banner
我是老王

我是老王

我用尽全力,过着平凡的生活!
github
twitter
email
follow
discord user

Mom Test questioning techniques and improving AI response accuracy

The Mom Test#

The Mom Test is a framework proposed by Rob Fitzpatrick for evaluating startup ideas and obtaining real customer feedback.

image

Its core idea is: “You cannot get real feedback by asking others (even your mom) if they like your startup idea, because people often say 'nice things' instead of 'true things' out of politeness or to avoid conflict.” The purpose of this test is to teach you how to ask the right questions to bypass this "social politeness," uncovering users' real behaviors, pain points, and needs, thus avoiding wasting time and money on wrong ideas.

The core of the methodology can be summarized in three basic principles:

Principle 1: Touch on their life, not just your idea#

Don’t immediately describe your idea; this will only put the other person in a "judge" role, where they will react to your idea (usually encouragingly) rather than revealing their true situation.

  • Wrong example: “I have an idea for an app that does XX, what do you think?”
  • Right example: “How do you usually deal with ‘a certain problem’?” “Can you tell me about the last time you encountered ‘a certain problem’?”

By focusing on “their past experiences, behaviors, and pain points,” you can obtain objective, real data.

Principle 2: Ask about past specific facts and behaviors, not vague future opinions and assumptions#

People's opinions and future commitments (“I might use it”) are cheap and unreliable. Specific past behaviors and facts are the reliable evidence.

  • Wrong example: “Would you use a product that solves this problem?”

This is a future assumption, and the answer is likely to be “yes.”

  • Right example: “How are you currently solving this problem?” “What tools or methods have you tried for this?”

These are questions about past facts.

Principle 3: Listen more to “bad news” and pain points, and talk less#

Valuable insights for startups often hide in users' frustrations, complaints, and dissatisfaction. Your goal is to discover and understand these pain points, as they are the reason for your product's existence.

[Case Study] Dialogue Comparison for iPad Recipe App#

Failed Mom Test (exposing ideas, receiving lies)

  • Son: “Mom, I have a business idea, can I tell you about it?” (exposing self, asking the other person not to hurt their feelings)
  • Mom: “Of course, dear.” (ready to lie to protect you)
  • Son: “You like your iPad, right? Would you buy an iPad recipe app?” (hypothetical question, leading to an answer)
  • Mom: “Um…” (internally: At my age, do I need any recipes?)
  • Son: “And it’s only $40, and you can share it with friends, plus there are videos from your favorite celebrity chefs!” (continuously selling until the other person says “yes”)
  • Mom: “Oh, that sounds great, dear.” (non-committal praise given to make you happy)
  • Son: “Great! Thanks, Mom!” (gaining false validation, heading towards failure)

Correct Mom Test (focusing on life, obtaining truth)

  • Son: “Mom, how’s the iPad been for you lately?”
  • Mom: “Oh, I love it! I use it every day.”
  • Son: “What was the last thing you did with it?” (asking about specific facts)
  • Mom: “I’m looking for places to stay for our trip.”
  • Son: “By the way, I noticed you have a few new cookbooks on your shelf, where did they come from?” (shifting to a relevant area, but still about her life)
  • Mom: “Oh, they were all gifts I received for Christmas. Honestly, at my age, I don’t need new recipes!”
  • Son: “When was the last time you bought a cookbook for yourself?” (anchoring past behavior)
  • Mom: “Speaking of which, I bought a vegetarian cookbook three months ago. Your dad wants to eat healthier.”

So what about in Prompts?#

Applying the questioning techniques of the Mom Test to Prompts will surely yield great results:

  • Eliminate illusions: Ask questions based on users' experiences, behaviors, and other real occurrences.
  • No flattery: Listen to bad news and dissatisfaction.

1. Avoid asking for opinions, instead ask for evidence#

Wrong example:

  • Do you think this plan is good?
  • Is this design reasonable?

Analysis: The model tends to give more vague compliments.

Right example:

  • Please give me 3 specific counterexamples that illustrate scenarios where this plan might fail.
  • Based on known facts/data, list the limitations this design might encounter.

2. Avoid future assumptions, ask about past performance#

Wrong example:

  • If you encounter problem X, how would you handle it?

Analysis: The model output lacks basis and may produce illusions.

Correct example:

  • Please list the cases of problem X that you have learned from the training corpus.
  • How has X been solved in past research or historical records?

Analysis: Output based on “facts that have occurred.”

3. Avoid vagueness, ask for specifics#

Wrong example:

  • Help me optimize this copy.
  • Give me some improvement suggestions.

Correct example:

  • Please rewrite this copy into 3 versions: ① for investors, ② for engineers, ③ for general users, with each version being 100 words.
  • Please point out which parts of the copy are vague and provide clearer alternative expressions.

4. Drive by behavior rather than attitude#

Wrong example:

  • If you were a user, would you like this product?

Correct example:

  • Assuming you are the target user, please simulate an actual usage process and write down the steps you would click, input, and hesitate.

5. Validate rather than seek praise#

Wrong example:

  • Can you confirm that my logic is correct?

Correct example:

  • Please check my logic, find at least one possible error, and explain why.
  • If you must refute me, please provide 3 points of argument from the opposing perspective.

Summary#

Mastering the five prompt techniques based on the Mom Test effectively reduces the model's output of illusions and flattery. If you find it a bit difficult to remember the five techniques, you can also remember one keyword: “brutal.” The usage is simple; when having professional discussions with AI, including this keyword will make the output sharper and effectively reduce flattery. For example:

Here is the description of my website's features,
Analyze its prospects,
How to actively validate user needs?
Please evaluate it in the most brutal way.

[Website feature description]
Loading...
Ownership of this post data is guaranteed by blockchain and smart contracts to the creator alone.