I Built a Backlink Machine in Claude Code.

Hey Builder 👋

Last week I told you my backlink outreach stood at 12 emails and 0 links. "So… that's next month lol."

Well. Next month came early.

This week Claude Code found the sites, vetted them, drafted the emails and logged every single reply. And let me tell you… the replies did not hold back.

The short version:

Quickly, the rest of my week:

I want to talk directly to the founder who knows backlinks matter, has a half-finished list of sites somewhere, and hasn't sent one email this month.

You're not lazy. Outreach is just 40 tiny boring jobs wearing a trench coat. Find sites, check them, find emails, write, send, follow up, answer, track, and then check the link actually went up.

So I didn't do it. I built it. In Claude Code, with a human (me) only where a human actually matters.

Quick disclaimer: I'm not an SEO agency. I'm one person building three products with AI. The email template came from Prafull at Outsmartt, an SEO agency I partner with. Everything around it is Claude Code.

Here's the whole machine below 👇

The logic is simple: a site that already linked to a resume builder will link to another one.

I gave Claude Code the Ahrefs MCP and 11 competitors (Canva, Enhancv, FlowCV, Kickresume, Novoresume, Resume.com, Resume Genius, Resume.io, Rezi, Teal and Zety). For each one it pulls every backlink, with one filter:

Ahrefs hands back at most 100 rows per call. So big pulls go to small helper agents (Haiku, low effort) whose only job is writing the CSV word for word. 955 raw rows, one merge later: 502 sites.

(For Switzerland I pointed it at 8 .ch competitors and got 189 more.)

Steal this: "Pull my 10 biggest competitors' backlinks, DR 10-40, dofollow, in-content, one per domain" is one sentence to an agent with the Ahrefs MCP.

Here's the part I like most. And it isn't even AI.

The vetting is plain rules in a script, so it gives the same answer every time:

Claude Code pulls the Ahrefs facts into a file. The script makes the call. No model decides who's "worth it".

The receipts: when the first 31 replies came in, I back-tested the rules on all of them. They passed the 5 sites I actually kept talking to, and failed 21, including all 17 sellers I'd turned down.

So now the vet runs before anything gets drafted. Sites that fail never get a follow-up.

Steal this: let the AI gather the facts and let code make the yes/no call. You can test code. You can't test vibes.

Before spending a cent, the script reads each site's contact, about, write-for-us and Impressum pages.

Only after that does it offer Mindcase, a paid email finder, and only with a cost estimate first and my yes.

Steal this: make your agent show you the bill before it spends. One line in the skill does it: no paid call without an estimate and a yes.

The email itself is Prafull's template, word for word. A script fills in the name and company. No AI rewriting: same email, every time. The heart of it is one question:

On Oct 19 Claude Code put 168 drafts in my Gmail. Then I sent them. Myself.

That's on purpose. In my Claude Code settings, Gmail's send, reply and forward always ask first. The agent can draft all day. Only I hit send.

Quick math break:

Every reply gets logged to one file: what they said, the price, a verdict (accept, counter, decline or wait) and the next step.

And the answer comes from a table, not from my mood:

The rule is to answer every reply the same day, and the agent only ever writes drafts.

Every Monday at 08:30 a routine emails me what's pending: drafts left, emails sent, replies, follow-ups due. I answer with one sentence ("log the outreach sends and draft the follow-ups") and it does the rest. I never type a command. That was the whole point.

A few of the quotes, for fun: $25, $35, $50, $60, $100, $150, $159, $200, $300, $500, $350 a year, and my favourite, "$2,500/post, or 25 for $500 upfront (PayPal)" 🙃 I paid none of them. Two of those sellers, Corp to Corp ($150) and Learn Architecture ($160 a post), ended up as free swaps.

A script opens every partner page and checks the link is really there: the right page, the right anchor, no nofollow, sponsored or ugc tag, and the page isn't noindex. Anything that fails flips to "missing".

Why it exists: on Oct 24 a reply went out saying our side of the Corp to Corp swap was live. It wasn't. Oops. It went live on Oct 26, and now the skill has a rule: never draft a claim before the checker passes, in the same session.

Steal this: if your agent writes "it's live" in any email, make it prove it first.

8 links to Careerkit are live: 5 from Learn Architecture (one company, five architecture sites), 2 from SearchOpal and 1 from Corp to Corp. Every one is a trade, and I link back to them from my blog. Two more partners have agreed and aren't live yet.

Nobody replied "great site, here's a link". The pool of sites that already link to your competitors is mostly sellers and swappers, which is exactly why the vet and the price table exist.

One caveat I take seriously: Google's spam policies call out excessive link exchanges. That's why the rule in my skill is one natural link from a relevant post, never site-wide.

The lazy version, this week:

If you take one thing away: let the AI do the 40 boring jobs, let code make the yes/no calls, and keep the send button for yourself.

Okay, that's it from me. Reply with your question, I answer one every week. And if you've ever been quoted for a "guest post", tell me the highest number you've seen 👀

Keep shipping,

Nishant