Research program

Research

By Vee Popat, founder of DankRank · Last reviewed September 22, 2026

Search gets noisy when everybody is quoting everybody else's study. Research is where we publish DankRank AI SEO work built on a defined dataset, prompt set or repeatable observation method, rather than one client story or one interesting screenshot.

I publish the methodology with the findings because I want you to see what the result actually covers.

Published studyPB-10

Published researchWe Asked ChatGPT, Claude, Gemini and Perplexity the Same 24 AI SEO QuestionsWe compared the same 24 AI SEO questions across ChatGPT, Claude, Gemini and Perplexity to see which sources each one surfaced and cited. Across 96 observations, the four runs cited 211 domains, with surprisingly little overlap.Read the study and download the data

Current studyRSR 002

RSR 002PUBLISHED

AI Crawler Controls in 2026

We checked robots.txt and llms.txt across the top 100 companies in the 2026 Fortune 500 to measure how their public controls treat AI search/discovery, training or broader AI use, and user-directed retrieval.

Only 8 of the 100 published a named rule for at least one measured search or training control. A real llms.txt text file appeared on 12. Partial-path restrictions were much more common than root-level restrictions, largely because general crawler policy can apply even when an AI agent is never named.

READ THE AI CRAWLER CONTROLS STUDY

METHODOLOGY LOCKED
IN PROGRESS
PUBLISHED— current stage

The State of AI Search Visibility 2026

The separate State benchmark remains in methodology/in-progress state. Its locked v1.2 method uses 32 prompts across eight topic strata and 176 final-study observations. Final collection is still deferred, so that page publishes the method, measurement rules and limitations rather than findings.

SEE THE LOCKED STATE STUDY METHODOLOGY

What We Research

The work starts with the sources and page types that keep showing up around commercial questions. From there, we compare citation overlap across AI systems and look at cases where a brand is mentioned without its own site being cited.

We separate buyer-intent prompts from informational ones, then repeat a smaller commercial subset to see how stable the recommendations and source patterns are. We can study those patterns even though we do not have access to the ranking systems behind them.

Research Is Not PROOF With More Rows

PROOF follows specific DankRank builds, baselines and experiments. Research looks across a larger sample and asks a broader question.

The DankRank launch baseline can tell us what happened to one new domain. I won't turn it into an industry benchmark simply because the chart looks interesting. Keeping those evidence types separate matters.

The Methodology Stays Visible

Every study should tell you enough to judge the finding:

sample size
prompt design
engines observed
dates
repeat frequency
what counted as a citation, mention or recommendation
competitor or brand selection
known exclusions
limitations
material methodology changes

If a later version changes the method, I’ll document the change.

A visibility study can tell us about the prompt set we observed. Questions about total consumer use, influence on commercial decisions or the revenue created by an AI citation need different evidence.

I keep the conclusion inside the boundaries of the data.

Research Should Be Useful Outside DankRank

The best outcome isn't a report that says "hire DankRank for your AI SEO." It's a dataset or finding another search practitioner can challenge, repeat or use to ask a better question.

Commercial “money” pages can link to the research when it supports a claim. The research itself shouldn't need a sales pitch to be worth reading.

For practical measurement, use How to Measure AI Search Visibility.

Related resourcesDR-263

Practical guides for measuring AI search visibility, understanding AI search systems and managing crawler discovery.

Explore all resources