↗Research
AI search research — what the evidence actually says
AI search attracts confident claims and very little evidence. This page collects the sources we rely on, labelled by type, so you can see what is documented by platforms, what independent researchers have measured, and what is our own interpretation.
Get an AI Visibility Audit
By Prithvi Somavarapu · Updated · 2 min read
How to read this page
Every source is labelled:
- Platform documentation — what a vendor says about its own product. Authoritative on mechanics; not independent.
- Independent research — measurements by third parties with published methods.
- Academic research — peer-reviewed or preprint studies.
- Prisol analysis — our own interpretation or data. Always marked, never presented as independent.
Platform documentation
| Source | What it establishes | Our note |
|---|---|---|
| Google Search Central — AI features and your website | AI Overviews and AI Mode have no extra eligibility requirements beyond being indexed and snippet-eligible; may use query fan-out; clicks are reported in Search Console under Web | The most direct guidance any platform has published. Undercuts claims that special markup or files are needed. |
| Google — common crawlers (Google-Extended) | Google-Extended controls Gemini training and grounding; does not affect Google Search inclusion | Important for robots.txt decisions. |
| OpenAI — crawlers | OAI-SearchBot (search), GPTBot (training), ChatGPT-User (user fetches) serve different purposes | Blocking GPTBot does not remove you from ChatGPT search. |
| Perplexity — crawlers | PerplexityBot surfaces sites in results and is not used for training; Perplexity-User handles user-initiated fetches | Allow PerplexityBot for citations. |
| Bing Webmaster Tools — AI Performance report (public preview, 2026) | Reports citation counts for your pages in Copilot and Bing AI answers | Rare first-party AI citation data; no click data. |
Independent research
| Source | Finding | Our note |
|---|---|---|
| Pew Research Center, July 2025 — browsing data from 900 U.S. adults, 68,879 Google searches, March 2025 | Traditional result clicked in 8% of visits with an AI summary vs 15% without; links inside summaries clicked in 1%; sessions ended more often after a summary (26% vs 16%) | Strong method (real browsing data). U.S.-only, one month, Google only. Google disputed the methodology. |
Academic research
| Source | Finding | Our note |
|---|---|---|
| Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan & Deshpande, GEO: Generative Engine Optimization (arXiv 2311.09735, 2023; KDD 2024) | Introduced the term GEO and a benchmark; found that content changes such as adding citations, quotations and statistics could increase a source’s visibility in generative engine responses | Tested on research systems, not on commercial products’ current versions. Useful directional evidence that clear, evidenced content helps. |
Prisol analysis
We have not yet published original research. Planned work will follow these standards:
- Methods published in full: prompts, platforms, dates, locations, run counts, coding rules.
- Sample sizes and limitations stated alongside every number.
- No claims about platform internals that the data can’t support.
- Raw data shared where client confidentiality allows.
Claims we treat with caution
- Precise “AI ranking factors” or percentages from vendors without published methods.
- Single-screenshot evidence of visibility.
- Claims that any file, markup or tool guarantees inclusion in AI answers.
FAQ
Research: frequently asked questions
Why doesn't Prisol publish its own AI visibility statistics?
We will only publish research with a sound, disclosed method and adequate sample. Until then we'd rather cite others' evidence clearly than publish numbers we can't stand behind.
How do you choose which studies to include?
We prefer primary sources — platform documentation, peer-reviewed or preprint research, and independent studies with published methods. Vendor marketing claims without methods are excluded.
Can I suggest a study?
Yes. Email us with a link to the primary source and its methodology.
Related to Research
- Guide Zero-click search and AI summaries: what the evidence says What independent research says about clicks when AI summaries appear in Google, how to interpret it, and what it means for SEO and AI visibility strategy. Read
- AI Search How AI search works How AI search engines turn a question into a cited answer — query rewriting, retrieval, grounding and citation — and what that means for which brands appear. Read
- Comparison Comparisons Clear side-by-side comparisons of SEO, AEO, GEO, AI search optimization, AI platforms, citations, mentions, backlinks and the metrics used to measure them. Read
- Company Methodology The workflow Prisol uses to assess, optimize and measure brand visibility across search engines and AI answers: prompts, sampling, diagnosis and reporting. Read
AI Visibility Audit
Is your brand visible where your customers are searching?
Our AI Visibility Audit shows how you appear across Google and AI answer engines today — and what to fix first.
Prefer email? prithvi@prisol.org