How HiveMind works

Plain answers about what you are buying.

What the crowd is

Thousands of AI people.
Built to match U.S. Census numbers for age, income, region, education and household.
Each one answers in their own words, in small batches (we measure and correct for copying inside a batch).
Fast and cheap, so you can test many ideas.

Who they are: real Census data for the exact place you pick (a city, county or ZIP). Income, schooling, kids and home ownership follow their age, the way they do in real life.
How they think: a personality (the Big Five), how much they hate losing money, how much they doubt ads, how stuck they are in habits, and how much they follow the crowd. All based on published research averages, with real person-to-person spread.
What they care about: two core values (like safety, family, or excitement).
What is going on right now: money this month, time, a recent life event, today's mood, and where they pay attention.
Fairness rule: background shapes daily life only. It never decides an opinion.

People answer on a 5-step scale, then we apply the standard "say-do" discount, because people who say "probably" often never buy.
Half the crowd is asked "buy it?" and half "skip it or buy it?", with the answers in reverse order. The gap shows how much wording alone moves the result.
We ask about a product that does not exist. Every yes to it is pure bias, and we report that number.
Groups only count as different if they pass a false-alarm filter across every comparison we ran.

Picking between options. Which headline, ad, offer or funnel to test first with real money.
Finding the reasons. Why people say no, in their words.
Finding the leaks. Which step of a funnel loses the most people.
Finding the groups. Which ages, incomes or job types lean in.

Use it to decide what to test with real people. Then test it for real.

Reports show a margin, like ±3%. That only tells you how much the answer would move if we asked a new AI crowd of the same size.

It does not tell you how close the AI crowd is to real people. That is what the scorecard below is for.

Bigger crowds help you split results into more groups (by age, by income, by job) and still read them.

Would they buy: yes, maybe or no, plus the main reason.
Message, ad and content tests: each person picks the one that moves them most and the one that turns them off. We shuffle the order for every small group so the first option does not win just for being first.
How they'd find you: where they'd look first, what they'd type into search, and what kind of content stops their scroll there. Search words are AI guesses, so we check them against real search numbers.
Funnel test: people walk your steps one at a time and leave when something bugs them. AI people click far more than real people. Compare funnels and find the leakiest step; don't read it as a real click rate.

We asked the AI crowd questions where the real answer is already known from Pew Research Center surveys.

The first calibration run is in progress. Results will appear here.