We Let Sprinkle Pick Its Own Topics for a Month. Here's What Happened

For 30 days, nobody on our team suggested a single blog topic. We turned on Sprinkle’s AI Content Plan for this site, pointed it at our own keyword data, and let it decide what to write about, when to publish, and in what order — the same automated pipeline every customer’s account runs on.

The setup

Sprinkle’s Content Plan works from three inputs: live keyword opportunity data, gaps between our content and competitors’, and performance feedback from posts already published. Every few days it proposes the next batch of topics ranked by expected impact. We didn’t override a single one for the full month.

What it picked, and why

The topic mix skewed more technical than we expected — audits, backlink mechanics, comparison content — rather than the broader “SEO tips” posts a human editor might default to. The system was optimizing for keyword gaps where we had zero coverage and a competitor was ranking, not for topics that felt intuitively “on brand.”

In hindsight, that’s the more defensible strategy. A human topic list tends to repeat whatever the last few winning posts were about. The algorithm doesn’t have that bias — it just looks at where the gaps are.

The results

• 9 posts published on schedule, zero missed days
• 3 of the 9 ranked in the top 10 for their primary keyword within three weeks — faster than our historical average of five to six weeks
• The lowest performer was a comparison post that cannibalized traffic from an existing page rather than adding net-new visibility, which the system flagged itself in week three and stopped producing similar topics

What surprised us

The system caught its own mistake faster than a human editor typically would. Once the cannibalization signal showed up in the traffic data, the next topic batch shifted away from that content type without anyone telling it to. That feedback loop — publish, measure, adjust the next plan — is the part that’s hard to replicate with a manual content calendar, because it requires someone checking performance data daily, not monthly.

We’re keeping the AI Content Plan running with the same “no override” policy on this blog going forward. If it starts drifting, that’s useful data for the product team, too.