What n8n is and how to use it to automate marketing

What n8n is, explained simply

n8n’s own documentation describes it as a workflow automation tool that combines AI capabilities with business process automation. A workflow is a sequence of actions that repeat every time: export data, move it into a spreadsheet, notify the team. Tools like this are also called app-integration services: they connect programs that don’t «talk» to each other on their own.

In n8n, you build a process as a workflow: a set of nodes connected together to automate it. A node is a single step in a workflow: fetch data, change it, send it on. Workflows are built on the editor’s canvas: you add nodes and connect them. Where the built-in options aren’t enough, you can write code: the Code node runs JavaScript or Python as a regular workflow step.

An example of a simple workflow:

1. A visitor submits a form on the site, and the site passes its data to n8n.
2. n8n checks whether the required fields are filled in.
3. The inquiry is added as a row in a shared spreadsheet.
4. The manager gets a notification in a messaging app.

Each step here is a separate node. A person no longer copies the inquiry by hand, but still answers the client.

Is n8n a neural network?

No, n8n is an automation platform, not a neural network. A language model — a neural network that writes and analyzes text — is connected to a workflow as a separate node. For example, n8n has an Anthropic node for working with that company’s models (Claude), and an Ollama Chat Model node for local models running on your own hardware.

Without such a node, n8n doesn’t generate anything: it only runs the steps you’ve defined.

How n8n is built: workflows, nodes and triggers

Every workflow starts with a trigger — a special node that starts the workflow under certain conditions. The most common options:

• On a schedule. The Schedule Trigger node starts the workflow at set intervals or at a set time.
• On an event. The Webhook node receives data from apps and services when something happens on their end. A webhook is an address that another service sends data to on its own, for example on a new inquiry.
• Manually. A workflow can be started with a button in the editor — that’s how you test it before putting it into production.

New workflows aren’t active by default. For a workflow to run on its own, you activate it, and from then on it runs every time the trigger’s condition is met.

After the trigger come the actions: fetch data, process it, send it or save it. Many services have ready-made nodes. If there’s no ready-made node, the HTTP Request node helps: it sends requests to any app or service that has a REST API. API is a programming interface through which one service gets data from another, or sends it commands.

A diagram of a typical workflow:

trigger → fetch data → process → send or save → notify

Three more concepts you can’t do without:

• Credentials — saved login data a workflow uses to connect to external services, APIs or databases.
• Execution — a single run of a workflow. The execution log can be viewed per workflow or for all of them at once. It shows what succeeded and what didn’t.
• Error handling. You can assign a separate error-handling workflow to each workflow, which runs if an execution fails — for example, to send an alert.

Cloud version or your own server

n8n has two main options: n8n Cloud, which fully runs n8n for you, and self-hosted — installing it on your own infrastructure.

	n8n Cloud	Self-hosted
Where it runs	On n8n’s side	On your infrastructure: your own servers or a private cloud
Technical knowledge needed	No	Yes, for installation and setup
Free option	Trial period	Free Community edition with most of the features
Paid options	Yes, several plans	Yes, paid editions with a license key

Self-hosted means you install and maintain the software yourself. The documentation describes several methods: a single-command install (requires Docker), Docker Compose, deployment with cloud providers. The documentation calls npm installation deprecated as of n8n version 3.0. Docker is an environment that runs a program in an isolated container with everything it needs.

Separate sections of the self-hosting documentation cover configuration, security and keeping an instance running. In practice, that means updates, backups, and protecting the server and access are your responsibility.

License. n8n is distributed under a fair-code model. The free self-hosted edition uses the Sustainable Use License (SUL). n8n’s Help Center explains: if only you and people from your organization create and edit workflows, you can use the Community edition for free. A commercial license is needed if workflows can be created or configured by people outside the organization, as well as to host n8n as a service, ship it under your own brand, or use features that require a license key.

Without a license key, n8n runs as the free Community edition. Some features are only available on paid plans — for example, environments, external secrets storage, single sign-on (SSO), shared access to workflows and credentials, and Git-based version control. Check current terms and pricing on n8n’s official site: they change, and we don’t quote them in this article.

What to consider if you work with client data

If a workflow processes inquiries, contacts or correspondence, answer a few questions in advance:

• where the workflows physically run and the data is stored — in n8n’s cloud or on your server;
• who on the team has access to the editor, the execution log and the credentials;
• what personal data passes through the workflow, and whether it all needs to;
• whether the option you’ve chosen meets legal requirements and your internal rules — this is worth discussing with a lawyer.

Check cloud access, payment and licensing terms for your country and company yourself, on n8n’s official site. The choice between the cloud and your own server is often decided by exactly these questions, not by convenience.

AI in n8n: language models and AI agents

An AI agent is a system that uses a large language model to parse a request and decide which actions to take. A regular workflow follows pre-defined steps. An agent chooses on its own which tools to use to answer the request.

In n8n, this is handled by the AI Agent node: depending on the settings, it works as different types of agent. Around it, you assemble:

• a language model — for example, via the Anthropic node, or a local model via Ollama;
• tools — add-ons through which the agent gets data or resources. This can be a call to another n8n workflow, your own code, or an HTTP request to an API;
• memory — a history of previous messages, so the conversation continues instead of starting over each time.

In general terms, a simple agent is put together like this: trigger → AI Agent node → connected model → tools and memory → a result that’s saved or passed on. Detailed setup instructions are in n8n’s documentation.

An important feature for marketing is human approval before an action. In n8n you can require a person’s approval before an agent runs a particular tool. The workflow pauses and waits for a decision: if approved, the tool runs; if declined, the action is cancelled, and the agent receives a message saying it was declined. The approval request can be delivered, for example, via chat, Slack, Telegram or Gmail.

It’s worth using this whenever an agent publishes something, sends something to a client, or changes data. An agent acts within the access it’s been given, and it can make mistakes or invent facts. So give it the minimum access it needs, keep the execution log, and check the result.

The same principle — AI prepares, a human approves — applies to content in general. Social media management, where a person approves every piece of material before it’s published, works this way, for example.

What you can automate in marketing with n8n: examples

Below are workflows built from n8n’s ready-made nodes and HTTP requests. Each one has a step that stays with a human.

Task	What triggers it	What the workflow does	Where a human checks
A summary for an ads and analytics report	Schedule	Requests data from advertising and analytics systems that have an API, and adds rows to a Google Sheet	Checks the numbers and draws conclusions: the workflow collects data but doesn’t explain it
Notification of a new inquiry	Webhook: the site passes the form data	Checks the fields and sends a message to Telegram	A manager contacts the client
Passing an inquiry to a CRM	Webhook from the form	Creates an inquiry in the CRM via its API. For Bitrix24, n8n offers a connection via HTTP Request; amoCRM and 1C — also via API, if your system has one available	The person responsible checks the fields, duplicates and the inquiry’s source
Drafts of posts and descriptions	Schedule or manual run	Passes a task to a language model and saves the draft to a spreadsheet or document	An editor revises the text, checks facts and approves publication
Checking important pages on the site	Schedule	Requests pages from a list and sends an alert on an error	A specialist decides what to fix and in what order
Consolidating SEO data	Schedule	Exports data from analytics services with an API into a shared spreadsheet	A specialist interprets the changes and makes decisions

For Telegram, n8n has a ready-made node: it can send, edit and delete messages. For most message operations, the bot needs to be added to the channel.

If drafts are your main workflow, see how a content factory is built, where AI prepares drafts and an editor checks them. A separate article covers which tasks on a site you can hand to a neural network, and which stay with a specialist only.

Which workflow to take on first is covered in the «Where to start automating marketing» section.

Templates and documentation

n8n’s templates are ready-made workflows from the n8n team and community members that you can import into your own instance. The template library opens from the editor and on n8n’s site, with a filter by category, including «Marketing.» The documentation warns that after importing a template, you may need to set credentials and adjust the settings for your task.

A template saves time at the start, but it’s someone else’s workflow. Before running it:

• figure out what each node does and where the data goes;
• connect your own credentials, with the minimum access needed;
• run the workflow on test data, not on live accounts;
• check whether it has any steps that publish, delete or send something out without your knowledge.

n8n’s documentation is the primary source on nodes, settings and installation. The interface and feature names change, so check it rather than retellings.

Limitations and risks of automation

• Access. A workflow works with your services’ credentials. Whoever has access to n8n also gets access to them.
• Mistakes multiply. If a rule is set up wrong, the workflow repeats the mistake every time it runs.
• External services change. A service’s API, interface or access-rights update can stop a workflow with no visible sign. You need error alerts and regular review of the log.
• Maintenance. Every workflow needs an owner: who updates it, who fixes it, who decides it’s no longer needed.
• AI nodes make mistakes. A model can invent a fact or misread the task. Its result is a draft until a person checks it.
• Personal data. The more data passes through a workflow, the higher the requirements for storage and access.
• Hidden cost. Even the free option takes time to set up, check and maintain, and your own server also takes time to administer.

The same approach applies to SEO tasks: it’s worth automating repetitive operations and leaving decisions to a human. We cover which SEO operations we automate, and where the decision stays with a human, on the service page.

Where to start automating marketing

Automation speeds up a process along with its mistakes. So you describe the process first, then automate it. We follow the same rule in our own work: «Don’t automate chaos. Build the system first.»

1. Pick one frequent manual operation — one that repeats regularly and takes time but doesn’t require decisions.
2. Describe the input, the output, and the check rule. Where the data comes from, what the result should be, and how to tell it’s correct.
3. Choose how to run it — the cloud or your own server, based on the data, access and your team’s skills.
4. Build the workflow on test data and run it manually a few times.
5. Add an error alert and agree on who watches the execution log.
6. Assign someone responsible for the workflow and for checking the result.
7. After a month, assess the result: how much manual work it saved, where mistakes came up, what’s worth fixing. Only then move on to the next process.

Frequently asked questions

Is n8n free?

It depends on the option. For self-hosting there’s a free Community edition; the cloud can be tried during a trial period, then paid plans apply. Free use is limited by the terms of the Sustainable Use License: per n8n’s own explanation, that’s when workflows are created and edited only by people from your organization.

Do you need to know how to code to work in n8n?

For many workflows — no: they’re built from nodes on the editor’s canvas. Code comes in handy for non-standard data processing — there’s a Code node with JavaScript and Python for that. Installing it on your own server requires technical knowledge.

Can you connect a service to n8n if there’s no ready-made node for it?

Yes, if the service has a REST API: the HTTP Request node sends requests to it. If the service can send data on its own when something happens, the Webhook node receives it. Check the service’s own documentation for exactly what’s available via its API.

Can an AI agent in n8n work without human oversight?

Technically, yes: an activated workflow runs on its own. But for actions that can’t be undone — publishing, an email to a client, changing data — it’s better to turn on human approval: n8n can pause a workflow until a decision is made.

Is it safe to give n8n access to live accounts?

n8n encrypts credentials and other sensitive data with a dedicated key. But security also depends on you: who has access to the instance, how well the server is protected for self-hosting. Create separate accounts for your workflows with the minimum access they need.