This study compares the types and structural characteristics of travel-related pages cited by two leading generative AI systems, Google’s AI Overviews and OpenAI’s ChatGPT.

We looked at a dataset of 2,279 URLs cited in AI Overviews and 1,483 URLs cited in ChatGPT. Results were analysed across six travel categories: luxury, cruise, family, beach, solo, and package tours using an in-house built tool utilising the OpenAI API.

Pages were categorised as homepages, landing pages, or blogs/articles, and subfolder depth was measured to assess structural hierarchy.

Why did we do this study?

Generative AI systems are rapidly transforming how users discover and interact with online information. Google’s AI Overviews integrates synthesised responses within search results, while ChatGPT, with browsing capabilities, retrieves and cites live web pages during conversational interactions.

By comparing AI Overviews and ChatGPT citation patterns across key travel-related categories, we aim to identify whether each system exhibits biases towards certain page types or subfolder structural depths, and to infer what this means for visibility and content strategy moving forwards in the travel industry.

Understanding what types of pages these systems cite provides important insight into how generative models assess topical relevance, authority, and content intent.

Is AI more likely to cite landing pages, blogs, or home pages?

Both systems overwhelmingly favour landing pages (73.4% for AI Overviews; 71.2% for ChatGPT), though ChatGPT draws slightly more from blogs and homepages. Which in turn outlines with a landing page depth analysis, that both systems prefer shallower content architectures, with most citations within two subfolders of the root domain, a consequence of the content types being preferred.

These findings suggest generative AI systems favour structured, commercially relevant, and topically coherent content in the travel sector, though ChatGPT demonstrates a more diverse citation mix.

Key findings

1. Landing pages dominate AI citations across all travel categories

Both AI Overviews and ChatGPT cite landing pages (71-73%) over homepages (4-5%) and blog content (23-24%), with ChatGPT showing marginally more content diversity.

This isn’t surprising that landing pages are the most cited as they’re often purpose-built for queries and AI models prioritise relevance and specificity – landing pages often deliver both by design.

Travel Category AI Overviews Home (%) AI Overviews Landing (%) AI Overviews Blog (%) ChatGPT Home (%) ChatGPT Landing (%) ChatGPT Blog (%)
Luxury 15.4 68.4 16.2 12.0 64.5 23.5
Cruise 1.8 71.3 26.9 3.2 79.2 17.6
Family 1.5 67.2 31.3 3.9 66.9 29.2
Beach 1.6 72.7 25.7 1.9 64.3 33.8
Solo 1.9 75.8 22.3 3.4 73.8 22.9
Package Tour 1.4 85.2 13.5 3.0 78.8 18.1

2. Blogs don’t perform as you would expect

Many expected blogs to do better in the AI era because:

  • They contain detailed, original content
  • They answer “why” and “how” questions
  • They show expertise and authority

The fact that they’re only getting 22-24% suggests AI still prioritizes transactional over editorial content – at least in travel.

This doesn’t mean that blogs are important, as content serves multiple needs for a business – being ranked in Search or cited in AI is just one of those needs.

3. The homepage almost doesn’t exist outside of branded prompts

At 4-5%, homepages are nearly invisible to AI citations. This is expected but stark. It confirms what many suspected: your homepage is for humans, not for AI.

4. ChatGPT has a slight edge in diversity

ChatGPT is citing slightly more blog content (24.2% vs 22.6%) and homepages (4.6% vs 3.9%) which suggests it may be pulling from a broader training set, more willing to cite editorial or brand-building content, and is less focused on transactional pages.

Average distribution:

  • AI Overviews: Home 3.9%, Landing 73.4%, Blog 22.6%
  • ChatGPT: Home 4.6%, Landing 71.2%, Blog 24.2%

Both systems favour citing landing pages, though ChatGPT incorporates slightly more editorial and homepage content, indicating broader diversity.

Landing Page Depth Comparison

Subf. Depth AI Overviews (Count) AI Overviews (% of Landing Pages) ChatGPT (Count) ChatGPT (% of Landing Pages)
1 494 26.4% 237 22.2%
2 552 29.5% 494 46.3%
3 237 12.7% 119 11.2%
4 131 7.0% 120 11.3%
5 80 4.3% 32 3.0%
6 21 1.1% 5 0.5%
7–10 18 0.9% 0 0.0%

In both systems, the most cited landing pages exist within two subfolders of the domain root.

ChatGPT demonstrates a stronger concentration at depth 2 (46.3% vs. 29.5%), suggesting it may retrieve deeper but still structured content more readily than AI Overviews.

The Package Tour and Solo Travel categories show the strongest bias towards landing pages in both systems, reflecting clear commercial intent.

Conversely, lifestyle-oriented categories like Beach and Luxury show more blog inclusion, consistent with inspiration or awareness-oriented searches.

What does this mean for marketers?

This is mostly expected, but the degree to which landing pages dominate (70%+) and how invisible homepages are (<5%) is starker than many might have predicted.

It’s a clear signal that AI rewards specificity and structure over brand storytelling.

Both AI Overviews and ChatGPT rely heavily on landing pages, which typically provide structured, high-authority content aligned with user intent.

This confirms that clarity of structure and commercial orientation are key drivers of generative citation.

Conclusion

This study finds that both AI Overviews and ChatGPT strongly prefer landing pages when citing travel-related content, correlating with the theory of structured, intent-aligned page architecture.

However, AI systems differ in meaningful ways.

ChatGPT shows a more varied citation pattern, drawing on a wider range of blog and homepage content and favouring second-level subfolders, while AI Overviews distributes citations more evenly across shallower subfolder depths.

Key takeaways from this research for travel brands and content strategists:

  • For AI Overviews visibility: Prioritize clear, shallower landing page structures (subfolder depths 1-3) designed around specific user intent categories.
  • For ChatGPT visibility: Maintain a diverse content ecosystem, combining structured hubs with engaging editorial narratives that align with user exploration intent.
  • For both platforms: Content buried beyond depth 5 is essentially invisible. Flatten your site architecture and move critical pages closer to the root.

These findings suggest that generative AI visibility requires both architectural precision and content diversity. Click depth and subfolder depth have long been theorised best practices within SEO, and this data goes some way to corroborating this practice.

Further research should test whether these patterns persist across other industries and how retrieval systems balance authority against contextual richness in evolving AI search experiences.

Study Methodology

The two datasets were collected for this analysis utilising Knowatoa, across a number of queries and questions relating to the different travel sections. The two datasets being:

  • AI Overviews dataset: 2,279 URLs cited in AI-generated summaries for travel-related queries.
  • ChatGPT dataset: 1,483 URLs cited for the same set of queries using ChatGPT’s web browsing mode.

Each cited page was manually or programmatically classified into one of three categories:

  • Homepage (the main domain index page)
  • Landing page (structured commercial or informational hub pages, distinct from general blogs)
  • Blog/article/magazine (editorial or narrative content)

For both datasets, page-type frequencies were calculated by travel segment.

We also looked at landing page depth (subfolder level) to evaluate whether citation likelihood correlated with subfolder depth.

For homepages where there was a language variation (e.g., /en or /en/home), this was factored in and classified as “homepage”, and treated as the first level on subfolder depth.

Limitations and Bias Assessment

  • Both systems update frequently; results may shift as retrieval algorithms evolve.
  • The ChatGPT dataset included fewer URLs (1,483 vs. 2,279), potentially affecting proportional accuracy. This was due to ChatGPT citing fewer URLs for the same prompts than AI Overviews.
  • Differentiating between landing and blog pages algorithmically involves interpretive judgement, particularly where hybrid content exists.
  • The study focuses on citation frequency, not user behaviour or ranking effects.

Future research should incorporate time-based tracking, content format tagging, and cross-industry comparisons to produce a level of “generalisability”.