The Lab
Experiments we run inside Meriin. Original research on SEO, GEO, and how AI engines actually behave, with published methods and every number traced to code.
7
Published Experiments
5
Research Domains
0
Client Data Used
7 experiments
GEO AI Engines Barely Agree on Sources: 559 Citations Across ChatGPT, Gemini, and Perplexity
28%
Domains cited by >1 engine
We logged 559 citations from ChatGPT, Gemini and Perplexity on 25 queries. Only 28% of domains were cited by more than one engine. There is no single AI search.
Read the Experiment →
AI Systems Can an LLM Judge a Sentence's Citability? Naive vs Compiled Prompts, Benchmarked
0.25 → 0.81
Judge–rubric correlation
We benchmarked an LLM judging sentence citability against a fixed rubric. A naive prompt scored near random; compiling the rubric tripled its agreement.
Read the Experiment →
Growth Loop Type Doesn't Predict SEO Trajectory: Six Public Companies, 2021 to 2026
16×
Same loop, opposite outcomes
We tracked organic traffic for six public companies by growth-loop type. Same loop, opposite outcomes: Reddit +16.7x, Stack Overflow flat. The loop isn't destiny.
Read the Experiment →
GEO Only One AI Engine Shows Its Work: Query Fan-Out Across ChatGPT, Gemini, and Perplexity
1 of 3
Engines that show fan-out
Of three AI engines, only Gemini reveals its query fan-out, expanding every question into ~2.8 sub-queries. ChatGPT echoes the query; Perplexity shows nothing.
Read the Experiment →
Product-Led SEO Programmatic SEO Is Mostly Dead Weight: 97% of Template Pages Earn 100 Visits or Fewer
97.5%
Pages earning ≤100 visits/mo
We checked the page-by-traffic distribution of four programmatic sites totaling 11M pages. An average 97.5% earn 100 visits a month or fewer; 35% earn zero.
Read the Experiment →
GEO The Extractable Sentence: AI Answers Draw From Short, Early Lines, Not Dense Stats
11 words
Median sentence AI quotes
We traced 377 source sentences that ChatGPT, Gemini and Perplexity drew from. The median was 11 words, near the top of the page, and not the number-heavy ones.
Read the Experiment →
SEO Mechanics What Google's Leaked Signals Actually Do: A Function-Classification of 119 Content-Warehouse Attributes
13.4%
Signals that are ranking levers
We classified the 119 leaked Google signals SEOs audit by function. Only 13% are positive ranking levers. 87% suppress, gate, store, or comprehend.
Read the Experiment →No experiments match that search.
Try a different term, or clear the filter to see everything we've published.
Because "Science" Has to Mean Something.
Anyone can claim data-driven marketing. Meriin Lab is what that claim looks like when you have to show your work, the same discipline that runs behind every client account.
Pre-Registered Hypotheses
We state what would prove us wrong before we collect a single data point, not after, when it's easy to explain away.
Every Number Traced to Code
Every figure in every experiment links back to the script that produced it. No unsourced claims, no cherry-picked screenshots.
Confidence, Stated Honestly
We separate hard census data from judgment calls, and we publish limitations alongside findings. That's what "Science" means at Meriin.
Want One of These Run Against Your Own Data?
This is what backs every strategy we build: hypotheses, tailored experiments, and results tied to commercial outcomes. Not a recycled playbook.
No commitment. No pitch. Just senior thinking applied to your business.