LLMs Cite AI Content More Than Human Content, Once You Control for Page Type

LLMs Cite AI Content More Than Human Content, Once You Control for Page Type

SEO & AEO
Slate
Research study
A page-type controlled look at what assistants send traffic to, where editorial pages skew 71% toward purely AI-written content.
Organisation
Slate
Written By
Akilan Arumugam
PUBLISHED
August 5, 2026
Read
3
min

What it is.

This looks at 41,853 assistant-referred sessions across 19,000 B2B SaaS URLs in the first quarter of 2026, classifying each landing page by how it was written.

The aggregate says 57% of sessions landed on pages with both AI and human input. Controlling for page type flips the reading: among editorial pages, 71% of sessions landed on purely AI-written content.

Page type decides the answer. Pricing posts run 92% purely AI, while comparison pages sit at 57%, the closest split anywhere in the data.

What you get.

  • The sample: 41,853 assistant-referred sessions across 19,000 B2B SaaS URLs
  • The aggregate at 57% AI-plus-human, against 71% purely AI once you look only at editorial
  • Page-type splits, with pricing posts at 92% purely AI
  • Per-assistant variation, from Claude at 74.4% down to Gemini at 61.3%
  • Where each assistant sends traffic, with Perplexity at 27% editorial against ChatGPT at 18%
  • The classification method, including the model used and its agreement rate

Read it at the source.

Opens on Slate. Free, and the figures on this page all come from it.
Read the guide

Frequently Asked Questions

Why does controlling for page type matter?
Where is the data from?
Do LLMs prefer human-written content?
Where should human editing go?
Does page type change the answer?
Which assistant is most likely to send traffic to AI-written pages?
Strengths
  • Controls for page type, which is the variable that reverses the headline reading.
  • States the classifier and its three-class agreement rate of 87%, so the labelling is auditable.
  • Breaks results out per assistant instead of collapsing them into one AI figure.
Limitations
  • The sample is GA4 data from TripleDart-managed client properties, so it is one agency's book.
  • Slate's own research, and Slate sells content and AI visibility software.
  • AI detection is probabilistic, so page labels carry error even at 87% agreement.
Skip this if
  • Skip it if you already know which page types earn your assistant referrals.
Ideal for
  • Content leads setting an AI policy
  • SEO and AEO leads

The plays behind the reading.

We build inbound GTM engines for B2B software teams, and these are the frameworks we build from. Bring us a pipeline target and we'll map the route.