I run a small content experiment on a test blog I've maintained since early 2025. I published 20 posts in three batches: purely AI-generated with no editing, AI-generated with a thorough human editing pass, and fully human-written. Same topic categories, similar target keyword difficulty, the same site. I tracked rankings and organic traffic over 6 months. Here's what I found.
The test setup
Twenty posts across three categories: informational ("what is [concept]"), commercial intent ("best [tool] for [use case]"), and local/specific ("how to do [task] in [city] in 2026"). Each category had a mix of AI-only, AI+editing, and human-written posts. I was careful to ensure topic difficulty was roughly comparable across groups — I wasn't comparing a low-competition keyword in one group to a high-competition keyword in another.
The AI-only posts: generated with Claude Sonnet, proofread for obvious errors, no structural editing, published as-is. The editing pass for AI+editing posts took 30–45 minutes per post and included: adding at least one specific example from real client experience, reworking the introduction to have a sharper hook, removing formulaic phrases and filler, adding real pricing or data in India-specific context, and adjusting the heading structure to reflect how I'd actually organise the information rather than the AI's default.
What the data showed after 6 months
AI-only posts: 3 of the 7 posts ranked on page 1 within 3 months. 4 remained on page 2–3. The page 1 posts were targeting informational queries with lower competition. None of the AI-only posts that targeted more competitive commercial queries reached page 1. Average CTR across AI-only posts: 2.1%.
AI+editing posts: 6 of the 7 posts ranked on page 1 within 4 months. 1 sat on page 2–3 for the most competitive keyword in the batch. The edited posts consistently outranked the AI-only posts targeting similar keywords. Average CTR: 4.3%.
Human-written posts: 5 of the 6 posts ranked on page 1. The CTR was 5.1% — slightly higher than the edited AI posts, which I attribute partly to the slightly better headline writing in human posts (a subjective quality I noticed when reviewing them) and partly to variance.
The difference between AI-only and AI+editing is more significant than the difference between AI+editing and human-written. The editing pass matters enormously. The AI origin, by itself, is not the determining factor.
The patterns in what ranked and what didn't
Posts that included specific India-context details — ₹ pricing, named cities and areas, Indian tool alternatives, Indian regulatory context where relevant — outperformed posts without that specificity regardless of writing method. Specific beats generic in Indian SERPs particularly because the competition is often thin and non-India-specific.
Posts with original structured data (a comparison table I built, a checklist from my own process) performed better than posts that were only prose. The structured content formats get featured snippet placement at a higher rate and increase time on page.
The informational posts declined in performance over the 6-month period — consistent with the AI Overviews impact I described in a separate post. The commercial and specific posts held their rankings better.
My current approach to AI-assisted content
I use AI to generate a detailed first draft. The draft is genuinely useful as a structure and saves me research time significantly. Then I do a thorough editing pass that takes 40–60% of the time the full post would have taken to write from scratch. Net result: I produce content approximately 2–3x faster than purely human-written, and the quality of the edited AI post is comparable to or better than a humans-written post I'd have produced under time pressure.
What I don't do: publish AI drafts without meaningful editing and call it a content strategy. The people doing that are mostly producing content that doesn't rank, doesn't convert, and gives AI-assisted content a worse reputation than it deserves.
The better frame: AI is a capable first-draft writer. The editor who knows the subject, knows the audience, and has real experience to draw on is still essential for content that performs. The two together are faster and often better than either alone.
What the editing pass actually looks like — my specific process
Since I've been claiming the editing pass makes the critical difference, let me be specific about what it contains. Vague advice like "edit your AI content" is useless. Here's the actual checklist I run:
First pass (10 minutes): read the full draft and mark every sentence that sounds like an AI wrote it. Typically these are: sentences starting with a hedge opener or AI throat-clearing phrase, any sentence using AI-signal vocabulary (the kind of words AI writing guides tell you to stop using), any opening paragraph that restates the article title in sentence form, any closer that wraps everything up with a summary paragraph. Delete or rewrite every marked sentence before doing anything else.
Second pass (15 minutes): add specificity. Every place the AI writes "many businesses" or "various strategies" or "significant results" — replace with actual numbers, actual business types, actual results. In an Indian context, this means ₹ figures, named cities, named tools, named platforms. "Many businesses in India use WhatsApp for customer service" becomes "A textile export business in Surat I worked with was handling 300+ customer enquiries per day through WhatsApp before they automated the first-response layer with AiSensy."
Third pass (10 minutes): add one genuine opinion or counterintuitive take that I actually hold. This is the part AI cannot fabricate. Something I disagree with in the conventional advice, something I think is overcomplicated, something I think works differently than most articles claim. This section is where the content becomes mine rather than a polished version of what everyone else publishes.
Fourth pass (5 minutes): check the structure. AI default structure is: intro, 5 equal-length h2 sections, conclusion. Sometimes that's right. Often it's not. Restructure so the most important section comes earlier, related sections flow naturally into each other, and there's a sense of escalating specificity rather than five roughly equal chapters with arbitrary headings.
Total editing time per post: 35–45 minutes. That's the investment that turns a 2-minute AI draft into a post that actually ranks and converts. Anyone skipping this step and wondering why their AI content doesn't perform — the answer is usually the skipped editing pass.
When AI content consistently underperforms: the actual patterns
From my cross-site testing, the conditions where AI content almost never competes effectively:
Highly contested branded queries where competitors are established businesses with years of domain authority and real customer reviews. AI-generated content about "best CRM software in India" doesn't rank near Capterra or G2 reviews. No editing pass fixes that; the competitive landscape has different inputs than content quality.
Content that requires visual evidence. "How to set up Google Analytics 4 in 2026" — users want screenshots of the actual interface. AI content without original screenshots will lose to a video tutorial or a competitor who photographed the exact UI steps. Text alone is insufficient for process-documentation queries.
Queries where personal authority matters for conversion even if they rank. "Should I quit my job and freelance in web design" — an AI-generated answer might rank for this. But the conversion from reader to enquiry only happens when there's a real person behind the content. If I'm publishing an AI draft under my name without genuine personal input, the readers who reach the contact form can tell. The credibility gap is real and it shows in conversion rates even where rankings look fine.
Related: How AI Overviews are changing the organic traffic game and How to automate your WordPress content workflow.