Five n8n workflows that do the judgement work, not just the plumbing.
Import a JSON file, connect your own API keys, and you have a working automation that reads a call transcript, a support ticket, a resume, a keyword or a research question and writes a structured result somewhere useful. Built for operators who already run n8n and are tired of templates that are three nodes and a prayer.
84 nodes across five workflows. Every one imports into a clean n8n instance, verified twice. No hardcoded credentials. Current model IDs. No workflow has been run end to end against live APIs, and that is stated plainly further down this page.
Who this is for
You already have n8n running.
Self-hosted or cloud. You have an Anthropic key. You have looked at the free template gallery and found a lot of "send a Slack message when a form is submitted" and almost nothing that makes a decision and explains it.
These five are the second kind. Each one takes unstructured input, reasons over it, branches on what it found, and writes a result with the reasoning attached.
If you do not run n8n yet, self-hosted n8n is free and n8n cloud starts around $20 a month. That is a separate decision from this one, and you should make it first.
The five workflows
Each one buys separately. Each one works alone.
1 · $69.99 · 16 nodes
Meeting Notes to CRM
The problem it removes: The rep finishes a call, the CRM stays empty for three days, and by the time anyone writes it up the objection they heard is gone.
Feed it a call transcript. It pulls out attendees, deal stage, action items with owners, objections raised, competitors mentioned and budget signals. Then it writes a HubSpot contact, a note on the record, and one task per action item.
The detail that matters: only seller-side action items become tasks. If the prospect said they would loop in their CFO, that stays in the note. It does not land on your rep's task list as something they have to chase themselves. Most transcript-to-CRM builds get this wrong and the rep abandons the automation in a week because their task list is full of other people's promises.
This one also ships with its own validator script, 42 checks, that you can run yourself before you trust it.
2 · $69.99 · 18 nodes
Support Ticket Triage
The problem it removes: Every ticket gets the same queue and the same wait, so the angry one about a failed payment sits behind a password reset.
It classifies each inbound ticket on four axes: category, urgency, sentiment, and its own confidence in that call. Then it branches.
High confidence plus low urgency, it auto-replies. Medium confidence, it drafts a reply and holds it for a human to approve. Low confidence or high urgency, it escalates to Slack with the reasoning attached so whoever picks it up knows why the machine backed off.
Every decision gets logged, including the confidence score. That is the part you will actually use, because after two weeks of logs you can see where your thresholds are wrong and move them with evidence instead of vibes.
3 · $69.99 · 19 nodes
Resume Screener
The problem it removes: 200 applicants, one hiring manager, and a shortlist that gets made by whoever's resume happened to be on top.
It scores resumes against a job description and writes out its reasoning per factor, so you can read why it scored something a 6 and disagree with it. Strong matches route to the recruiter. Maybes go in a queue. Everything else logs to Google Sheets so nobody disappears.
It is instructed to ignore name, gender, age, nationality and school. That is a design choice in the prompt, not a certification of anything.
Read this before you buy this one. Automated hiring tools are regulated. NYC Local Law 144 covers automated employment decision tools, and the EU AI Act classifies employment screening as high risk. This is a screening aid that requires human review of every outcome, including the rejections. Compliance with the law in your jurisdiction is yours, not mine. If you were hoping to point this at an inbox and let it reject people unattended, do not buy it, because that is not what it is and using it that way could put you on the wrong side of a statute.
4 · $69.99 · 15 nodes
SEO Content Pipeline
The problem it removes: The gap between "we should write about this keyword" and a draft somebody can actually edit.
Give it a keyword. It fetches what currently ranks, builds an outline, names the gap between the outline and what is already ranking, writes a draft, and produces a title under 60 characters plus a meta description. Output lands in Google Docs.
If you have not set a search API key, it does not quietly make up competitors. It drops into a clearly labelled "no live research" mode so you know the gap analysis is missing and the draft is working from the model's own knowledge. A pipeline that invents its sources is worse than no pipeline, because you cannot tell which drafts are poisoned.
This produces a draft and a set of decisions. It does not produce rankings. Nothing produces rankings.
5 · $69.99 · 16 nodes
AI Research Brief
The problem it removes: The "quick research" ask that eats a Tuesday, and the AI answer you cannot use because you have no idea where any of it came from.
Ask it a question. It searches, fetches and reads the top sources, then writes a brief where every finding carries its source URL and a verbatim quote from that source. It calls out where sources disagree with each other instead of averaging them into a confident mush, and it states what it could not determine.
Then it validates its own citations against the URLs it actually fetched. Anything it cannot verify gets quarantined rather than shipped in the brief. This is the workflow I would hand to someone who has stopped trusting AI research output, because the failure mode it is built around is exactly the one that broke their trust.
What you actually get
For each workflow you buy
The workflow JSON. Import it into n8n. That is the whole install step.
A setup guide, roughly 3,000 words. Not a README. Prerequisites listed up front so you find out about a missing key before you start, numbered setup steps, one real sample input with the exact output it produced, the five likeliest ways it breaks with the fix for each, and an honest section on what it does not do.
Credential references, not credentials. Every node points at an n8n credential you create. There is no key of mine, and no placeholder key of yours, baked into any file.
Current model IDs. All five call claude-sonnet-5. Not a 2024 model ID that gets deprecated out from under you three weeks after purchase, which is the single most common thing wrong with workflow templates you find for free.
The meeting-notes workflow also ships with a validator script, 42 checks, that you run against the file to confirm the structure is intact before you wire it to your CRM.
What I verified before putting this up
All five import successfully into a clean n8n instance. I ran that import myself, twice.
84 nodes total across the five. Every connection resolves to a real node. No orphans, no dead branches. Checked with a script, not by eye, because eyes miss things in a 19-node graph.
No hardcoded credentials anywhere in any file.
All five reference claude-sonnet-5.
Tested on n8n 1.109.1.
What this is not
I would rather lose the sale than have you find these out after you pay.
These have not been run end to end against live APIs. Structure and importability are verified. Runtime behaviour is not. I have confirmed that every workflow imports cleanly, that every node connection resolves, and that no node references a dead model or a baked-in key. I have not confirmed that a HubSpot task lands correctly in your HubSpot instance, because I have not run these against live production APIs. You will be the first person to run yours against your accounts. Budget an hour of wiring and testing per workflow, and run each one on sample data before you point it at anything real.
You bring your own API keys and pay your own per-run cost. Anthropic for all five. Then HubSpot, Google, Slack or a search API depending on which workflow. Nothing here is metered by me and nothing routes through me, which also means nothing is subsidised by me. The per-run cost is yours.
These are workflows to run, not a service I host. There is no dashboard, no login, no uptime for me to be responsible for. You run them on your n8n. Self-hosted n8n is free, n8n cloud is about $20 a month.
Tested on n8n 1.109.1 only. Version 2.x was not tested. If you are on 2.x, assume some node parameters will need adjusting.
AI output needs a human between it and anything that matters. Before a reply goes to a customer, before a resume decision affects a person, before a draft gets published with your name on it. The workflows are built with approval branches and logged reasoning specifically so a human can sit in that gap. Take that gap out and you own whatever comes through it.
All sales are final. That is why this section is this long and this specific. Read it, decide, and buy only if it still sounds like what you want.
The bundle
All five workflows, $270.
Separately they are $349.95.
Buy the bundle if you want the set: five guides, 84 nodes, and five patterns you can read and steal from for whatever you build next. Honestly, the guides are where most of the reusable value sits. Once you have seen how the confidence-branching in the triage workflow is wired, or how the research brief validates its own citations against fetched URLs, you can rebuild that pattern in anything.
Buy a single workflow if exactly one of these is a problem you have this week. There is no penalty for coming back for the rest later, and no expiring discount pushing you to decide today.
Yes. Enough to import a workflow, create a credential and read an execution log. If you have never opened n8n, start with the free self-hosted version and a simple workflow of your own first. These are not a tutorial.
Do I need to code?
No, but you will be editing prompt text and node parameters, and reading JSON when something does not line up. If that is uncomfortable, this line is not for you yet.
What will it cost me to run?
Depends on your volume and your model usage, and I am not going to guess at your numbers. All five call Claude, so your Anthropic usage is the main variable. Run one on ten real inputs and read your own bill. That is the only estimate worth having.
Which one should I buy first?
Whichever maps to the thing that is annoying you right now. If you genuinely cannot pick, the Research Brief is the one with the fewest external dependencies to wire up, so it gets you to a working run fastest.
Will these work on n8n 2.x?
Untested. Built and verified on 1.109.1.
Can I modify them?
Yes. They are JSON in your instance. Change prompts, swap nodes, rip out the branch you do not want. That is the point of buying a workflow rather than a subscription.
Can I resell them?
No. Personal or single-company use.
What if a workflow does not do what I need?
All sales are final, which is why every honest limit is on this page above the buy button rather than in a footnote below it. If the "what this is not" section describes a dealbreaker, that is the section doing its job.
Is the resume screener legal to use?
That depends entirely on your jurisdiction and how you use it. It is a screening aid built to require human review. NYC Local Law 144 and the EU AI Act both apply to automated employment tools, and there may be more that applies where you are. Compliance is your responsibility. If you do not have an answer for how you would handle a bias audit or a candidate notice requirement, get that answer before you use this on real applicants.