# AI-Written vs Human-Written Content: What Otterly’s Experiment Found

> Source: https://go-seo.com/ai-vs-human-content-ai-seo/  
> Last updated: November 18, 2026

**Yes, AI-written content can beat human-written content in AI search and in Google, but only when a careful multi-step process is behind it, not a one-click tool.** In a new experiment from Otterly, a 100% AI-written site earned 70% of the two sites’ AI citations and 65% of their Google clicks.

Otterly published the full results this week in [AI vs Human Written Content](https://otterly.ai/blog/geo-experiment-ai-vs-human-writing/) by [Rick Tousseyn](https://otterly.ai/blog/author/rick/), and I attended their webinar walking through it. Here is what they found, what “100% AI-written” really meant, and a phrase from another SEO podcast that explains what to do about it.

Key takeaways

- **The test:** two new sites, the same 100 sloth questions, one written by humans and one by AI.

- **The result:** the AI-written site got 70% of AI citations (2.3 times as many) and 65% of Google clicks.

- **The catch:** “AI-written” meant a 10-step workflow with research, company data, editing, and fact-checking, not one click.

- **Commodity vs non-commodity:** pure AI output is a commodity. AI plus your own stories and experience is not.

- **The name:** more people now say “AI SEO” than GEO. It’s the same work.

## What did Otterly actually test?

Otterly published the same 100 sloth articles on two brand-new websites, one article per question. Their copywriters wrote one site without AI. The other was written entirely by AI using a workflow built by Hans van Gent, Head of SEO at [Seeders](https://seeders.com/).

They tracked four sources over about five months: Google Search Console, Bing Webmaster Tools, Otterly’s citation data from 100 prompts across seven AI engines, and AI crawler visits from server logs.

## What were the results?

The AI-written site won in AI search and in Google. The human-written site won in Bing and in brand mentions.

- **AI citations:** the AI site took 70% of citations and led on all seven engines. ChatGPT never cited the human site.

- **Breadth:** 90 of the AI site’s 100 articles were cited at least once, against 69 for the human site.

- **Google:** the AI site earned 65% of clicks and impressions, with an average position of 9.5 against 12.9.

- **Bing:** the human site led, with 85% of clicks.

- **Brand mentions:** AI answers named the human site 62% more often.

- **Crawling:** AI bots visited both sites about equally. The AI site had only 5% fewer visits.

Keep the scale in mind. Click-through rates stayed under 0.5% on both sites, and together they earned just 0.6% of all citations on those prompts. These are small, new sites.

## Does “100% AI-written” mean one-click AI content?

No, and this is the most important detail. Every AI article went through a 10-step workflow that took about 40 to 45 minutes per article. Hans has spent two years building it.

- Pick one of 200+ content templates.

- Research the Google results and build an outline.

- Add statistics and internal company data.

- Plan each section.

- Write, with internal and external links.

- Run an AI editorial review.

- Rework sections so AI engines can extract them.

- Fact-check and send corrections back.

- Rewrite the intro, outro, and calls to action.

- Create images.

Compare that with Otterly’s earlier 2,000 AI Blogs experiment. Two sites published about 1,000 AI posts each in under a day, and Google deindexed both without warning. The pipeline here published nearly all 100 articles over 17 days.

## Was this a fair fight?

Not entirely, and Otterly says so. The human articles were basic and general, with no expert depth, while the AI site got a much heavier process. So the result can’t be pinned on authorship alone.

Other limits: it was one low-competition niche, the sites are small, and one experiment can’t prove a cause. Otterly is now testing three more sites in different niches. The fair takeaway is that bad content loses whoever writes it, and human authorship by itself is no advantage.

A simple way to read the result is through three rules of thumb for getting a post to rank or get cited:

- **It delivers on the promise.** The page does what the search result or question implied, with no bait and switch.

- **It’s a good experience.** It loads fast, reads well, and the content is solid.

- **It has a good reputation.** Other sites link to it and mention it.

Otterly’s test mostly compared the first two. Both sites were new and small, so neither had much reputation, and together they earned just 0.6% of citations. The Sloth Conservation Foundation alone took 8.5%, and Wikipedia took 5.3%. Reputation is its own lever, and no writing process replaces it.

## What are commodity and non-commodity content?

Commodity content is written entirely by AI, so anyone with the same prompt can produce the same thing. Non-commodity content is AI-assisted work with a real human layer added, such as a personal anecdote, a client story, or a hands-on experience. Another SEO podcast coined these two terms, and I’m adopting them.

Otterly didn’t test an anecdote layer, so this part is my reading of their data. But one result points that way: the human site was named in 62% more AI answers even though it was cited less. Otterly’s own advice is to use AI for coverage and structure and keep humans in the loop for brand.

## How do I turn commodity AI content into non-commodity content?

Use a process for the draft, then add what only you can add. Here is what I’d do:

- **Skip the single prompt.** Use steps: research, outline, draft, edit, fact-check.

- **Add a story.** A wedding photographer can describe a real venue. A contractor can describe a job that went sideways. That is the “experience” in [E-E-A-T](https://go-seo.com/what-is-e-e-a-t-in-seo/).

- **Add a client result,** with permission.

- **Add proof:** photos, prices, and numbers you can stand behind.

- **Match the question.** Otterly’s AI site used short, question-shaped URLs (a median of 5 words versus 9.5), a habit that fits [question headings for AI search](https://go-seo.com/question-headings-for-ai-search/).

- **Make sections easy to lift.** See my guide to [content chunking for AI search](https://go-seo.com/content-chunking-for-ai-search-the-need-to-know-seo-strategy/).

- **Fact-check everything,** and publish at a steady pace.

A good first move is your five most important pages. Add one real story to each, then check that AI bots can reach them with the free [AI Visibility Checker](https://go-seo.com/tools/ai-bot-checker/). Being crawled is not the same as being cited: ChatGPT’s fetcher read the human site 322 times and never cited it.

## Is it called AI SEO or GEO?

Both names describe the same work: getting your business cited in ChatGPT, Perplexity, Google AI Overviews, Claude, and Copilot. Otterly and most of the industry used GEO (generative engine optimization) first. Lately I hear more people say “AI SEO,” so I use both.

The label doesn’t change the job. Clear content, a site AI bots can reach, and real expertise do the work. My guide explains [what generative engine optimization is](https://go-seo.com/what-is-generative-engine-optimization-geo/), and my [Los Angeles SEO agency](https://go-seo.com/los-angeles-seo-agency/) work covers both traditional SEO and AI SEO.

## Frequently Asked Questions

### Will Google penalize AI-written content?

Google’s guidance says generative AI can help with research and structure, but using it to produce many pages without adding value may violate its spam policy on scaled content abuse. It also advises fact-checking AI content before publishing. In Otterly’s 2,000 AI Blogs test, Google deindexed two sites with about 1,000 posts each.

### Did ChatGPT cite the human-written site?

No. ChatGPT never cited the human-written site on Otterly’s 100 prompts, even though ChatGPT’s on-demand fetcher visited its pages 322 times. It shows that an AI bot reading your page does not guarantee a citation.

### Should I stop hiring human writers?

No. Otterly’s own advice is to use AI for coverage and structure and to keep humans in the loop for brand. The human-written site was named in 62% more AI answers, even though it was cited less often. Real stories, clients, and experience are what AI cannot supply.

### How fast should I publish AI-assisted content?

Steadily. Otterly’s winning pipeline published nearly all 100 articles within 17 days. The sites that Google deindexed had published about 1,000 posts each in under a day. A steady pace with real review behind every page is the safer approach.

### Is one experiment enough to change my content strategy?

No. This test used one low-competition topic and two small, new sites, and the two workflows differed in more ways than authorship. Treat it as a strong signal, not a rule. Otterly is running the same test on three more sites in other niches, so watch for those results.

Want to know whether your pages read as commodity or non-commodity to AI engines? I’ll record a free video [SEO and AI SEO audit](https://go-seo.com/contact-seo/) of your site. All data in this post comes from Otterly.ai’s [AI vs Human Written Content](https://otterly.ai/blog/geo-experiment-ai-vs-human-writing/) experiment, and I recommend following their monthly GEO experiments.
