Auditing 158 Articles: What Does ChatGPT Actually Cite?

A Real-World AI Visibility Audit

As more users turn to ChatGPT, Perplexity, Google AI Overviews, and Claude for information and answers, a critical question has emerged: what content do these AI answer engines actually cite? Which articles get "seen" by AI, and which ones get filtered out?

Recently, a brand team partnered with AI analytics firm humanswith.ai to conduct a transparent, publicly verifiable AI visibility audit. They reviewed 158 publicly published pieces of content, evaluated 78 of them against 28 criteria, and tracked citation behavior across four major platforms: ChatGPT, Perplexity, Google AI Overviews, and Claude. All scan IDs and timestamps were made public so anyone can independently verify the results.

Some of the findings are deeply counterintuitive — even challenging core assumptions many content creators hold about how to get AI to cite their work.

Writing Quality Barely Affects Whether AI Cites Your Content

Perhaps the most surprising finding: writing quality has almost no impact on whether content gets cited by AI.

The data showed that articles cited by AI had an average quality score of 60.4, while articles that were not cited actually scored higher at 62.5. In other words, the articles AI ignored were, by conventional standards, better written.

This challenges one of content marketing's core beliefs — that quality is king and good writing gets rewarded. In the world of AI answer engines, models aren't judging your prose or narrative craft. They're looking for structured, extractable, trustworthy information fragments. Elegant, essay-style writing may be no more valuable to a machine than a simple bulleted list.

Your Publishing Platform Is the Decisive Factor

If quality doesn't matter, what does? The answer: your publishing platform.

One of the most striking comparisons in the audit involved the same author and the same material published on different platforms — with dramatically different citation rates:

• Published on an industry publication platform (e.g., a Medium partner publication): ~52% citation rate
• Published on a company's own blog: 0%
• Published on the author's personal Medium page: 0%

This means AI models are heavily dependent on platform authority and credibility signals when selecting what to cite. No matter how good the content, if it's published on a channel that lacks authority signals, it has almost no chance of being cited by ChatGPT, Perplexity, or similar AI answer engines.

This raises a sharp, practical challenge for enterprise content strategy — your carefully maintained company blog may be essentially invisible in the AI era.

Structured Content Beats Beautiful Writing

The audit data further reinforces a "structure first" principle:

• Articles using H2 headings phrased as questions saw citation rates increase by ~19%
• Articles containing tables and lists saw citation rates increase by ~18%

The logic is clear: AI answer engines are fundamentally trying to match specific user queries with relevant answers. Question-format headings align naturally with how users search and prompt; tables and lists present information in a highly structured way that models can directly extract and cite.

For content creators, this offers an actionable direction: rather than chasing elegant prose, break down topics with Q&A-style headings and organize key information using lists and tables.

Marketing Language Kills AI Visibility

Another clear signal from the audit: any promotional or sales-oriented language significantly reduces AI visibility.

Content with sales-oriented language saw citation rates drop by 26%. Models actively filter out promotional phrasing.

This is especially important for brands. Enterprise content has traditionally carried product promotion and service advocacy, but in the eyes of AI answer engines, that "self-promotional" tone is actually a liability. To be cited by ChatGPT or Perplexity, content must return to a neutral, objective, information-first stance — and resist the impulse toward commercial promotion.

Why Isn't Freshly Published Content Being Cited?

The audit also surfaced an easily overlooked variable: time since publication:

• Articles published more than 2 months ago: ~43% citation rate
• Freshly published articles: only ~7% citation rate

This suggests that AI answer engines prefer content that has been time-tested — more broadly indexed and validated. New content needs a "settling period" before it enters a model's citation pool.

For content strategies seeking quick wins, this is a reality check: AI visibility isn't immediate — it's a long-term investment that requires patience.

Which Content Formats Are Never Cited by AI?

Beyond these patterns, the audit also identified several content formats with near-zero citation rates:

• Short-form content (shorts)
• Overly narrow, niche-vertical headlines
• Single-company case studies

Across all four platforms — ChatGPT, Perplexity, Google AI Overviews, and Claude — these formats had citation rates of essentially zero. This reinforces the earlier conclusions: AI answer engines favor broadly applicable, structured, non-promotional informational content, and show little interest in content that is too short, too niche, or centered on a single company.

How to Get Your Content Cited by AI Answer Engines

While this audit was limited in scale (78 deeply evaluated pieces), its transparent methodology — publishing the full prompt set and scan IDs — provides a rare, verifiable reference point for AI visibility research.

Taken together, as AI answer engines increasingly dominate how information is distributed and discovered, content strategy needs a fundamental reorientation:

1. Platform selection over content polish — Publishing on authoritative industry platforms is more effective than endlessly refining content on your own channels.
2. Embrace structured writing — Use question-format headings, tables, and lists to organize content.
3. Eliminate marketing tone — Maintain a neutral, objective, information-first voice.
4. Accept the time cost — AI visibility requires a settling period; don't expect overnight results.

As the gateway to information gradually shifts from traditional search engines to AI answer engines, understanding "what AI actually cites" is becoming essential knowledge for every brand and content creator. The answers this audit delivers may not be what you want to hear — but they're honest.