Ethical gut-checks

Three fast tests for when something is legal but still feels wrong.

The idea

"It's allowed" is a low bar. Some decisions clear every rule and still land you in trouble — with users, with the press, with yourself. Before you ship a borderline call, run it through three quick tests. Each one catches a different kind of blind spot, so a decision has to clear all three, not just one.

And when you raise the concern out loud, voice it as a question, not an accusation. "How would we explain this on the front page?" invites the room to think. "This is unethical" makes them defend.

How it works

Run the decision through each gate in turn. A gate you clear is a class of trouble you've closed off; a gate you fail names the specific harm to fix.

1. Front-page test. Picture the decision, and your reasoning, printed as tomorrow's headline with your name on it. If you'd wince, the problem is usually secrecy — you're relying on people not finding out.
2. Stakeholder-harm scan. List who bears the downside. For each, ask two things: how bad is it, and did they agree to it? Harm that lands on people who didn't consent is the sharpest signal of "legal but wrong."
3. Letter vs spirit. Every rule has a text and a purpose. When the text permits what the purpose was written to prevent, you're exploiting a loophole. Privacy terms exist to protect people; a clause that lets you profile them anyway satisfies the letter and betrays the spirit.
4. Voice it as a question. Surface the concern without indicting anyone: "How would we explain this to the shopper afterward?" beats "you're being creepy." Questions keep the room problem-solving instead of defending.

When to use it

Reach for it when…	Keep in mind
The lawyers said yes but your gut said pause.	Passing is a floor, not a verdict — these tests flag risk, they don't grant permission.
A choice quietly moves harm onto people who never agreed.	Clearing one gate isn't enough; a decision has to clear all three.
You're about to exploit a technicality or a buried clause.	Fast checks, not a substitute for real ethics or compliance review on high-stakes calls.

Watch out for

• Treating "legal" as the finish line. Terms of service are written to protect the company, not to answer "is this right?" The front-page test exists precisely because compliance can't.
• Counting only visible harm. The loudest stakeholder isn't always the most harmed. Non-consented harm to quiet users outweighs a small, agreed-to cost that someone signed up for.
• Fake consent. A pre-ticked box or a clause on page 40 isn't consent. If the harmed party wouldn't have agreed had you asked plainly, treat it as no consent.
• Winning the letter to lose the spirit. "Technically allowed" is the phrase that shows up in every post-mortem. If you're reaching for it, the spirit gate is probably already failing.
• Accusing instead of asking. Framing a concern as a character charge ("that's dishonest") gets you a defensive room and no fix. The question form keeps everyone on the same side of the table.

Worked example

Interviewer: "Marketing wants to email lapsed users a fake 'your data will be deleted in 48 hours' warning to boost re-engagement. It's not illegal. What do you do?"

Run the gates. Front page: "Company scared its own users with a fake deletion threat" — you'd wince, so it fails on secrecy and deception. Harm scan: the downside lands on trusting users who never consented to being manipulated; that's the heaviest, least-consented harm on the board. Letter vs spirit: no statute forbids it, but the whole point of honest communication norms is not to trick people — letter passes, spirit fails. Then voice it as a question in the room: "If a user later learned the warning was fake, how would we explain it?" That reframes it from "who's the bad guy" to "what's our story," and the answer is usually a redesign: send a genuine win-back offer instead.

Check yourself

A team ships a dark-pattern cancel flow: legal, and it does lift retention. Which read is strongest?

It cleared legal review, so the ethical question is settled. The harm (frustrated users who can't leave) lands on people who never agreed to it — it fails the harm scan. Retention went up, so on balance it's fine.