35 Percent of New Websites Since ChatGPT Are AI-Generated and the Internet Has Become Aggressively Positive

New research from 404 Media indicates that approximately 35 percent of all newly published websites since ChatGPT's November 2022 launch were AI-generated or AI-assisted — and the resulting internet has become, in the researchers' words, "aggressively positive." The finding, based on linguistic analysis of nearly 11 million domains registered between 2022-2026, is the most concrete measurement to date of how generative AI has reshaped the public web.

What "AI-Generated" Actually Means in the Study

The 35 percent figure includes pages that were either fully generated by an LLM (writeup, structure, navigation, all of it) or substantially assisted by one — typically meaning the prose was AI-drafted and only lightly edited by a human. The methodology used a combination of well-known LLM-detection heuristics, perplexity analysis, and stylistic markers that AI text consistently exhibits but humans typically do not (over-balanced clauses, transitional phrase patterns, certain low-frequency vocabulary clusters).

The "aggressively positive" finding is striking. AI-generated content is statistically much more likely to contain superlatives ("transformative," "revolutionary," "groundbreaking"), affirmative framings, and absence of contrasting viewpoints. The tone difference between human and AI-written corporate copy is now mathematically detectable.

The SEO and Search Implications

Google has been quietly absorbing this. The March 2024 helpful-content updates and subsequent algorithm tweaks have specifically targeted low-quality AI-generated content. SaveDelete's own SEO performance is part of why we know that — sites that lean too heavily on AI without editorial input are seeing visible ranking decay, while sites with clear human editorial voice and original analysis (which we'd argue includes our own work) have generally held up.

The deeper implication: Google's value is no longer just retrieval. It is curation against an AI-generated background. That is a much harder product to defend than pure search, and it explains why Google has been so aggressive on Gemini integration even at the expense of search ad revenue. The competitive moat for search is shifting from "find the right page" to "find the page written by a human who actually knows what they're talking about."

What This Means for Content Creators

Three consequences worth thinking about:

First, the marginal value of authentically human, editorially-distinct content is rising — not because human writing is intrinsically better, but because it is increasingly rare. Sites that maintain a clear editorial voice and contrarian or specific perspectives stand out more than ever in a sea of AI-written homogeneity.

Second, AI-assisted content is not the enemy. The 35 percent figure includes both bot-farm SEO spam AND legitimate creators using AI as a drafting tool. The distinction Google increasingly cares about is editorial judgment, not whether the words came from a model. Sites that use AI to draft + a human to edit and add perspective are still ranking well.

Third, the "aggressively positive" finding suggests an opportunity: critical, contrarian, opinion-forward writing is mathematically distinguishable from the AI-generated baseline. That is exactly the editorial voice we've been building at SaveDelete with the "My Take" sections in every article.

My Take

Honestly, this is good for editorial sites that have been doing the work properly. The internet is getting flooded with AI-written content that is too positive, too balanced, too generic. The only thing that stands out in that environment is opinion, friction, and specific perspective. SaveDelete's own positioning — opinionated, logical, not politically correct — was built specifically because the alternative is being one more "aggressively positive" generic-tech blog that nobody can distinguish from a bot.

The cold concern for the broader internet is that 35 percent and rising is a structural problem. As AI-generated content compounds, search becomes harder, social media becomes harder to read, and the signal-to-noise ratio for any kind of public-knowledge work degrades. Editorial human work is now a public good with declining supply. Whoever figures out how to commercially fund that supply at scale will be writing the next decade of media business model history.

Frequently Asked Questions

How was the 35 percent figure measured?

404 Media analyzed approximately 11 million domains registered between November 2022 and April 2026 using LLM-detection heuristics, perplexity analysis, and stylistic markers. The methodology has known limitations — LLM detectors are imperfect — but the directional finding is robust.

What does "aggressively positive" mean?

AI-generated text is statistically more likely to use superlatives ("transformative," "revolutionary"), affirmative framings, and absence of contrasting viewpoints than human-written equivalent text. The tonal homogeneity is now detectable through linguistic analysis.

Is AI-generated content ranking on Google?

Increasingly less. Google's helpful-content updates have specifically targeted low-quality AI content. Sites with clear editorial voice and original perspective have held up; pure-AI bot farms have seen significant ranking decay over the past 12 months.

Should creators use AI to draft content?

Using AI as a drafting tool with strong human editorial input is still effective. The distinction that matters is editorial judgment and original perspective, not whether the words started in a model. Pure AI output without human editorial layer is the failing pattern.

The Bottom Line

35 percent of new websites since ChatGPT being AI-generated is the cleanest measurement to date of how generative AI has reshaped the web. The "aggressively positive" tonal finding suggests editorial voice and opinion are now scarce signals against a homogenous background. For content sites with editorial discipline, that is a competitive advantage. For the internet broadly, it is a structural decay in signal-to-noise that needs new business models to fix.