1 Million URLs Studied: Which URL Structures Gets You Cited The Most on AI Search
URLs are the first signal AI Search engines see, before any content loads. Shorter URLs are “better.” Descriptive slugs “help.” Clean canonical structure “signals authority.” These conventions carry over from traditional SEO, where URL attributes have measurable influence on rankings.
OtterlyAI analyzed 1,028,959 unique URLs cited across six AI Search platforms (ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini, and Microsoft Copilot) over a 24-hour observation window to test whether those conventions still apply to AI Search citations.
Of 1,932,200 AI citation instances observed, the average URL was cited 1.9 times. The median was 1. Against that distribution, we tested 15 URL-level attributes, from character length to page type, for relationships with citation frequency.
Key Findings (TL;DR)
- Shorter URLs do not get cited more. We found near-zero correlation between URL length and citation count (r = -0.025). The average cited URL was 63 characters long, but 40-character URLs and 120-character URLs got cited at almost identical rates.
- “Guide” pages are the most-cited page type. URLs with /guide/ in the path averaged 2.7 citations, 42% above the overall average of 1.9. Blog posts came next (2.0 avg). Pricing pages performed worst (1.5 avg).
- Clean URLs beat messy ones by 24%. URLs without query strings (no “?”) averaged 2.1 citations. URLs with query strings or UTM tags averaged only 1.6. URLs containing digits showed the same pattern (1.6 vs 2.0).
- Path depth does not matter. A URL with 5 subfolders (/blog/category/subcategory/topic/post/) got cited as often as a URL with 1 subfolder. Correlation was 0.002, essentially zero.
- TLD choice does not affect citation rates. .com, .org, .io, and .ai domains all averaged 1.7 citations per URL, effectively identical. Despite handling 528,000+ .com URLs in the sample, the domain extension itself was not a signal AI Search engines use to pick citation sources.
- Over 50% of URLs only get cited once. The median citation count is 1. The average is 1.9. The most-cited URL was cited 965 times. A handful of URLs carry most of the AI Search visibility.
Why URL Structure Matters (or Does Not) for AI Search
AI Search engines select citation sources by retrieving pages, parsing content, and deciding which URLs to reference in generated answers. URLs are the first metadata an AI system sees.
Most SEO guidance assumes URL structure matters. Shorter URLs are faster to crawl. Descriptive slugs help keyword targeting. Shallow paths are easier to index. These heuristics carry over from traditional search, where URL length correlates weakly but measurably with rankings.
The question for AI Search Optimization is whether AI systems inherit those same preferences, or whether they treat URLs as simple addresses and rely on content for selection. To test that, we measured URL-level attributes across a large population of already-cited URLs and compared them to citation counts.
Scope of Study
OtterlyAI monitored 1,028,959 unique URLs cited across ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Gemini, and Microsoft Copilot over a 24-hour window. In total, 1,932,200 AI citation instances were recorded.
For each URL, we extracted the following data:
| Attribute | What we measured |
|---|---|
| URL Length | Total character count of the full URL |
| Domain Length | Character count of the domain only (e.g., “nightwatch.io” = 13) |
| Path Depth | Number of subfolders in the path (count of / after the domain) |
| Hyphen-Word Count | Total hyphens in the URL (proxy for slug word count) |
| Trailing Slash | Whether the URL ends with / (yes/no) |
| Has Query String | Whether the URL contains ? (e.g., ?id=123, ?utm_source=…) |
| Has Fragment | Whether the URL contains # (e.g., #section-2) |
| Year in URL | Whether the URL contains a year between 2023 and 2027 (freshness proxy) |
| Digit in URL | Whether the URL contains any number 0 to 9 |
| Question Pattern | Whether the slug contains what-is, how-to, or why- |
| Comparison Pattern | Whether the slug contains -vs-, best-, top-, alternatives, or -review |
| TLD | Top-level domain extension (.com, .org, .io, .ai, .uk, etc.) |
| Domain Dots | Number of dots in the domain (1 = root, 2+ = subdomain or multi-part TLD) |
| Is Homepage | Whether the URL is the domain root (path depth = 0) |
| Page Type | Inferred from path: Blog, Guide, Docs, Help, Learn, Pricing, Product, Compare, News, Homepage, or Other |
We used the Pearson correlation coefficient (r) for continuous attributes and compared average citation counts across categorical flags. Pearson r ranges from -1 to +1, with values near 0 indicating no linear relationship.
Two interpretive notes:
- Correlation does not imply causation.
- A near-zero Pearson r does not rule out categorical effects. A lift in group averages (for example, Guide pages at 2.7 vs 1.9 baseline) can be practically meaningful even when the overall correlation is negligible.
What URL Attributes Get Cited Most in AI Search?
1. URL Structure Basics Show Near-Zero Correlation
| URL Attribute | Correlation (r) | Interpretation |
|---|---|---|
| URL Length | -0.025 | Negligible (negative) |
| Domain Length | -0.007 | Negligible (negative) |
| Path Depth | +0.002 | Negligible (positive) |
| Hyphen-Word Count | -0.013 | Negligible (negative) |
| Domain Dots | +0.039 | Negligible (positive) |
Every continuous URL attribute we measured returned a Pearson correlation below 0.04 with citation frequency. In practical terms:
- A 100-character URL is no more or less likely to be cited than a 40-character URL.
- A deeply nested page at /blog/category/post is not penalized relative to /post.
- Hyphens/total words in slugs do not predict citations in either direction.
- Domain character length does not matter.
This result is consistent with broader AI citation research. Other studies observed no linear relationship between page attributes like word count and citation count in AI platforms. An Ahrefs analysis of 1.9 million AI Overviews citations showed a similar pattern, with near-zero correlation between content length and citation frequency.
What this means: optimizing URL length, depth, or slug formatting as a GEO tactic delivers effectively minimal return at this dataset scale.
2. Landing Pages are the Strongest URL-Level Signal
| Page Type | URL Count | % of Total | Avg Citations | vs Baseline |
|---|---|---|---|---|
| Other/ landing page | 828,097 | 80.5% | 1.9 | baseline |
| Blog | 99,683 | 9.7% | 2.0 | +5% |
| Homepage | 28,676 | 2.8% | 1.9 | baseline |
| News | 24,075 | 2.3% | 1.7 | -11% |
| Product/Service | 23,028 | 2.2% | 1.6 | -16% |
| Guide | 8,649 | 0.8% | 2.7 | +42% |
| Learn | 7,448 | 0.7% | 1.7 | -11% |
| Help | 3,297 | 0.3% | 2.0 | +5% |
| Compare | 3,425 | 0.3% | 1.7 | -11% |
| Docs | 1,871 | 0.2% | 1.6 | -16% |
| Pricing | 710 | 0.1% | 1.5 | -21% |
When URLs are grouped by inferred page type (based on path segments like /blog/, /guide/, /pricing/), clear differences emerge. Guide pages outperform every other category. Pricing pages perform worst. The gap from top to bottom is 80%.
Interpretation: AI Search prioritizes reference-style content. Guides, long-form blog posts, and help articles provide structured, citable answers. Pricing and product pages deliver transactional information that AI systems deprioritize for informational queries.
So why are pricing/product/service pages not cited more? It seems that brands tend to focus on creating blogs and landing pages. It is plausible that a majority of websites on the web do not wish to publish pricing information or do not have clear enough information that an AI likes to reference on their product/service pages. If this is the case, then that should be a priority in your strategy.
What this means: Do be the most polished brand voice. Focus on filling in gaps in what a buyer wants to know about a product or service. A useful frame for spotting those gaps is Chantal Smink’s QPAFFCGMIM framework:
- Questions
- Problems
- Alternatives
- Frustrations
- Fears
- Concerns
- Goals
- Myths
- Interests
- Misunderstandings
This framework helps you surface the unanswered questions buyers carry into a purchase decision. The takeaway: audit your service and product pages for missing information a buyer would still need to choose, then fill those gaps with clear, bottom-funnel content that resolves alternatives, comparisons, fears, and concerns. Those are the pages AI engines cite when a user is close to choosing.
filing it under /guide/ or /blog/ rather than /product/ or /pricing/ aligns it with how AI Search selects citation sources.
3. Dynamic & Query URLs Reduce Citations by 24%
URLs with query strings (?param=value) averaged 1.6 citations.
URLs without averaged 2.1.
That is a 24% reduction across 430,628 query-bearing URLs.
The digit-in-URL pattern mirrors this: 1.6 vs 2.0. These two signals likely overlap. URLs with digits are often:
- Query-parameter URLs like ?id=12345
- Tracking-tagged URLs like ?utm_source=…&utm_medium=…
- Session-specific or paginated URLs
AI systems appear to treat these as lower-value references, likely because they are:
- Non-canonical versions of a canonical page
- Dynamic content that AI crawlers may not index consistently
- Less stable and less shareable over time
What this means: for pages you want cited by AI Search, serve clean canonical URLs. Use rel=”canonical” to consolidate signals. Strip tracking parameters from URLs intended for external citation or sharing.
4. Country-level Domains Carry Outsized Weight (.uk in this example)
| TLD | URL Count | % of Total | Avg Citations |
|---|---|---|---|
| .com | 528,314 | 51.3% | 1.7 |
| other | 374,657 | 36.4% | 1.7 |
| .uk | 43,331 | 4.2% | 3.0 |
| .org | 32,395 | 3.1% | 1.7 |
| .net | 12,549 | 1.2% | 1.6 |
| .io | 11,235 | 1.1% | 1.7 |
| .ai | 8,326 | 0.8% | 1.7 |
| .co | 6,170 | 0.6% | 1.6 |
| .edu | 6,095 | 0.6% | 1.5 |
| .gov | 5,887 | 0.6% | 1.5 |
Most TLDs in the dataset cluster tightly around 1.5 to 1.7 average citations. One exception: .uk domains, averaging 3.0 citations per URL.
We are careful about this finding. A 24-hour window is narrower than studies with stronger temporal coverage. Without prompt-level metadata, we cannot determine whether .uk domains are genuinely preferred by AI systems, whether the prompts that returned .uk citations were high-volume ones, or whether UK-oriented queries dominated the observation window.
What this means: we flag .uk performance as a finding worth structured investigation in a longer study, not a recommendation to shift to .uk TLDs.
5. Citation Volume Follows a Heavy Power-Law Distribution
Across 1,028,959 URLs:
- Average citations per URL: 1.9
- Max citations on a single URL: 965
- 15.8% of all URLs account for 50% of all AI Citations
- 20% of all URLs account for 54% of all AI Citations
(Quick note on what you’re seeing: The X-axis in this graph uses a non-linear spacing so the head of the distribution stays visible.)
Roughly half of all cited URLs appear exactly once in the 24-hour window. A small minority (less than 1%) accumulate hundreds of citations each. This is a textbook power-law distribution called a long-tail distribution, the same pattern seen in link-building, social sharing, and most discovery systems.
What this means: the value of being cited once is limited. URLs that drive meaningful AI visibility at scale sit in the rare, high-frequency citation tail. Structural URL factors that improve odds by a few percent are less valuable than earning a single page into that tail through content quality, authority, and reference value.
URL Patterns With No Observed Citation Lift
Despite being widely used SEO tactics, the data shows several familiar optimization patterns have no observable connection to AI citation counts. The signals below appear roughly as often in low-citation URLs as in high-citation ones.
| Attribute | Avg (Y) | Avg (N) | Effect |
|---|---|---|---|
| Year in URL | 1.9 | 1.9 | None |
| Question pattern (how-to, what-is) | 1.8 | 1.9 | None |
| Comparison pattern (-vs-, best-, top-) | 1.9 | 1.9 | None |
| Is homepage | 1.9 | 1.9 | None |
The URL Playbook for AI Search
Based on what we observed, here is what we would actually prioritize if the goal is AI citation visibility.
1. Product pages still matter to avoid zero click
Despite that guide, blog, and help pages consistently earn higher citation averages than product, pricing, or unclassified pages. It does not mean you should only focus on top funnel content. Keep in mind that in a world where zero clicks for informational content is steadily increasing. Product/landing (bottom funnel) pages still serve as the most reliable way to get actual traffic clicks.
If a page is built to answer a question comprehensively, file it under /guide/ or /blog/. This is directionally aligned with how AI Search selects citation sources..
2. Serve Clean Canonical URLs
Query strings reduce citation frequency by 24% in this dataset. For content you want cited:
- Use rel=”canonical” to consolidate signals toward the clean URL version
- Strip unnecessary additional symbols and parameters from URLs.
- Keep parameters out of internal navigation
3. Do Not Over-Engineer URL Structure
URL length, path depth, hyphen count, and slug word count do not predict citations. Optimizing these attributes delivers marginal returns at best. The URL is a routing address. AI systems do not use its structure as a quality signal. What’s on your content is more important.
4. Invest in Content, Not Only URLs
URL structure alone is a poor citation predictor. Content structure (headings, definitions, chunking), author attribution, freshness, and page-level reference value consistently show larger effects in citation research. If you have a limited GEO budget, spend it on content quality, not URL refactoring.
Final Conclusion
Across 1,028,959 URLs and nearly 2 million AI citation instances, URL structure basics alone do not predict citation frequency in AI Search. The common SEO heuristics around URL length, depth, slug formatting, and domain structure are not visible in AI citation data at this scale.
What does matter:
- Page type. Guide , Blog , Help, Pricing, Docs, and Product. Reference-style pages get cited more.
- Clean canonical URLs. Too many query strings, symbols and parameters cost roughly a quarter of your potential citations.
- The content itself. URL structure cannot substitute for reference/content value.
AI Search rewards pages that answer questions clearly. URLs route users and crawlers to those pages.
Got a GEO experiment idea? Let’s collaborate!
At OtterlyAI, we don’t just write about AI Search. We test it! With the community, we run and update a public GEO Experimentation Tracker with what we’re running and what moved AI Search. Running your own GEO experiment? Email rick.tousseyn@otterly.ai and we may feature it with your name attributed.
👉 Want to measure your AI Visibility? Start with OtterlyAI and our GEO Guide.
