Research

Original-data research on AI-mediated visibility and work: how systems find, parse, cite, and route, measured from real data rather than opinion. Each study starts from primary sources, states its method and its limits, and publishes what survives inspection.

How to measure brand visibility in AI search results

September 28, 2026

Mention, answer inclusion and citation come apart the moment you sample an assistant more than once. One prompt can carry a brand's entire presence, a parser can move its citation rate 2.5x, and a hallucinated license still counts as a mention. A workflow for technical teams, built on 359 answers.

Method: A fixed 10-prompt panel (8 non-branded category and comparison prompts, 2 entity prompts) run 10 times per prompt (5 for entity prompts) on four surfaces through one API gateway on 2026-09-28: gpt-5-mini without tools, gpt-5-mini with web search, Gemini 2.5 Flash without tools, Perplexity Sonar (gpt-5-mini at reasoning effort low, web search at search_context_size low). Mentions matched by regex, citations parsed from annotations and inline links, answer inclusion and entity accuracy graded by hand.
Primary sources: 359 model answers collected via Vercel AI Gateway (OpenAI Responses API with web_search, chat completions for Gemini and Sonar) · First-hand: Beetroot, a Windows clipboard manager I build (max.nardit.com/beetroot, github.com/mnardit/beetroot-releases)

The Quality Paradox: Why Search Improved for Google While It Degraded for Publishers

May 23, 2026

Three layers of evidence (publisher traffic decay, Alphabet's revenue mix, the leaked Google ranking schema) showing why no algorithmic recovery is coming.

Method: Triangulation across three independent evidence layers: a publisher-traffic decay measurement, an Alphabet revenue decomposition, and analysis of the leaked Google ranking schema.
Primary sources: 27 publisher sites (Ahrefs) · 12 quarters of Alphabet filings · the March 2024 Google ranking-schema leak

Tracing the Pizza Day bitcoins

May 18, 2026

Every year the same articles. Counterfactual valuations, the road-trip rumor, the man-bites-pizza headline. Nobody traces the actual coins. So I did. Here is what the chain shows and what 'those specific bitcoins' even means.

Method: On-chain forensic UTXO trace under strict positional FIFO, followed to depth 5 from the original transaction.
Primary sources: The Bitcoin blockchain · the original 10,000 BTC payment of 2010-05-22