I published 50 AI-written blog posts. Here's what actually ranked.

Three months ago I set up an AI agent to write blog posts for a client's website. We published 50 posts over 12 weeks, all targeting long-tail keywords in the home renovation space. I wanted to know: does AI content actually rank?

The short answer: some of it did, surprisingly well. Some of it sank. The difference wasn't what I expected.

The setup

The agent identified keywords using search volume and competition data. It wrote first drafts targeting those keywords. I edited every post before publishing — some lightly, some heavily. We published about 4 per week.

I deliberately varied my editing approach to see what mattered. Some posts I barely touched. Some I rewrote the intro and conclusion. Some I added personal anecdotes and specific examples. Some I restructured completely.

The results after 3 months

Of the 50 posts:

• ·12 reached Google's first page for their target keyword
• ·8 more ranked on page 2
• ·15 sat on pages 3-5 with minimal traffic
• ·15 didn't rank meaningfully at all

Overall, 24% hit page one. For a brand new blog with almost no domain authority, that's actually decent. Most SEO agencies would be happy with that rate from human-written content too.

What the winners had in common

The posts that ranked well shared three things.

They answered a specific question directly. Not "everything you need to know about kitchen remodeling" but "how much does it cost to remodel a galley kitchen in a 1960s house." Long-tail, specific, answerable.

They included real numbers. The posts where I added actual cost ranges, timelines, or measurements from the client's projects outperformed vague posts by a wide margin. Google seems to reward specificity. So do AI engines — this aligns with the Princeton GEO research showing statistics boost visibility by 37%.

I edited them enough to add something the AI couldn't. A paragraph about a specific project. A caveat based on experience. An opinion about when not to do something. The posts where I just fixed grammar and published performed worst.

What the losers had in common

Posts that flopped were almost always too generic. The AI would write a perfectly competent overview of a topic, but it read like it could've been on any website. No unique angle, no specific data, no reason for Google to rank it over the 50 other pages saying the same thing.

Some posts failed because the keyword was too competitive. An AI-written post on a new domain isn't going to outrank Home Depot for "best kitchen faucets." Keyword selection mattered more than content quality for those.

A few failed because I published them with minimal editing and they had that flat, slightly over-explained AI tone. Not wrong, just... not interesting enough to keep someone reading.

What Google actually cares about

Google published guidelines saying they evaluate content quality regardless of how it's produced. From what I observed, that's accurate but incomplete.

Google doesn't penalize AI content. But Google does reward content that has something unique to offer. If your AI post says the same thing as everyone else's AI post (because you're all using similar prompts), nobody ranks.

The posts that worked treated the AI draft as raw material, not finished product. Like getting a research brief from an intern — useful starting point, needs your expertise layered on top.

The editing spectrum

I started categorizing my edits:

Level 1 (grammar only): 5% of these ranked. Not worth publishing.

Level 2 (restructure + clarity): 20% ranked. Better, but still generic.

Level 3 (add data + examples): 35% ranked. This is where it started working.

Level 4 (add opinion + unique info + data): 50% ranked. The sweet spot.

The difference between level 2 and level 4 was maybe 15 extra minutes per post. The ROI on those 15 minutes was enormous.

What I'd do differently

I'd skip the broad keywords entirely. Every post should target a long-tail phrase that a real person would type when they're close to making a decision. "Kitchen remodel cost" is too broad. "Cost to add an island to a small kitchen" converts better and is easier to rank for.

I'd add FAQPage schema to every post. Two of my best-performing posts had FAQ sections, and they appeared in Google's featured snippets. That's free visibility.

I'd publish consistently rather than in bursts. Google seems to reward regular publishing cadence. The weeks where we published 4 posts outperformed the weeks where we published 6 then 2.

I'd set up the agent to update old posts, not just write new ones. Refreshing existing content with new data is one of the highest-ROI SEO activities, and it's perfect for automation.

The bottom line

AI content ranks. But "AI content" isn't one thing. A barely-edited AI draft is a different product from an AI draft that's been enriched with real data, specific examples, and human judgment.

The agent handles the 80% that's tedious: research, outlining, first drafts, meta descriptions. You handle the 20% that makes it worth reading.

If you're doing this manually, ClawKit automates the agent side. But the principle works regardless of what tool you use: treat AI as your first-draft machine, not your publishing button.