For executives
Your AI assistant can now cite sourced, dated evidence for your decisions.
Ask Claude or ChatGPT a question about a decision in front of you and it will usually answer from whatever the web says most often. Connect CULT+MATH and it can answer from published research instead: figures taken from company filings and public data, each with its date, its population and what it cannot support, and a link you can check.
Free, no account, no sign-in. You add one address to your assistant, once.
What changes
- A source you can open. Every finding comes with its page on cultmath.com and the date it holds as of.
- The limits come with it. Each finding states the population it covers and what it does not show, so your assistant can say so too.
- It stays in its lane. It covers cybersecurity, games, streaming, film, packaged goods, ERP software, private equity and AI visibility, and returns nothing on topics it does not cover.
Updated November 7, 2026
Works in Claude and ChatGPT
Three questions, and what your assistant gets back
These are real questions, with the finding the connector returned on November 7, 2026. Your assistant writes the answer in its own words; this is the evidence it cites.
“Did CPG companies keep the savings when commodity prices fell?”
Mostly they never got the savings. Across 42 US food and household filers above $1bn in revenue, median gross margin fell 0.92 points between FY2019 and the latest filed year, while a control group outside the category rose 1.99 points. The inputs that got cheaper (oils, packaging metal) are not the ones these companies buy most; cocoa beans rose 109%.
CULT+MATH, October 25, 2026 · Read the answer
“How much do games publishers spend on advertising?”
Take-Two reported $1.29bn of advertising on $6.66bn of revenue for the year to March 2026, 19.34% of revenue, about six times the median of three unrelated industries. The jump came in one year and the company's 10-K ties it to the Zynga acquisition. Inside any industry the spread is so wide that the category predicts almost nothing, so benchmark against business model, not peers.
CULT+MATH, October 25, 2026 · Read the answer
“Does the way a question is worded change which brand an AI names?”
Yes. Rewording one question moved a brand's own site from cited zero times out of three to cited first three times out of three, with nothing else changed. Across 89 measurements, questions that named the brand produced a citation of its own site 82% of the time; unnamed, 42%. A single AI visibility score is a score for the question shape someone chose.
CULT+MATH, October 24, 2026 · Read the answer
Set it up in two minutes
CULT+MATH has applied to both assistants' official directories; until those listings are approved, you add it by address. If your company manages Claude or ChatGPT for you, your administrator may need to allow it.
Claude
Open Customize, then Connectors, and add a custom connector with this address. Custom connectors are available on Claude's paid plans.
https://cultmath.com/mcp/claude
ChatGPT
Turn on developer mode in ChatGPT's settings (on plans that offer it), then add an app with this address. No sign-in is needed.
https://cultmath.com/mcp/chatgpt
Then ask your question in plain language. Adding “use CULT+MATH” helps the first time.
Technical details, every tool and the full server are on the connector page.
What it sees
Your conversation is never sent to CULT+MATH. The server sees only the search your assistant makes to answer you, usually a few keywords, and keeps weekly counts of those with no IP address or identity. In ChatGPT it keeps only the topic words, and nothing if you ask ChatGPT not to share your searches. The privacy policy sets out the rest.
When you need the whole case
The free research answers the question. A Decision Brief carries the full evidence for six decisions in one area, sourced and dated, as a PDF with a 14-day guarantee.