Case studies
We changed what the AI recommends.
Same prompts. Same competitors. A different recommendation over time, and every number verified against the client’s own domain.
Each engagement runs a locked head-to-head benchmark: the same buyer-intent comparisons, against the same competitor field, put to ChatGPT, Perplexity, Gemini, and Grok every month. The content we publish on the client’s own domain raises the AI-perceived authority of that domain. That is what changes the platforms’ opinion, and the next run verifies whether the recommendation moved. And a human approves every piece before it publishes: the system creates velocity, people keep judgment.
Measured results
SMS & messaging
+480% wins
verified AI wins in 11 weeks: 5 → 29 of 80 queries
Soapbox Bulletin
From sixth place to the most recommended brand in its benchmark, on identical prompts, every result verified against soapboxbulletin.com.
Read the case study
Wealth management · brand redacted for compliance
+300% wins
May to June (1 → 4), from a zero April baseline, across 3 of 4 platforms
An independent wealth manager
More than a dozen firms share the client's name inside the AI engines. URL-grounded measurement found the real story: win the name first, then the market.
Read the case study
How the benchmark works
Discovery prompts measure whether a brand can be found. Head-to-head prompts measure whether a brand can be chosen.
Locked prompts
The prompt set is locked at the baseline and rerun without modification, so every measurement is like-for-like. The questions never change; only the market’s answer can.
URL-grounded verification
Every prompt requires the AI to return each company’s official domain, so every answer carries its own verification key. A result counts only when that domain is the client’s. Name-matching alone inflates results; we exclude it.
Directional confidence
AI answers are sampled point-in-time and platforms update continuously. Trends across identical prompt sets are meaningful; single-run values carry sampling noise, and we say so.
Human in the loop
Core to how we work: the system drafts, people decide. Every article is curated, edited, and approved by a human before it goes live, and every byline belongs to a real person. Nothing publishes on its own.
The objective is not to increase mentions. The objective is to change which company the AI recommends first.
01 · Build evidence
Publish authoritative, structured content AI systems can reliably crawl, understand, and cite. Drafted at machine speed, approved by a human every time.
02 · Establish identity
Make the engines associate the brand with the correct company and domain — entity resolution, URL-verified.
03 · Win the comparison
Head-to-head recommendation wins on a locked prompt set against a constant competitor field. The leading indicator.
04 · Expand influence
Wins accumulate into broader referencing. Mentions, visibility, and conversions are intended to follow the preference.
Who does the AI recommend first in your market?
A walkthrough shows you, live, how ChatGPT, Perplexity, Gemini, and Grok answer your buyers' questions today, and what we would publish to change their answer.