AI-Written Content Isn't the Problem — Thin Content Is

_The debate over "AI content" is aimed at the wrong target. Whether a human or a model typed the words matters far less than whether the page says anything worth citing. The real axis isn't human vs. AI — it's thin vs. information-dense._

▶ Free tool — 60 seconds: Check your AI visibility → See which prompts already mention (or ignore) your brand across ChatGPT, Perplexity, and Gemini.

---

TL;DR

• Two assumptions drive most of the fear about AI content, and both are wrong: (1) that AI-written automatically means spam, and (2) that AI content is inevitably thin and keyword-stuffed.
• On a recent episode of the Distribution Podcast, Ravish Agrawal of Gamma argued the axis that actually matters is thin vs. information-dense — not who or what produced the words.
• His "steel glass vs. plastic glass" framing captures it: a thin page names a product; a dense page explains depth, benefits, and how it compares to the alternatives.
• Ravish notes that Google's May-21 search update penalized thin, mass-posted content (attributed; verify before republishing) — and because ChatGPT pulls a large share of its web results from Bing's index, thin pages lose on AI surfaces too.
• Princeton's 2023 GEO research found that adding quotes, statistics, and citations measurably lifts how often LLMs cite a source — density is a ranking input, not a nicety.
• Use the quality rubric at the end to keep AI-assisted drafts on the dense side of the line.

---

The two myths about AI content

Most objections to AI-assisted writing collapse into two claims. Both sound obvious. Both are wrong.

Myth 1: AI-written means spam. The assumption here is that provenance determines quality — that a sentence produced by a model is inherently lower-value than one typed by a person. But answer engines don't grade drafts on how they were made. They read the finished page for whether it resolves the query. A model can produce a sharp, original, well-sourced explanation, and a human can produce a vague, recycled one. Provenance is not the signal.

Myth 2: AI content is inevitably thin and keyword-stuffed. This is the more understandable myth, because so much AI content is thin — cranked out at volume, optimized for a keyword, and empty of anything a reader couldn't get from ten other pages. But that's a description of how the tool is often used, not a property of the tool. Thin is a choice. As Ravish Agrawal of Gamma put it on the Distribution Podcast, the distinction people should be drawing is not human vs. AI — it's thin vs. information-dense.

Once you move the axis, the whole "AI content penalty" conversation reframes. Google, ChatGPT, Perplexity, and Gemini aren't hunting for machine-written text. They're demoting thin text. The fix isn't to write by hand. It's to write something dense enough to be worth citing.

If you want the deeper framing on how AI engines decide what to surface, our answer engine optimization guide covers the mechanics.

---

Thin vs. thick content: what LLMs actually reward

Ravish's clearest illustration is the steel glass vs. plastic glass example (attributed to Ravish Agrawal of Gamma). Imagine two product pages. The first says: "Buy our glass." The second explains what the glass is made of, why steel outperforms plastic for this use, who it's for, and how it stacks up against the alternatives on price and durability. Same topic, roughly the same word count — radically different information density. The second page gives an answer engine something to extract, quote, and attribute. The first gives it nothing but a name.

That's the whole game. Thick content carries:

• Original claims — a point of view or finding a reader can't get from the ten pages that came before yours.
• Data — numbers, ranges, and measurements that anchor the claim in something concrete.
• Comparisons — how the thing stacks up against the alternatives, which is often the exact shape of the user's question.
• Specificity — named conditions, named tradeoffs, named use cases, instead of hedged generalities.

Ravish reinforces the point with an observation from Gamma's own citation tracking: a 400,000-view YouTube video and an 18,000-view one both earned AI citations (attributed to Ravish Agrawal of Gamma) — so raw view count alone didn't explain which got selected. Something other than popularity was doing the work. Density and relevance were better predictors than reach.

This isn't only a Gamma anecdote. Princeton's 2023 GEO research (introduced in the paper that coined "generative engine optimization") tested content changes against how often large language models cited a source, and found that adding quotes, statistics, and citations measurably increased visibility in AI-generated answers. In other words, the very things that make content dense are the things that make it citable. Density isn't a stylistic preference — it's a documented input to whether you get named.

There's a mechanical reason to care beyond AI engines. Ravish points out that Google's May-21 update penalized thin, mass-posted content (attributed to Ravish Agrawal of Gamma — verify the specific update and date before republishing). And because ChatGPT pulls a large share of its web-search results from Bing's index, a page that reads as thin to a traditional search crawler tends to lose on AI surfaces as well. Thin content fails twice: once in classic search, once in the answer layer built on top of it.

Our guide to generative engine optimization goes deeper on which content structures earn citations across engines.

---

A quality rubric for AI-assisted content

The point isn't to stop using AI to draft. It's to hold every draft — human or AI — to a standard that keeps it on the dense side of the line. Run each piece against these five checks before it ships:

1. Unique claims. Does the page say at least one thing a reader can't find on the first page of results? If every sentence could appear on a competitor's site unchanged, the page is thin regardless of length. Add a point of view, a finding, or a stance.

2. Primary evidence. Are the claims backed by data, examples, or first-hand observation — not just assertion? A number with a source beats an adjective. This is the lever Princeton's research measured directly: quotes, stats, and citations lift visibility.

3. Concrete examples. Does an abstract claim come with a specific instance? The steel-glass example works because it's specific. "Be more detailed" is advice; "explain the material, the use case, and the comparison" is an example someone can act on.

4. Expert input. Has someone who actually knows the domain shaped or reviewed the argument? AI drafts fluently, but it doesn't know which tradeoff your buyers actually lose sleep over. A practitioner's edit is what turns competent prose into a citable point of view.

5. Fact-checked numbers. Every statistic is verified against a primary source, and anything you can't verify is either cut or flagged. Attribution matters: a claim you heard on a podcast is a claim from that speaker until you've confirmed it independently. Density built on wrong numbers is worse than thin content — it's thin content that also erodes trust.

A draft that clears all five is information-dense whether a person or a model produced the first pass. A draft that fails them is thin whether or not a human typed every word. That's the standard the myths keep everyone from applying.

To see which of your pages already clear that bar in the eyes of AI engines — and which get ignored — monitor your brand's prompt-level visibility across ChatGPT, Perplexity, and Gemini.

▶ Start free: Spin up your workspace and track which prompts cite you — open Asva AI →

---

FAQs

Does Google or ChatGPT penalize AI-written content?

Not for being AI-written. As Ravish Agrawal of Gamma argues, engines demote thin content, and Google's May-21 update targeted thin, mass-posted pages (attribution; verify the update details before citing). A dense, original, well-sourced page produced with AI assistance is treated on its merits. Provenance isn't the signal — information density is.

What is "thin content" in the context of AI search?

Thin content resolves a query with little more than a keyword and a name — no original claims, data, comparisons, or specificity. In Ravish's "steel glass vs. plastic glass" framing, it's the page that says "buy our glass" instead of explaining what the glass is, who it's for, and how it compares.

Why do LLMs cite information-dense content more often?

Princeton's 2023 GEO research found that adding quotes, statistics, and citations measurably increased how often large language models cited a source. Dense pages give an answer engine specific, attributable material to extract — thin pages give it nothing but a name to skip.

Does ChatGPT use Google or Bing for web results?

ChatGPT's web search draws a large share of its results from Bing's search index rather than Google's. That's part of why thin content loses twice: pages that read as thin to a traditional search crawler tend to be weak candidates for AI answers built on the same index.

Can AI-generated content rank and get cited?

Yes — if it's information-dense. The tool that produced the first draft doesn't determine outcome; the density does. Gamma's own tracking showed a 400K-view and an 18K-view video both earning citations (attributed to Ravish Agrawal of Gamma), suggesting relevance and depth predicted selection better than reach.

How do I make sure my AI-assisted content is dense enough?

Run every draft through a five-point rubric: unique claims, primary evidence, concrete examples, expert input, and fact-checked numbers. A draft that clears all five is dense regardless of how it was written; one that fails them is thin regardless of who typed it.

---

The takeaway

The "AI content" panic is a category error. The question was never whether a human or a model wrote the words — it's whether the page is thin or information-dense. Thin content was losing before generative AI existed; the answer layer just made the penalty faster and more visible. Write pages with original claims, real evidence, concrete comparisons, and verified numbers, and it won't matter who drafted them. Ship keyword-stuffed filler, and no amount of hand-typing will save it.

If you're not sure which side of that line your content falls on, the fastest way to find out is to look at what the engines already say about you.

▶ See where you stand: Run your free AI visibility report → or book a 30-minute walkthrough to map a content plan around it.

---