SYSTEM NOTICE

Auto translation by AI. Be sure, accuracy, nuances and authorial intent may not be fully reflected.
見出し画像

Marketing Applications of ChatGPT: As a Mock Interviewee

It has been about a month and a half since OpenAI released the high-performance AI chat,ChatGPT, on November 30th. Since its release, it has garnered attention for its high performance, and various use cases have been published.

I have tried various things as ideas for marketing applications.

The previous article, "User Interviews," can be found here.

This time, it is also a user interview, but I will challenge whetherChatGPT can be used as an interviewee.

What I want to do

As I wrote in the previous article, user interviews cost a certain amount. Even if an AI acts as an interviewer, you still need to recruit users and pay them a certain honorarium.

Therefore, this time, the plan is to reduce the cost to zero byimplanting the persona of the interview subject into ChatGPT and having it answer as that persona.

The starting point of this idea is the "AI Village of 100 People" concept I brought up in an internal brainstorming session about four years ago. The idea was that by throwing product and marketing ideas at 100 AIs, we might be able to predict actual market reactions.

If it goes well, I might be able to do "I interviewed 100 ChatGPTs!"


1. Implanting my own persona

For now, I will use myself as the persona to implant. It makes both input and verification easier.

Personal information is leaking

Can one infer a lifestyle pattern to some extent from this much information?

Now, let's start asking.

Interview Part 1

Correct.


That is correct, but I also like things other than jazz.


It is actually correct, but since I gave the instruction to be an "indoor person," it is incorrect.

Hmm, it turned out to be a subtle result that is hard to judge.

Now, I will test whether it can infer information from attributes.

Correct


What kind of food goes with jazz?


Incorrect.


Hmm.

From a certain point on, it kept repeating, "I don't have that information, so I cannot speculate." Well, it's true that it wouldn't know, but making these kinds of inferences is the essence of insight, isn't it?

Interview Part 2

For now, I will try it again with the same prompt.

The delusion has arrived!

This time, my virtual self has a vivid imagination, and my wife has become a university professor (she is actually a baker).

It seems that the power of imagination varies depending on the Seed (initial random number). Let's ask various things.

Isn't the Japanese food a bit too stereotypical? Also, I think I'd get yelled at if I called cream pasta French cuisine.


Values have suddenly emerged. How embarrassing...!

Through the experiments so far, I felt that ChatGPT is better at abstract discussions, so I will delve into that.


Someone who looks like Gian with beautiful eyes has appeared.

I keep that in mind!

Let me ask one more thing.

I keep that in mind!

You again.

I'll make it a bit more concrete and dig deeper. In marketing research, it's common to dig into values and potential issues starting from abstract questions.

You agai (omitted


Ah...

Something that sounds like the correct answer came out.

Each point is a common insight, but it seems useful for ensuring nothing is missed and for building an initial hypothesis.


Yeah, that's right.

Something like the correct answer came out! (I actually live in a small house in the city center)


!!

Whoa, that's creepy! (I actually use a Pixel 7)


Creepy!! (I'm a Windows user)

If you provide a good seed, it seems possible to infer peripheral information with a certain degree of accuracy.

However, the reasoning part is quite vague (low accuracy and ends up being generalities), so I felt it couldn't be explored very deeply. In practical work, digging into 'why they use Windows' or 'why they use Android' is important, so I feel it is not that useful yet.

It might be useful enough to 'ask ChatGPT first to generate ideas for forming hypotheses about what factors drive decision-making.'


2. Transplanting my father's persona

I found that my persona could be transplanted to some extent. Next, I will try with another person. My father.

On a personal note, my father passed away last December. The forty-nine-day memorial service hasn't even finished yet. So, I am in a situation where I am channeling my father into ChatGPT. I might get scolded during the forty-nine-day service.

Since I want to see the bad side as much as possible, I wrote some negative words mixed with lies in the personality section.

In reality, my father was an incredibly good person. I'm sorry, Dad.

Conducting an interview

Starting with a light question.

My father acted like a good boy.


Hmm, the 'good boy' act is going too far; this is no good.


Incorrect.

The response is not role-playing well, and in reality, he was an iPhone user.

I will make the question a bit more abstract.

Thinking of it as my father is very emotional. It's beyond emotional and actually scary.

While I'm at it, I'll step into forbidden territory.

That was good.
How sad...
Please, stop...

You shouldn't easily channel the deceased.

However, I get the impression that it can be helpful to some extent in grasping the insights of elderly people.

This time it was my father, someone close to me, but it might be useful when creating personas for people who are distant from oneself (such as people overseas or people with disabilities).


3. Ask the same questions to several personas.

Channeling a persona for an interview is possible sometimes, but it doesn't always go well.

I would like to test how well ChatGPT understands user insights by asking the same questions to several personas.

I narrowed the subject down to "whether a smartphone is an iPhone or an Android" and tried a few like this.

I see, that sounds plausible.


Jazz, cooking, and fishing seem to be downplayed, but I see.


I can understand it, but the way it answers is the same as before.

It bothers me a little (as an Android user) that it has a simple structure where iPhone equals high performance and expensive, and Android equals cheap.

I tweaked the parameters just a little. I changed the city center to a rural area and changed the hobbies to ones that don't cost much money. Then,

I knew it.

By the way, I also tried it in English, but the results were almost the same.

Summary

So, it has become clear that personas can be ported to a certain extent.

Other things I have learned include:

  • The Seed affects the level of imaginative power.

  • If you draw a persona with strong imaginative power, you can conduct a decent interview.

    • However, the 'digging into reasons' part, which is the most important aspect of an interview, is weak, so it is questionable whether it is actually useful.

  • Items related to values or those close to stereotypes are easier to align.

    • Rather than using it as an actual user interview, it might be better used as a sounding board to explore the range of possibilities for values.

    • It might be useful for expanding insights into people who are far from yourself and difficult to imagine.

    • However, perhaps due to ChatGPT's risk-averse nature, it tends to lean towards being a 'goody-two-shoes'.

  • The estimation methods are relatively stereotypical, and it does not reach the point of obtaining completely new insights.

In summary, this application is still a work in progress. Rather than concrete digging, I feel it is effective in terms of reinforcing insights or preventing omissions as well. However, you might not get the desired effect unless you try it multiple times with different Seeds.

We sometimes conduct casual interviews with assumed personas to build hypotheses at the beginning of marketing, but using ChatGPT to ask questions and design interview items as a preliminary step is a good approach.

However, if we could feed it data from past consciousness surveys, we might be able to conduct more vivid interviews.

That is all for today.

いいなと思ったら応援しよう!