Automate Perplexity research with n8n

#01What does the Perplexity node do?

According to Perplexity's n8n integration guide (checked November 2026):

Older n8n tutorials show a "Chat" or "Message a Model" operation with Sonar model names. Sonar support ended on 27 September 2026, so update those workflows to the Agent resource.

#02How do you set it up?

1. Update n8n to 2.14.0 or later (Cloud updates automatically, but self-hosted needs an upgrade).
2. Create an API key in the Perplexity API console and set a spend alert.
3. Add the credential in n8n: Settings → Credentials → Perplexity API.
4. Add a Perplexity node to a test workflow and run one Search and one Agent call.

New to n8n? Start with n8n for Marketers and your first workflow.

#03What is the core workflow pattern?

Schedule Trigger (Mon 06:00)
  → Google Sheets: read "prompts" tab (prompt, cluster, priority)
  → Loop Over Items (batch of 5)
     → Perplexity (Agent): answer prompt, JSON schema output
     → Perplexity (Search): same prompt as query, top 10
     → Code: extract brands mentioned, cited domains, our rank
     → Google Sheets: append to "results" tab
     → Wait (2s)
  → Code: summarise week vs last week
  → Slack: post summary to #ai-visibility (internal channel)

Use structured output

Ask the Agent resource for JSON so parsing is reliable:

{
  "type": "object",
  "properties": {
    "brands_mentioned": { "type": "array", "items": { "type": "string" } },
    "recommended_first": { "type": "string" },
    "cited_urls": { "type": "array", "items": { "type": "string" } }
  },
  "required": ["brands_mentioned", "cited_urls"]
}

System instructions: "Answer as you normally would for a buyer, then fill the schema. List every brand you mention in the order you mention it."

Our view: the instruction slightly changes the answer compared with a plain query. Keep the wording fixed so week-on-week comparisons stay fair, and spot-check against the consumer app monthly.

#04Which other workflows are worth building?

#05How do you handle errors and costs?

• Retry on Fail on Perplexity nodes, with a few retries and a pause between them.
• 429 rate limits: batch with Loop Over Items, and add a Wait node between batches (n8n error handling).
• Error workflow: route failures to a Slack alert with the prompt that failed.
• Cost controls: use the fast effort or fast search, keep prompt sets focused, and log the cost per run.
• Executions: on n8n Cloud, each workflow run counts as an execution. Batch prompts into one run rather than running per prompt (n8n costs).

#06How should you store results?

A sheet works for up to a few thousand rows a month. After that, move to a database such as Postgres or Supabase.

#07Worked example: Ledgerline

Ledgerline is a fictional accounts-payable software company. Details are illustrative.

Ledgerline moved its weekly API script (lesson 11) into n8n so marketing ops could own it.

Walkthrough

1. Updated n8n and added the Perplexity credential.
2. Built the core pattern with 40 prompts in batches of 5.
3. Added the JSON schema and a Code node to compute Ledgerline's position.
4. Added the Search resource for the same queries.
5. Set Retry on Fail and an error workflow.
6. Ran three weeks in parallel with the old script and compared the results, which matched within normal answer variation.
7. Retired the script.

#08Common mistakes

• Copying old Sonar tutorials. Use the Agent resource.
• One execution per prompt. That's expensive on Cloud, so batch instead.
• No version logging. Record the effort version in each row.
• Auto-posting externally. Always add an approval step.

#09Try it yourself

Build the core workflow with five prompts and one Slack message to yourself. Run it manually, check the sheet, then schedule it weekly. Add the Search resource in week two.

#10Quick quiz

1. What are the three resources in the current Perplexity node?

Answer: Agent, Search and Embeddings.

2. How do you handle 429 errors?

Answer: Use Retry on Fail, batching and Wait nodes.

3. Why use a JSON schema?

Answer: So brands and citations can be parsed reliably.

#11What should you check before the next lesson?

• n8n is version 2.14.0 or later, and the credential works.
• The core workflow writes to a results sheet.
• An error workflow and spend alerts are on.

Previous: The Agent API and Search API. Next: How Perplexity chooses sources. Or return to the Perplexity for Marketers track. Related: Computer and Automations, n8n AI visibility monitoring, n8n competitor monitoring.

#12Common questions

Which n8n version do I need for the Perplexity node? Perplexity's integration guide lists n8n 2.14.0 or later for the current node with Agent, Search and Embeddings resources. Update n8n first if you don't see them.

Should I use n8n or Perplexity Computer Automations? Use Computer Automations for quick, marketer-owned jobs. Use n8n when you need exact API settings, logging, approvals and connections to your CRM, sheets or databases.

How do I stop a workflow from running up costs? Limit the prompt set, use the fast effort or fast search where possible, add a Wait node between batches, and review API spend weekly at first.