ChatGPT citations prioritize ranking and precision, not length. I recently came across an intriguing study conducted by AirOps that examined how ChatGPT assigns citations. It revealed that pages with precise, narrow answers are favored over lengthy, broad content.
After reviewing 16,851 queries, AirOps found that pages with well-matched headings and focused content rank higher in citations. Impressively, the top retrieval result was cited 58% of the time, indicating a strong preference for relevance over mere volume.
Why this matters to us. These findings are crucial if we’re aiming to earn more ChatGPT citations. To succeed, we need to prioritize winning retrieval spots, mirroring queries in our headings, and providing highly precise answers.
Key insights. The study emphasized retrieval ranking as a pivotal factor. Top-ranking pages were cited 58.4% of the time, compared to only 14.2% for pages positioned tenth. This highlights the significant impact of retrieval rank on citation frequency.
Another crucial point I noted was the importance of heading relevance. Pages where the heading strongly matched the query were cited 41% of the time, significantly outperforming less matched options.
It also showed that narrowly focused pages outperform comprehensive guides, challenging the typical “ultimate guide” approach many of us might consider effective.
Factors driving citations. From what I gathered in the study, being well-ranked, using query-matching headings, and maintaining content focus are key to earning citations from ChatGPT.
Additional structural insights: While structure like JSON-LD markup offered a slight boost in citations, it wasn’t as critical as I initially thought. Pages with this markup had a citation rate of 38.5% versus 32.0% for those without. Interestingly, articles with 4 to 10 subheadings performed notably well.
Furthermore, content length had diminishing returns. Pages with 500 to 2,000 words performed best in citations, whereas those exceeding 5,000 words were cited less than even the briefest ones.
Freshness matters, but only to an extent. Content published within 30 to 89 days had the best performance in terms of citations, while newer content underperformed slightly, suggesting the need for time to build retrieval signals.
Older content, particularly those older than 2 years, struggled in citations, implying the potential benefits of refreshing existing content if it currently ranks well for target queries.
Understanding the data. AirOps examined 50,553 responses derived from 16,851 unique queries, each run three times. The exhaustive dataset encompassed 353,799 pages across various sectors and query types.
The detailed analysis is documented in the report titled The Fan-Out Effect: What Happens Between a Query and a Citation.
Inspired by this post on Search Engine Land.
Leave a Reply