Skip to content
AskElephant Knowledge Base home
AskElephant Knowledge Base home

Output Error

Introduction

If your workflow completes without errors but the result is blank, incomplete, or wrong, the cause is almost always upstream of the output — a missing variable, a missing upstream node, or a prompt that doesn't give the AI enough to work with.

Start With Missing Variables

This is the #1 cause of empty output. Review every prompt node and verify it references the data it's supposed to analyze.

  1. Open each prompt node.

  2. Confirm it uses {{variable_name}} to reference upstream data.

  3. If it doesn't, type {{ in the prompt to see available variables and pick the one you need.

  4. Double-check the Input Requirements section lists every variable the prompt depends on (for example, - Transcript: {{transcript}}).

🚩 Red flag. Output that reads "I don't have the information to process this" or similar almost always means a prompt is missing a variable.

When the AI Asks for Info Instead of Running

If the output reads like questions the AI is asking you — "Can you provide the transcript?" or "Which meeting are you referring to?" — the AI didn't receive the data it needed. This is a prompt or variable issue, not a platform bug.

How to fix it:

  1. Open the prompt node and confirm the Input Requirements section explicitly lists each variable the prompt depends on.

  2. Move data references to the top of the prompt and instructions below. Models handle that order better.

  3. Run the workflow and check the upstream node's output to confirm it actually produced the variable you're referencing.

Verify Your Trigger Fired

  • For a New Meeting trigger: record a test meeting and confirm the workflow runs in Workflow History.

  • For a Deal Updated trigger: update a test deal in HubSpot.

  • For a signal or tag trigger: confirm the signal is set up correctly.

Check That the Data Exists

The AI can't create data from nothing. If you ask "what company is this from?" but the contact doesn't have a company assigned, the AI will say it doesn't have that information.

Before assuming the workflow is broken, check the upstream system (HubSpot, Salesforce, etc.) to confirm the data you expect is there.

Review Your Prompt Instructions

Vague instructions produce vague results.

  • Vague: "Tell me about this deal."

  • Specific: "Summarize the deal size, expected close date, and next steps from our last meeting with this customer."

Output Isn't Arriving Where Expected

If the workflow completes but the email doesn't arrive, the Slack message doesn't show up, or the CRM field doesn't update, check the output node.

Email:

  • Is the email address spelled correctly?

  • Does that email account actually exist?

Slack:

  • Is the correct channel selected?

  • Is AskElephant added to that channel? (AskElephant can't send to channels it's not a member of.)

  • Is the Slack integration connected at both workspace and personal level?

HubSpot or Salesforce:

  • Is the object type (Deal, Contact, Company, etc.) selected?

  • Is a specific record selected?

  • Do you have permission to update that record? Ask your workspace owner or CRM admin to verify.

Conversation:

  • The output appears in a chat conversation with the name set in the Create Conversation node. Refresh the page if you don't see it.

Workflow Runs Partially Then Stops

If the run shows Failed partway through:

  1. Click into Workflow History and open the failed run.

  2. Find the node marked Failed and click it to see the error.

  3. Common failure points:

    • Incorrect node configuration: sending to Slack without a channel, or creating a HubSpot record without required fields.

    • AI can't access required data: the prompt asks for something the workflow doesn't have. Check upstream variables.

Writing Effective Prompts for Property Updates

When a prompt node needs to write exact content to a property — especially HTML or rich text — follow these rules:

  • Avoid markdown code fences. Wrapping templated content in triple backticks can confuse the model.

  • Be explicit about HTML. If the variable contains HTML, say whether to preserve markup exactly, convert to plain text, or extract meaning — otherwise the model may strip tags.

  • Use clear, direct instructions. State what to derive, where it comes from, and format constraints in plain language.

  • Keep instructions concise for verbatim copying. For near-verbatim use of another node's output, use short explicit language like "reproduce the given string without summarizing."

Example prompt for verbatim HTML property writing:

Use the following string as the entire property value. It may contain HTML
(tags, attributes, entities, and whitespace). Copy it character-for-character:
do not add a preamble, do not summarize, do not fix or reformat markup, and
do not strip or decode HTML unless the string is already plain text.

Value to write:
{{node_output}}

If the string above is empty or only whitespace, write an empty string.

Still Stuck?

If output is still empty or wrong after checking all of the above, contact support with the workflow name, a link, and a specific run ID from Workflow History. See Workflows for the full troubleshooting guide.

Powered by Plain