Research, as served to AI agents
# What AI agents get at /research
--- title: Research | Tanuj Sharma url: https://tanujsharma.xyz/research --- # Research Working studies on how Google and AI assistants like ChatGPT find, index and recommend websites. Hypotheses stay marked as hypotheses until there is first-party data to back them. ### [What 400K+ crawled-but-not-indexed URLs can teach us about large-scale SEO.](https://tanujsharma.xyz/research/crawled-not-indexed-large-scale-seo) Programmatic SEO · Ongoing · Dataset: 400K+ URLs · Updated October 2026 A working study of crawl behaviour, page differentiation, internal linking and site architecture across a large programmatic URL inventory. ### [How AI referral traffic behaves differently from traditional search.](https://tanujsharma.xyz/research/ai-referral-traffic-vs-search) AI search · Ongoing · Dataset: External benchmarks + first-party measurement in progress · Updated October 2026 A study of what happens after someone arrives from ChatGPT, Perplexity or another AI assistant instead of a traditional search engine. ### [What happens when you remove thousands of low-value URL combinations?](https://tanujsharma.xyz/research/removing-low-value-url-combinations) Technical SEO · Experiment in progress · Dataset: Large programmatic URL inventory · Updated October 2026 A controlled experiment in reducing redundant programmatic URL inventory and measuring what changes in crawling, indexation and search visibility. ## How these studies work - **One question per study.** Each study starts with a question I could not answer from existing advice. - **Every claim is labelled.** Fact: published by a primary source such as Google, and linked. Observation: something I saw in the data. Interpretation: what I think it means. Hypothesis: what I expect but have not tested yet. - **Sources are numbered.** Every outside claim links to where it came from, so you can check it. - **Results need a comparison.** I only call something a result after comparing before and after, or changed pages against unchanged ones. - **Studies change in public.** When new data comes in, the study is updated and the change is logged with a date. - **Failures get published.** If an experiment does not work, that is still the result, and it goes up too. ## Research into practice I publish what I learn here. At MarketingPal, I use what holds up to run search and marketing for lean teams. https://marketingpal.co