← Back to the Lab
Complete Experiment·Last updated: April 1, 2026
Can AI content rank on Google?
I tested how AI-assisted content performs in search when it’s edited, structured, and fact-checked like a real editorial workflow.
The problem I’m testing
AI makes it easy to publish a lot of content—but most of it reads the same, and there’s real concern about whether Google will trust or surface it.
Hypothesis
If I treat AI as a drafting assistant—not the final writer—and combine it with strong outlines, editorial judgment, and real examples, I can ship content that both ranks and builds trust.
How I’m running it
- Picked a fresh domain in a niche with real but reasonable competition.
- Defined topics and outlines first, based on search demand and business-relevant queries.
- Used AI to draft sections, then heavily edited for clarity, specificity, and brand tone.
- Added unique data points, examples, and internal links that AI couldn’t invent on its own.
- Tracked indexing, rankings, and click-through over several months.
Data sources
- Google Search Console (impressions, clicks, queries, pages)
- Rank tracking (optional)
- Index status checks (site: queries / URL inspection)
What I’m seeing
- Pages that combined AI drafts with strong editing and unique examples earned impressions and rankings similar to human-written baselines.
- Thin, unedited AI drafts underperformed—even when the outline was solid—confirming that “just paste from the model” is not a strategy.
- Longer-form, structured guides with clear subheads and internal links performed best, regardless of whether AI was involved.
What it means
- AI can speed up production, but you still need a clear point of view, strong outlines, and a human editor.
- Google is rewarding usefulness and clarity, not whether a model touched your draft.
- For most brands, the win is using AI to go from blank page to solid draft faster—then investing editing time where it matters most.
What I'd do differently
- Capture a baseline set of queries/pages before publishing any AI-assisted content.
- Ship fewer pages earlier, then iterate based on GSC query data instead of guessing.