, whether run manually or automatically.

Before You Start — Diagnosis

Confirm the following before attempting any fixes:

  • The agent did run and produce output — if no output was generated at all, see the article on agents not running when expected

  • You can identify what's wrong with the output — is it using the wrong documents, missing key information, or structured incorrectly?

  • You have Firm Admin permissions if this is an Auto Agent with trigger conditions that may need refinement

  • You know which documents the agent should have focused on versus which it actually used

Fixes

Possible Cause: Agent Instructions Are Too Vague or Incomplete

Cause: The agent's instructions do not specify what to include, what to exclude, or how to structure the output. Without clear direction, Eve makes broad assumptions about what to read and what to produce, leading to unfocused or irrelevant results.

Steps:

  1. Navigate to the agent in question — for Drafting Agents, go to "Draft a document" in any matter and locate the agent; for Chat Agents, access the agent from the Agents page or matter-level agent panel

  2. Select "Edit Agent" to open the agent configuration

  3. Review the existing instructions and identify missing specificity:

- Does it specify which document types to focus on? (e.g., "medical records," "adjuster correspondence," "discovery responses")

- Does it define recency expectations? (e.g., "documents uploaded in the last 30 days")

- Does it explain what should be excluded? (e.g., "do not include internal firm emails")

- Does it describe the desired output structure or length? (e.g., "provide a 2-paragraph summary," "list findings in bullet points")

  1. Rewrite the instructions to be explicit about scope, source material, and output format — treat the instructions as if you are briefing someone who has never seen the case before

  2. Save the updated agent

  3. Run the agent manually on a test matter to confirm the output improves

Expected Outcome: The agent produces output that is focused, relevant, and structured as intended. The instructions now guide Eve to the correct documents and clearly define what the output should contain.

Possible Cause: Agent Is Reading From the Wrong Documents

Cause: For Auto Agents, if instructions do not explicitly define which documents to focus on, Eve assembles matter context dynamically based on trigger conditions and available files. If the trigger condition is broad (e.g., "any document uploaded") and instructions are vague, the agent may pull from irrelevant or outdated documents.

Steps:

  1. Open the Auto Agent Log (navigate to Agents in the sidebar, then select the Auto Agent Log tab)

  2. Locate the agent and expand the matter row where poor output was generated

  3. Review the "Why the agent produced output" explanation — this shows which documents Eve identified and why it selected them

  4. Compare the documents Eve used against the documents you expected it to use

  5. Edit the agent's instructions to specify exact document types, folders, or recency filters — examples:

- "Focus on documents in the Medical Records folder"

- "Summarize medical records uploaded in the last 7 days"

- "Use only adjuster correspondence, not internal firm emails"

  1. If the agent has a trigger condition, confirm it aligns with the document scope in the instructions — for example, if instructions say "focus on new medical records," the trigger condition should include "new medical records uploaded" not "any document uploaded"

  2. Save the updated agent and run it manually on a test matter to confirm the correct documents are now being used

Expected Outcome: The agent focuses on the correct subset of documents as defined in the instructions. The Auto Agent Log shows Eve selecting the intended files, and output quality improves.

Possible Cause: Central Library Reference Files Are Outdated or Missing

Cause: If the agent is configured to reference Central Library files (such as objection banks, style guides, or firm templates), but those files are out of date or missing, the agent produces output that does not align with current firm standards or fails to incorporate required content.

Steps:

  1. Navigate to the agent configuration and confirm which Central Library files are selected under the reference files section

  2. Navigate to the Central Library (accessible from the main sidebar)

  3. Locate the files the agent is referencing and review their content:

- Are they current and accurate?

- Do they contain the information the agent's instructions expect to draw from?

- Are there newer versions that should replace them?

  1. If the files are outdated, upload updated versions to the Central Library with the same file names so the agent automatically references the new versions

  2. If the agent's instructions reference specific content that is missing from the Central Library files, either add that content to the files or update the instructions to remove the reference

  3. Save any changes and run the agent manually on a test matter to confirm it now incorporates the correct reference material

Expected Outcome: The agent draws from current, accurate Central Library files and incorporates that information into its output as instructed. Output aligns with firm standards and includes required content.

Possible Cause: Drafting Agent Example Documents Are Poor Quality or Misaligned

Cause: Drafting Agents learn structure and tone from example documents uploaded during agent creation. If those examples are inconsistent, poorly formatted, or do not represent the firm's current standards, the agent produces output that mirrors those flaws.

Steps:

  1. Navigate to the Drafting Agent configuration and locate the example documents section

  2. Review the example documents currently uploaded — ask:

- Are they high-quality, well-structured examples of the output type?

- Do they reflect the firm's current style and formatting preferences?

- Are there multiple examples, and do they vary in a way that might confuse the agent?

  1. If the examples are poor quality or outdated, remove them and upload new, high-quality examples that clearly represent the desired output

  2. Ensure the examples are consistent in structure, tone, and formatting — if the agent is meant to produce formal demand letters, all examples should be formal demand letters in the same style

  3. Update the agent's instructions to clarify how the examples should be used (e.g., "follow the structure and tone of the example demand letters")

  4. Save the updated agent and run it manually on a test matter to confirm output quality improves

Expected Outcome: The agent produces output that mirrors the structure, tone, and formatting of the updated example documents. Output quality is consistent and aligns with firm standards.

Possible Cause: Agent Instructions Conflict With Trigger Condition Scope

Cause: For Auto Agents, if the instructions expect the agent to focus on a specific document type but the trigger condition fires on a different type, the agent runs at the wrong time or uses the wrong context, leading to irrelevant output.

Steps:

  1. Review the agent's trigger condition (navigate to Agents, select the agent, and view the "When to Run" configuration)

  2. Compare the trigger condition's "Additional Condition" filter to the document types specified in the instructions — examples of misalignment:

- Trigger: "any document uploaded" but Instructions: "summarize new medical records" → agent runs on every upload but instructions expect only medical records

- Trigger: "new adjuster correspondence uploaded" but Instructions: "summarize all case documents" → trigger is too narrow for what instructions expect

  1. Align the trigger condition and instructions so they expect the same scope:

- If instructions are specific, narrow the trigger condition to match (e.g., change "any document" to "new medical records uploaded")

- If the trigger is specific, update instructions to clarify the agent should focus only on the documents the trigger fires on

  1. Save the updated agent configuration

  2. Use the trigger condition testing panel to validate the condition now fires on the correct matters and document types

  3. Run the agent manually on a test matter to confirm output is now relevant

Expected Outcome: The agent's trigger condition and instructions are aligned — the agent runs when the right documents appear and focuses on those documents in its output. Irrelevant runs and off-target output are eliminated.