Psychology Of AI
People Built It. People Trust It.
AI can look like a black box. Look closer and there's a person at each end. At the front, the teams who build the models and write down what they want them to value. At the back, your buyers and your team, who talk to AI like it's a person, then trust it or quit on it. We study the minds at both ends, then build your marketing and your AI systems around what we find.
Trusted And Recommended By the AI your buyers ask, and by the people who use yours.
Why People Talk To Machines Like Friends
Your buyers know AI isn't a person. They treat it like someone anyway, and they forgive it far less than they'd forgive a colleague. 3 findings explain the pattern, and each changes how we build.
They Give It A Mind
Anthropomorphism · Epley, Waytz and Cacioppo, 2007
People read humanlike intentions into things that aren't human, more so when they're trying to predict what the thing will do and when they're short on human connection. A chatbot that answers in full sentences checks both boxes. So we decide on purpose what character your AI shows customers, before your customers decide it for you.
They Mind Its Manners
The media equation · Reeves and Nass, 1996
People apply social rules to computers. In 1 test of it (Nass, Moon and Carney, 1999), people rated a computer more kindly when that same computer asked for the rating than when a different computer asked, the way you'd soften feedback to someone's face. So your AI's tone, patience and handoff to a human count the way an employee's would.
They Forgive People Faster
Algorithm aversion · Dietvorst, Simmons and Massey, 2015
After watching a forecasting algorithm make mistakes, people picked a human forecaster instead, even when the algorithm had beaten the human. So an AI system's early mistakes cost more trust than a new hire's, and we start it on jobs it can get right before we hand it harder jobs.
Read The Mind Behind The Model
Every AI model has makers, and the biggest makers publish what they want their models to value. The company behind ChatGPT publishes a model spec that tells its models not to flatter people, to say when they're unsure, and ranks a confident right answer above a hedged right answer. Google says large language models are probabilistic, always producing new and different responses. Put those together and you have the opponent's whole mindset: the exact words change every time, but what the makers reward doesn't. So we stop chasing any single answer and work on what holds steady: true, specific claims a model can repeat without hedging, and no hype it would have to talk its way around.
ChatGPT
Its maker's spec says the facts in an answer shouldn't shift with how a question is phrased, so we work on your facts, not on clever prompts.
Gemini
Google calls these models probabilistic, so we judge how you show up over many runs, never from a single reply.
Google's AI Features
Google says they need no special files, markup or structured data, so we skip the gimmicks and fix the page.
Your Own AI Systems
Built by people too, so we write down what yours should value before it says a word to a customer.
Why We Let Your Team Overrule The Machine
You'd think an AI system works best when nobody touches it. The research points the other way. In a set of forecasting studies, people were far more willing to use an imperfect algorithm when they were allowed to adjust its answers, even when the adjustment allowed was tiny, and their forecasts improved as a result (Dietvorst, Simmons and Massey, 2018). They didn't need much control. They needed some. So when we build an AI system into your business, we give the people using it a small, visible way to correct it.
Hand your team a little control over the machine, and they'll actually use it.
How We Win The Mental Game
-
Read The Opponent
What the makers of the AI tools your buyers use say they reward, in their own published words.
-
Find The Nerves
Where your buyers trust AI, where they doubt it, and what they ask it before they ever ask you.
-
Say It With Confidence
Claims a model can check and repeat without hedging, and nothing it would have to walk back.
-
Build In Composure
AI systems that start on jobs they can get right, show their work, and let your team correct them.
-
Read The Trend
AI answers shift from run to run, so the trend over weeks says more than any 1 reply.
Confidence Shows Up On The Scoreboard
Anybody can sound sure about AI. The score that matters is whether the machines recommend you and whether people keep using what gets built.
-
Recommended By Name
How often the AI assistants your buyers use put you forward when they ask who to hire.
-
Kept As Written
The share of an AI system's work your team accepts without redoing it.
-
Still Switched On
AI tools your team still uses months after launch, not just on demo day.
-
Hours Handed Back
Manual work the system takes off your team each week, counted, not guessed.
Who Keeps Your Head In The Game
- Jeremy Wayne Howell Reads Both Minds
Personally directs the work, from what the model makers reward to how your buyers and your team actually treat AI.
- Azra Hajdar Keeps You Focused
Your day-to-day contact, so every question about what the machines are saying gets answered between check-ins.
Fair Questions, Straight Answers
Why Does An AI Menu Start With Psychology?
Because there's a person at each end of every AI system. Somebody decided what the model should value, and somebody decides whether to believe what it says. Understand both and the machine in the middle gets a lot more predictable.
Are You Trying To Trick The AI?
No, and it wouldn't last. The makers publish what they reward, and honesty sits near the top: ChatGPT's maker tells its models not to lie and not to flatter. Tricks are what those rules are written against. We give the machine true, specific facts it can repeat with confidence.
Why Do AI Tools Give Different Answers To The Same Question?
Because they're built that way. Google describes large language models as probabilistic, always producing new and different responses. So 1 answer proves very little. We ask the same questions many times, across tools, and read the pattern.
Should Our Chatbot Pretend To Be A Person?
No. People already treat chatbots socially, so it doesn't need a disguise to feel natural. Even ChatGPT's maker says its assistant shouldn't pretend to be human. We give yours a clear identity, good manners and a fast route to a real person.
Why Won't Our Team Use The AI Tool We Bought?
Often because they watched it get something wrong early, and people give up on a machine after a mistake faster than they give up on a colleague. We restart it on narrow jobs it can get right, with a way for your team to correct it, and widen its job as trust comes back.
The Next Play
AEO And GEO
Ask 3 AI assistants about your company and you may hear 3 different stories. Here's how to tighten the plot.
Door #2AI Agents And Automation
Your team will forgive a new hire's mistake long before a machine's. That changes how we build.
Door #3Psychology-First Marketing
The algorithm is only part of the story. The rest has a pulse, a budget and a deadline.
What Does The Machine Think Of You?
Bring us the questions your buyers ask AI, and any AI tool your team stopped using. Let's talk about the people at both ends.
Get a Free Audit