You've invested heavily in content, optimized structure, added microdata. But ChatGPT and Perplexity still ignore your page. The culprit might be a detail many overlook: the URL structure. AI bots don't "read" URLs the way humans do. However, they use URLs as a technical signal: how logically the page fits into the site hierarchy, whether the topic is clear from the address, and whether the URL shows signs of a "spammy" page.

This article explains how URL structure affects visibility in AI-driven search, which formats work, which kill citations, and how to build URLs for AI models.

What AI Models "See" in a URL

AI bots don't analyze URLs the way search engines used to. Keywords in URLs no longer carry decisive weight. But models use URLs for three tasks. First, determining page type. "example.com/blog/how-geo-works" — the model understands it's a blog article. "example.com/catalog/product-123" — a catalog page. Second, assessing hierarchy and depth. A URL with 2-3 segments is seen as an important top-level page. A URL with 6-7 segments ("example.com/blog/seo/geo/2026/how-geo-works") is treated as a secondary, deep page. Third, filtering out temporary pages. If a URL contains "?", "=", "&", "id=123", "sort=", the model may consider the page technical or temporary and exclude it from answers.

Important. There's no direct link between "good URL → high citations." But a bad URL can prevent the model from considering your page, even if the content is perfect. A professional SEO site audit includes checking URL structure.

URLs That Kill Citations

Dynamic parameters and sessions. Example: "example.com/page?id=12345&session=abcde&sort=desc". The model sees question marks, equals signs, ampersands, and long parameter chains. AI bots perceive such URLs as temporary, technical pages not meant for permanent use. Even if the page is a full-fledged article, the model may ignore it. Solution: configure clean URLs via server config files or plugins.

Excessive nesting depth. Example: "example.com/blog/seo/geo/2026/march/how-to-optimize-for-chatgpt". The model sees 6-7 segments. AI bots interpret such URLs as a sign that the page is too deep in the hierarchy and not important. It might be a "draft" or "archive." The ideal depth for an important page is 2-3 segments after the domain. Example: "example.com/blog/geo-optimization" or "example.com/services/seo-promotion".

Numbers and IDs instead of words. Example: "example.com/posts/123456". The model sees only digits. AI bots can't determine the page topic from the URL. Models have to rely solely on the content inside. But for preliminary source selection, they may use the URL. Solution: replace numbers with transliterated keywords.

Stop words and extra characters. Example: "example.com/how_to_choose__seo__contractor". The model sees double underscores. AI bots perceive messy URLs as a sign of a low-quality site. Underscores, uppercase letters, Cyrillic characters in Latin text, extra hyphens — all reduce trust. Solution: use only lowercase Latin letters, digits, and hyphens. No underscores, no capitals, no spaces.

Changing parameters on every request. Example: "example.com/article/123?utm_source=google". The model sees different URLs for the same page. AI bots may get "confused" and think these are different pages. As a result, citations may not be attributed to any of the variants. Solution: set up canonical URLs and avoid using parameters in URLs.

Effective GEO optimization requires clean, understandable URLs.

Correct URLs for Generative Optimization: Five Principles

Principle 1: Use short URLs. Ideal length is 50-80 characters after the domain. Too short (under 30 characters) may be uninformative. Too long (over 100 characters) signals deep nesting. Example of a good URL: "example.com/geo-optimization". Example of a bad one: "example.com/blog/seo/ai/generative-engine-optimization/2026/how-geo-works-detailed-guide-for-beginners".

Principle 2: Use transliteration of keywords without stop words. No conjunctions, prepositions, or particles. No "how", "to", "for". Example of a good URL: "example.com/choose-seo-contractor". Example of a bad one: "example.com/how-to-correctly-choose-seo-contractor-for-your-business".

Principle 3: Use hyphens to separate words. Hyphens are the standard separator for search engines. Underscores don't separate words; they glue them together. The model may not understand that "geo_optimization" is two words. With a hyphen — "geo-optimization" — it's clear. No spaces or other characters.

Principle 4: Reflect hierarchy but don't go too deep. Maximum 3 segments after the domain: 1st level — section ("services"), 2nd level — category ("seo-promotion"), 3rd level — page ("comprehensive-seo"). Deeper only for archives and secondary pages. Important pages — no deeper than the third level.

Principle 5: Avoid parameters and identifiers. No "?id=", "=", "&", "?sort=". Only static, clean URLs. If your CMS generates parameterized URLs, configure clean URLs via plugins or server config files. Keep parameters only for technical pages (search, filters, sorting) that don't aim for citations. Quality content creation will be more effective with clean URLs.

Impact on Different Page Types

Homepage. URL — "example.com" or "example.com/". Short, clean. High impact on AI-driven search: the model often uses the homepage as a source of general brand information. Long URLs with parameters for the homepage are unacceptable.

Service and product pages. URL — "example.com/service/seo-promotion" or "example.com/product/laptop". The structure should reflect the category and name. High impact: the model may cite these pages as sources of specific data. Avoid identifiers and numbers in product page URLs.

Category pages. URL — "example.com/services" or "example.com/products". Short. Medium impact: the model may use the category for answers to general questions. If a category page is just a list of product links, the model may ignore it. Add content to the category page, not just a list.

Blog and article pages. URL — "example.com/blog/article-title". The structure should include the blog section but not go into deep nesting. Not "example.com/blog/seo/geo/2026/how-geo-works". High impact: AI models actively cite informational articles. Short, clean URLs help the model identify the topic.

About and Contact pages. URL — "example.com/about", "example.com/contacts". Short, without extra segments. Medium impact but important for trust signals. The model checks these pages to assess brand reliability. Don't hide them in deep nesting.

How to Check Your URL's Readiness for AI-Driven Search

Method 1: Visual check. Look at the URL in the browser's address bar. Can you tell what the page is about without opening it? If not — the URL is bad. If the URL contains question marks, equals signs, ampersands, identifiers — bad. If the URL is too long — bad.

Method 2: Check for parameters. Open the page in incognito mode. Click through links. If parameters appear in the address bar where they shouldn't — configure clean URLs. Use webmaster tools to see which parameterized pages are in the index.

Method 3: Analyze depth. Count the number of slashes in the URL after the domain. If there are 4 or more — the page is too deep. Restructure the hierarchy: move important pages up a level. Exception — archive pages and secondary materials.

Method 4: Check via webmaster tools. Section "Crawl" → "Crawl stats". See how often bots crawl pages with different URL types. If parameterized pages are crawled more often than static ones — there's a problem. Configure clean URLs and block parameters from indexing in robots.txt.

Read about new search trends in the article new search trends. You can comprehensively improve your site's visibility with comprehensive SEO promotion.

Conclusion

URL structure is not the main factor for AI system citations, but it's an important technical signal. Models don't rank pages by keywords in URLs, but they use URLs to assess page type, importance, and quality. A bad URL can prevent the model from considering your content, even if it's perfect.

Five principles of a good URL for AI-driven search: short URLs; transliteration without stop words; hyphens to separate words; maximum three levels of depth; no parameters or identifiers. These principles don't guarantee citations, but they eliminate a technical reason for your page being ignored.

Start by auditing your most important pages — homepage, services, articles. Check URLs for parameters and depth. Configure clean URLs if you use dynamic ones. Remove identifiers and numbers. These are basic actions that take 1-2 days but eliminate one of the potential causes of invisibility. And remember: a perfect URL won't save bad content. But a bad URL can kill good content.

Frequently Asked Questions

Should I change old URLs that are already indexed?
Changing old URLs is risky. If a page already has traffic and citations, changing the URL could lose it all. If the page isn't cited and has no traffic, you can change it, but be sure to set up a permanent redirect from the old URL to the new one. Prioritize new pages. For old ones, fix only critically bad URLs (with parameters, too long).

Which format is better: transliteration or English words?
For the Russian segment — transliteration. "geografiya-okhvatov" is clearer to the model than "location-coverage". Exception — if you work with an international audience. The key is consistency. Don't mix transliteration and English words in the same URL. Transliterate keywords from Russian.

Does HTTPS affect citations?
Indirectly, yes. HTTPS is a basic security signal. Sites without HTTPS may be flagged by browsers as "not secure". The model may consider this a trust factor. You should switch to HTTPS anyway. Certificates are now free. It won't give you an advantage, but lacking HTTPS could be a disadvantage.

What if my CMS doesn't allow clean URLs?
Switch to a different CMS or use plugins. If that's not possible, leave it as is, but accept that citation chances decrease. In such a system, compensate for bad URLs with perfect content and microdata. But the best solution is migrating to a modern CMS that supports clean URLs.