Why we built Lyra: from zero to millions of blog impressions

We did not start with a content strategy. We started with a product nobody could find.

Our search presence was a flat line near zero. We were shipping features and posting them to a feed that our existing users saw and nobody else did. The product was good. The distribution was missing. Every founder eventually hits this wall: the thing you built is real, and the internet does not know it exists.

So we did the unglamorous thing. We started writing.

The boring channel that actually compounded

We tried the usual channels first. Paid acquisition worked until we stopped paying. Social got us a spike and a flat week after. Launches got us a day. None of it compounded, which is the only property that matters when you are small and cannot outspend anyone.

Blog content was different. A single post that answered a real question kept pulling traffic months after we wrote it. We would forget a post existed, check the analytics, and find it quietly bringing in readers every day. That is the part nobody tells you about SEO: the work is front-loaded and the payoff is a annuity. You write once and it pays out for years.

The trick was to stop writing for ourselves and start writing for the search bar. Not keyword stuffing, the opposite. We watched what developers actually typed when they were stuck, then wrote the post we wished existed when we had the same problem. Specific titles. Real numbers. Code that runs. No throat-clearing intros.

One post a week. Then two. The line that had been flat for a year started to bend.

What "millions of impressions" actually looked like

It did not arrive as a hockey stick. It arrived as a slope.

The first three months felt like shouting into a well. Google does not rank a new domain quickly, and patience is the entire game. Around month four, a few posts cracked the first page for long-tail queries, the kind with low competition that bigger players ignore because the volume looks small. Small volume, stacked dozens of times, is not small.

By the time we looked up, the blog was doing the heavy lifting for the whole product. Millions of monthly search impressions, most of it from posts targeting questions our competitors never bothered to answer. The compounding had kicked in: new posts ranked faster because the domain had earned trust, and old posts kept climbing as they accumulated internal links.

If you want the playbook we used, we wrote it down in our guide to SEO for SaaS. The short version: pick winnable keywords, publish consistently, link everything together, and never ship a claim you have not checked.

The part that broke: doing it by hand

Here is the problem with a content engine that works. It demands to be fed.

Every week was the same loop. Research what to write. Check it against what we had already published so we did not compete with ourselves. Write the draft. Verify every statistic and every link, because one broken link or stale number quietly erodes the trust you spent months building. Add the internal links to the right pages. Generate a banner. Open the pull request. Review it. Merge.

It worked, and it was relentless. The research was mechanical. The fact-checking was mechanical. The interlinking followed rules we could write down. The only genuinely creative part was the angle, and even that came from data more than inspiration. We were spending senior time on a process that was ninety percent rules.

That is the moment a founder either hires a content team or builds a machine. We build machines.

Lyra is the machine

Lyra is the loop we ran by hand, handed to an agent that does not get tired.

She watches for topics worth ranking for in your focus area and dedupes them against what you have already published, so she never cannibalizes your own rankings. She reads your repo on connect and learns where posts live, your frontmatter, your slug rules, and how you sound, then writes in that voice. She fetches and confirms every external link or drops it. She checks claims against current sources. She scores each draft out of ten across content, SEO, technical accuracy, readability, and linking, leaves herself review comments, and rewrites until it clears the bar.

Then she opens a pull request, labeled and ready, with a banner generated to match your blog. She tags you. Nothing merges until you hit the button. That last part is not a limitation, it is the point. You keep editorial control and lose the grind. If your blog lives in a Git repo, that is the whole case for an AI blog writer for developers.

This very post is the kind of thing Lyra ships on her own. The structure, the interlinking, the fact-check pass, the banner: that is her loop, not a human one.

What we learned, compressed

If you are where we were a year ago, with a real product and no search presence, here is what we would tell you:

• Pick the boring channel that compounds. For us it was blog content. Write the post you wished existed.
• Target the questions bigger players ignore. Low competition stacked dozens of times beats one impossible head term. We go deep on this in internal linking automation and answer engine optimization.
• Verify everything. A blog grows on trust, and trust dies one broken link at a time.
• Be consistent for longer than feels reasonable. The slope starts before the spike.
• Then automate the parts that are rules, so you spend your time on judgment instead of mechanics.

We grew a product from invisible to millions of impressions on the back of a blog. The strategy was simple and the execution was a grind. Lyra is what happens when you refuse to keep doing the grind by hand.

We built Lyra because we ran this exact playbook ourselves and got tired of running it manually. If you have a blog in a GitHub repo and an Anthropic key, she can run it for you, from topic discovery to a pull request you merge.

Talk to the founder → · Join the waitlist