AI search platforms like ChatGPT, Google AI Mode, and Perplexity are altering how content material will get found. However what makes one piece of content material get cited whereas one other will get ignored?
To reply this, we analyzed hundreds of citations and in contrast them to comparable pages rating in Google. Our aim was to establish which text-only qualities most strongly correlate with AI quotation conduct, and whether or not these patterns differ from conventional web optimization alerts.
Key takeaways:
Primarily based on our analysis, we discovered 5 content material qualities that confirmed a robust optimistic correlation with AI citations, plus one which confirmed a unfavourable correlation:
- Readability and summarization: +32.83%
- EEAT alerts: +30.64%
- Q&A format: +25.45%
- Part construction: +22.91%
- Structured information parts: +21.60%
- Non-promotional tone: -26.19%
In different phrases: Content material that leads with clear solutions, demonstrates experience, and makes use of structured formatting will get cited extra typically.
The 5 Content material Qualities That Enhance AI Citations
These 5 standards stood out as a result of the distinction between cited and non-cited pages was particularly sturdy. Beneath is a breakdown of every criterion and the way it appeared within the pages that LLMs chosen.
- Readability and summarization (33%)
- E-E-A-T alerts (30%)
- Q&A format (25%)
- Part construction (22.91%)
- Structured information rating (21.60%)
One attention-grabbing parameter right here was non-promotional tone, which confirmed a unfavourable correlation. This doesn’t essentially imply LLMs choose promotional language.
A extra seemingly clarification is that professionally written articles, which are usually well-structured, well-sourced, and optimized, typically use a business or persuasive tone.
As Roma Chereshnev, information scientist at Semrush explains:
Articles written by skilled copywriters are well-optimized for web optimization, well-structured, or just comprise helpful data. And since these articles are sometimes written by professionals to draw site visitors or to supply providers or merchandise, it is attainable that these articles extra typically make use of a promotional tone.
Due to this fact, it is attainable that the problem is not that they use a promotional tone, however that usually, if an article is sweet in itself, it is as a result of it is written by knowledgeable.
Cecilia Meis, senior editor at Semrush, notes that these findings additionally align with core rules of high-quality content material:
Readability and construction aren’t web optimization shortcuts. They merely make data simpler for each folks and AI methods to interpret. When content material is organized, direct, and backed by clear experience, fashions can perceive it extra reliably.
The best way to Use This Analysis
With this information, contemplate experimenting together with your content material to enhance your AI visibility. Search for pages in your web site which can be rating properly on Google however carry out poorly in AI search, and examine them utilizing the factors above.
Add a short, structured abstract originally of the web page that clearly states the important thing takeaway
- Strengthen E-E-A-T alerts by together with writer credentials and linking to dependable sources
- Use Q&A formatting in sections the place readers profit from direct solutions
- Add construction with headings, lists, tables, or charts to assist LLMs section the content material
- Monitor efficiency in Semrush utilizing Natural Analysis (for Google rankings) and the Visibility Overview report (for AI citations)
Content material Qualities With Minimal Influence on AI Citations
The scope of our research checked for 13 content material parameters in complete. 5 of them had been optimistic, one was unfavourable.
The “impression” right here is the % distinction in our optimistic and unfavourable scores
Listed here are the content material qualities that confirmed little to no correlation with AI quotation conduct:
This does not imply these qualities are unimportant for content material high quality, Chereshnev explains. As a substitute, they merely didn’t distinguish cited pages from non-cited ones in our dataset. These qualities appeared in each samples at comparable charges, in order that they didn’t meaningfully differentiate cited URLs from non-cited ones.
Issues change quick within the AI search area, so sooner or later (hold a watch out for an up to date research) these parameters may look extra impactful.
How We Ran the Examine
Our aim was to know how LLMs assume when choosing URLs to quote. Due to this fact, when creating the factors, we did not attempt to match the LLM’s reply with a human’s opinion. We performed analysis attempting to think about the factors from an LLM perspective.
On this research, we centered solely on the textual content seen on the web page. We didn’t consider metadata, HTML construction, schema markup, web page format, or any technical web optimization components. The aim was to know how LLMs reply to content material as textual content, impartial of web page code or key phrase focusing on.
To establish which content material qualities correlate with AI citations, we in contrast two pattern teams of URLs:
- A “optimistic pattern” of URLs cited by AI platforms for a set of prompts
- A “unfavourable pattern” of URLs rating in Google’s high 20 for associated key phrases
We scored each pattern teams throughout 13 content material parameters and measured the % distinction in scores.
That means, we may see which particular content material parameters appeared extra typically on pages that had been cited by AI—even after they weren’t rating on Google for related queries.
Examine interval: July 15 – August 6, 2025
Pattern measurement:
- 11,882 prompts (throughout ChatGPT Search, Google AI Mode, and Perplexity)
- 59,410 key phrases (on Google search)
- 304,805 URLs cited by LLMs (optimistic pattern)
- 921,614 URLs rating on Google search (unfavourable pattern)
- 337,785 complete distinctive URLs
What We Constructed
These findings straight knowledgeable the event of our Content material Toolkit, which helps you optimize content material for AI quotation potential. The software helps you establish the place your content material aligns with these patterns, and the place it could want enchancment.
AI search citations are unstable, however they’re not random.
Our information sheds some mild on what works, so audit your content material to provide it clear construction, E-E-A-T alerts, and Q&A formatting to make it AI search-ready.
