Every link building tool publishes a case study eventually. Most of them are unverifiable: screenshots of a rankings graph, "we built 40 links for a client we can't name", trust us.

We can do better, because the customer is us. On July 5 we pointed the BacklinkGPT agent at viral.app, our own creator-marketing product, connected one Gmail inbox, and let it run. This post is the unedited log of the first 14 days — what the agent judged, what it wrote, what humans rejected, what a prospect replied (summarized — we quote our own emails verbatim, not other people's), and what it cost.

The scoreboard after 14 days

Two numbers deserve the spotlight. 7.8% keep rate: the agent threw away 92% of what it found — junk listicles, competitor-adjacent blogs, unreachable giants — before writing a single word. And 19 rejected drafts: we reviewed everything before it sent, and we said no a lot. The agent learns from every rejection; the point of this post is not that it's flawless, it's that the whole loop is visible.

The link: how a DR 50 placement closed in 3 days

The agent found an analytics article on fontly.io (DR 50, ~6,000 monthly organic visits), judged it a strong match for viral.app's audience, and drafted this. We approved it, and it went out on July 13 at 05:14 UTC, signed by Mike from our team, whose inbox the agent works from:

Fourteen minutes later their editorial contact replied with their sponsorship terms: $80 for a permanent placement. The agent handled the negotiation from there — again, every message below waited for our approval before sending:

They came back with submission terms: a new article, one client link, 24–72h turnaround, PayPal. The agent confirmed the details and flagged one thing a template blast never would:

From there, honesty requires saying: humans did two things. We wrote and delivered the 1,600-word guest article (the agent requested the invoice and coordinated the handoff), and when the editor couldn't open the Google Doc, Felix jumped into the thread to fix the sharing settings. On July 16 at 10:45 the article went live with the link — 3 days and 5 hours after the first cold email. We paid the $80 invoice ten minutes later.

1. 1

   Jul 13, 05:14 — cold email sent

   The agent's first touch on the fontly.io analytics post. Reply with sponsorship terms arrives 14 minutes later.

2. 2

   Jul 14 — deal accepted

   The agent confirms the $80 placement and asks whether to supply a new article or extend the existing post.

3. 3

   Jul 15 — terms settled, article delivered

   Submission guidelines confirmed, invoice requested, and the 1,600-word guest article (written by us) handed over.

4. 4

   Jul 16, 10:45 — link live, invoice paid

   The article publishes with the link. We pay the $80 ten minutes later. First touch → live: 3 days, 5 hours.

Is a paid placement a "worse" link than an earned one? It's a different link. The agent's job was to find a relevant DR 50 site, open the conversation, and negotiate the terms — the same motion an agency retainer buys you for a few thousand a month. The fee went to the publisher, disclosed here because case studies that hide the money are marketing, not evidence.

The negotiation still running

The second conversation is with a round-up site in the UGC space. The agent pitched a listicle inclusion, got a $500 sponsored-listing offer back within a day, and countered — questioning the price against the site's authority metrics and proposing a partial link exchange instead. That thread is still open. We're sharing the pattern rather than the emails because the negotiation is live and the counterparty didn't sign up to be in a blog post.

The takeaway isn't the outcome yet — it's that the agent didn't fold at the first price quote, and didn't ghost either. It negotiates with context: it knows what we've paid for comparable placements.

What the agent got wrong

The honest column. Out of 63 drafts written in these 14 days, we rejected 19 in review — wrong angle, too pushy, or a page we didn't want our name on. Two emails bounced on dead addresses. Nine LinkedIn connection requests went out; zero accepted so far — LinkedIn is a slower channel and two weeks is nothing. One early pitch referenced a tool mention that merited a fact-check before sending; review caught it.

That's the actual product thesis: the agent does the hunting, judging, writing, and chasing — the parts that eat your week — and you keep a veto on every send. The rejections aren't waste, either; they feed the agent's guidance so the next batch drifts toward what you'd approve.

What about the links from before the agent?

viral.app's backlink portfolio also holds five older links (DR 28–78) that Felix won manually over the preceding two months — guest-post packages, a link exchange, one earned editorial mention — for about $200 in placement fees and a lot of founder hours. They're tracked in the same portfolio view, marked with how each was acquired.

That manual era is the comparison that matters. Manually: five links in roughly eight weeks, every one bought with an evening of prospecting, writing, and follow-up. With the agent: the first qualified conversation happened within 72 hours of connecting the inbox, and the first link was live nine days later — while the humans involved spent their time approving drafts and writing one article.

Run the same experiment on your site

The agent that did this is the product: BacklinkGPT reads your site, hunts prospect pages every night, judges them like a link builder, finds the right person, and drafts outreach that waits for your approval — 7-day free trial, card required, nothing sends without your OK.

Not ready for that? Run the free opportunity scan: it does a miniature version of exactly what you just read — real SERP hunting, real page verdicts, and the first pitch it would send for your site — and emails you the results.