How good is AI at BS? This isn’t a memo about Large Language Models and their tendency to hallucinate. I’m talking about a different kind of BS. Behavioral Science.
Today’s post brings together two threads: behavioral science’s insights into how our brains actually make decisions, and where AI fits (and doesn’t) in our creative process.
The Question: Can AI helpCan AI help in the editing process by analyzing whether a draft makes good use of behavioral science principles or violates them?
To find out, I wrote a one-page message that intentionally uses behavioral science principles well in some sections and poorly in others.
The Behavioral Science Test Draft
October 12, 2026
Dear Democracy Defender,
Let me put this bluntly. Everything you’ve done so far to help win the November 3rd elections could all be for nought unless you take the final step I’m about to request.
We have to contact a lot of people in these final weeks of the campaign and I’m urging you to play a critical role which I will detail in a moment.
I’m a realist. I know most of the people receiving this message won’t do what I’m asking. But you’re different.
You have proven time and time again that you are a dedicated defender of democracy. You’re someone with strong beliefs about many critical issues and a deep commitment to acting on your beliefs.
You’ve done so much to put our campaign within striking distance of victory. Now isn’t the time for you to back off.
I’m going to be specific about what I am asking of you. I’m not asking for your financial support. What I need now is your time. I’m urging you to commit one hour every day between now and Election Day making urgent get-out-the-vote calls to people who are still on the fence about whether to vote and who to vote for.
They are the people who will decide the outcome of this election and you personally reaching out might be the thing that gets them off the sideline and onto our side.
I know I’m asking a lot of you. But I wouldn’t be asking if I didn’t think it might really help.
All you have to do is call this number right now – 1.234.567.1234. One of our voter contact specialists will provide you with a script for your calls and a list of voters who our analysis suggests might well be persuaded by your call.
I’m counting on you and I know you won’t let me down . Prove to me you’re not the kind of person who backs off at make-or-break moments like this. I really appreciate what I am sure will be your positive response.
On to victory,
Ima Candidate
Prompting an AI Behavioral
Science Review
Next, I prompted Claude and ChatGPT to review the draft in strictly behavioral science terms.
Both produced similar results. I’ll focus on ChatGPT’s, which were a bit sharper. Here’s a sampling of the ways it found behavioral science positively at work and the places where behavioral science suggests the draft is weakening itself. The following quotes ChatGPT verbatim.
Three Ways Behavioral
Science Strengthens the Draft
#1: “Everything you’ve done so far … could all be for nought unless you take the final step …” Loss aversion + sunk-effort-salience.
This is one of the strongest behavioral moves in the draft. People generally work harder to avoid losing something they believe they have already earned than to acquire an equivalent new gain. By framing the requested action as necessary to protect prior effort, the letter makes inaction feel like a loss rather than a neutral choice. It also activates a related “completion” impulse: I’ve already come this far; I shouldn’t stop now.
#2: “But you’re different.” – Identity-based motivation.
The letter moves quickly from describing what others will do to defining the recipient as a different kind of person. Behavioral research consistently shows that identity can be a more powerful motivator than abstract argument. The question subtly shifts from “Do I want to make calls?” to “What would someone like me do?”
#3: “We’ll provide you with a script and a list …” – Friction reduction
The draft anticipates two likely barriers: What do I say? And Whom do I call? Removing those burdens reduces cognitive effort and choice friction. In behavioral terms, this is one of the most useful parts of the letter because making a desired behavior easier can outperform trying to make people want it more.
Three Ways Behavioral Science Suggests
Weaknesses in the Draft
#1: The ask is behaviorally very large: “one hour every day between now and Election Day.” – Friction + present bias.
This is probably the biggest behavioral weakness. The reader is being asked not for a single act, but for a recurring commitment with substantial time cost. Present bias makes immediate inconvenience loom larger than future civic benefit. The larger the initial behavioral hurdle, the greater the likelihood of doing nothing at all.
Behavioral science would generally predict greater uptake from a much smaller initial commitment followed by escalation.
#2: “Most of the people receiving this message won’t do what I’m asking.” – Negative social proof.
This may be the clearest behavioral-science error in the draft. Social proof tells people what people like them actually do. Saying that most recipients will not act normalizes nonparticipation. The next sentence – “But you’re different.” – tries to counteract this with identity, but the message has already established an undesirable descriptive norm: Most people don’t do this.
#3: “Prove to me you’re not the kind of person who backs off …” – Psychological reactance.
This is risky behaviorally. When people perceive their autonomy is being threatened, they can resist precisely because they feel manipulated. The language moves beyond identity reinforcement into an implied challenge to the reader’s character. That may motivate some highly committed supporters but provoke reactance from others.
The Overarching Behavioral Science
Tension ChatGPT Identified
Most impressively, ChatGPT points out the most critical flaw that I built into the test draft:
The most important behavioral-science tension in the draft is interesting: its identity architecture is quite strong, while its action architecture is comparatively weak. It does a good job of making the reader think, “I’m the kind of person who should do this.” But then it makes doing it feel unusually demanding. Behavioral science would argue that the winning combination is strong identity motivation + very low friction + a small first commitment + visible social proof.
Adding a New Step to My Proposed
Human First, AI Forward Approach
to the Creative Process
Now, back to me writing, not ChatGPT.
So how good is AI at BS? You should, of course, decide for yourself. But my own verdict is that the AI audit did a good job identifying and explaining the behavioral science issues I built into the draft. For that reason, I’m amending my recommended approach to AI in the writing process.
Here’s a quick recap of the now four-step approach:
We’re all striving to find the right way to incorporate AI into our creative process. The key is finding the places where it really helps while maintaining a firm commitment to a human-centered, human first approach.
So, here’s my advice: On your next project, try the four-step process laid out above. And pay special attention to the third step – the behavioral science audit. I’m willing to bet you will find it gives you a productive, new angle of vision on ways to strengthen the persuasiveness of what you’ve created.
