Use ChatGPT Prompts and Priming to Create Personas - Be More Persuasive with Your Communications.
HTML-код
- Опубликовано: 6 июн 2023
- This ChatGPT Priming approach helps you create PERSONAS for the people with whom you most often communicate: work colleagues, bosses, employees, customers...
Once you define a persona, chatgpt can tailor your messages to be more persuasive and engaging for your target audience.
ChatGPT Resources:
Not another newsletter!! You've got plenty of those... This email won't discuss the news and it will only show up when I've got something very worthwhile to share:
lawtonsolutions.com/subscribe/
You can opt-out anytime with one click and it's totally free.
STEP 1: DEFINE A PERSONA
I am going to provide you with definitions of components that you will use for creating a persona. We will call this information = [Persona]. I want you to store these components for later use.
[Persona] =
“Demographics: These include age, gender, income, education, occupation, marital status, and family size. Demographics provide a basic understanding of who the persona is and helps in segmenting the target audience.
Psychographics: This covers the individual's values, beliefs, interests, lifestyle, and personality traits. Psychographics provide insights into the motivations and attitudes that drive decision-making processes.
Behaviors: The way users interact with a product or service, their purchasing habits, and preferred communication channels. This information helps tailor marketing strategies to be more effective and personalized.
Goals and aspirations: Identifying the user's personal and professional goals, as well as their aspirations, enables marketers to align their product or service with the user's desires, creating more compelling messaging.
Pain points and challenges: Understanding the problems and obstacles that users face allows marketers to position their product or service as a solution to these issues, ultimately creating value and building trust.
Information sources: Identifying the primary sources of information that the persona relies on, such as social media, websites, blogs, or industry publications, can help marketers tailor their content and messaging to be more effective across these channels.”
Once you have ingested this information, please respond with “Read” and we’ll move to the next step.
STEP 2: TELL CHAT YOU'RE GOING TO FEED IT WITH INFORMATIONAL SOURCES (WHICH IT WILL LATER USE TO POPULATE THE PERSONA)
I am going to provide you with informational sources that I will define via the following = [information]. I want you to remember this information for later use. Then we will use these informational sources to create specific content that I request. Please confirm you understand these instructions by responding, “Read,” and we will proceed to the next step.
STEP 3: PROVIDE CHAT WITH PUBLICLY AVAILABLE INFORMATION (SOCIAL MEDIA PROFILE, RESUME, SPEECHES, BLOGS, ETC)
Here is an information source for you to ingest for later use = [SOURCE NAME]
[SOURCE NAME] = Provide the source data
Please confirm you have ingested this information by responding, “Read,” and we will proceed to the next step.
REPEAT STEP 3 WITH AS MANY RELEVANT INFORMATION SOURCES AS YOU'D LIKE
STEP 4: DEFINE CHAT GPT'S ROLE
Next, I want you to please act as a professional therapist with more than two decades of experience studying, analyzing and publishing on human personality traits and characteristics. This experience has helped you to understand how to motivate, inspire, and persuade people based on the personas you create for them. For the next set of tasks, this is your role and how you should act.
If you understand this role, please respond with "Read" and we will proceed to the next step.
STEP 5: ASK CHAT GPT TO POPULATE THE PERSONA BASED ON THE INFORMATIONAL SOURCES
Based on your role, please use the information [SOURCE NAME 1... 2... 3...] to inform the persona components from [Persona] that you have previously stored. Please create an individualized persona using the format:
“Demographics:
Psychographics:
Behaviors:
Goals and aspirations:
Pain points and challenges:
Information sources:”
STEP 6: ASK CHAT GPT TO WRITE YOU A PERSUASIVE EMAIL / TEXT / REQUEST TO THIS PERSON IN A NEW CHAT
Copy the persona output provided by ChatGPT in Step 5. Then start a new chat.
IN THE NEW CHAT WRITE:
"I am going to provide you with a persona for PERSON, with whom I want to send communications. I want you to remember this information for later use. Then we will use this information to create specific content that I request.
[PERSONA] = PASTE THE PERSONA FROM THE PRIOR CHAT HERE"
Please confirm you understand these instructions by responding, “Read,” and we will proceed to the next step.
After Chat responds with "Read, " write:
"Please act as a highly persuasive and deeply experienced writer. Based on PERSONA, please write me a {FRIENDLY}, {PROFESSIONAL}, {FUNNY} email to this PERSONA about TOPIC...
NOW THAT YOU HAVE A NEW CHAT WITH THIS SPECIFIC PERSONA, USE THE SAME CHAT FOR ONGOING COMMUNICATIONS WITH THEM.
Your channel is underrated. You should have subscribers in thousands.
I just binge watched all 4 of your videos. They were immensely enlightening.
Yours is the most valuable RUclips channels I subscribed recently.
Eagerly waiting for more videos on the same topic.
Wow! That’s very kind of you. I really appreciate it. I’m working on more!
Just added a new video that I think you’ll like. Hope it continues to be valuable content!
Agreed!!
Thank you. You're the first person I've seen use variables and I appreciate your explanation
Much appreciated. Glad I could help!
You only have four videos, and I love your channel already. You've also helped debunk a lot of these fly-by-night SAAS CHATGPT apps as just basic prompts in fancy wrappers.. Thank you!
Wow, thank you! Glad to help.
I just added a new one for you on a different type of prompt methodology. Hope you find it useful.
@@LawtonSolutions Thanks buddy, watching it now..
You're amazing. I subscribed. Thanks! 👍🏾
Thanks so much! Really appreciate the kind words.
Just what I was looking for. Very useful for creating a paper from multiple complex sources.
Glad to hear it! Sounds like the perfect use case.
Wonderfully helpful information. Thank you for sharing all of these great ChatGPT tips. Highly recommended. #gptprompt #chatgpt #GTP4 #AI #prompttutorial
Thank you so much. Happy to help. Just curious… What’s your typical use case for ChatGPT?
Your channel is the best on the subject of Prompt Engineering that I have found so far, and I have watched a lot of videos on this topic recently!! Congratulations and please keep up the good work and keep these videos coming! You have a new subscriber here!! 😀
Super cool. Thanks so much for your thoughtful words. I’m working on a few more videos for ya!
Well done! I'm new to this, but based on my research thus far, the principles you outline here, and elsewhere, represent an elegantly distilled explication of the (current) STOA of prompt engineering. Subbed!
Thank you for the kind words. I really appreciate it and am glad to hear the prompt was helpful!
Highly informative and well-designed videos. Can't wait to apply this approach! Thank you.
Thanks for the great feedback. It really means a lot to me. I’d love to hear how it goes. Check back and let me know!
Wow, top notch stuff on this channel. This is a savage play on communication.
Very nice of you to say. It's a strange new world we're living in. Just hope I can keep up!
Hey Lawton, please do at least 1 video every week. You can make this content in your sleep. If you are too busy I understand.
Thanks for the motivation. Working on it! Also working on sleep. :)
Hey Lawton, you a bad Man. This is Heavyweight AI master class level videos. Best I have seen. I am going to check out your site. Hopefully you have AI training courses. Please keep making the videos.
Wow! Thanks for the kind words. So happy to hear it helped!
Super Awesome! Loved the content, It was Super helpful. Thank you so much
Thank you. Glad it was helpful! What’s your top use case for ChatGPT?
Fantastic ChatGPT video! This explanation of ChatGPT was extremely clear and insightful!
Wonderful to hear. It’s always nice to know that I’ve been helpful with AI information.
Great content, subscribed!!
Welcome aboard! So nice to have your confidence. I’ll do my best to bring more helpful content on ChatGPT.
You could put this all into a single prompt and either post the info in that prompt or post the extra info in the second. Sooner or later ChatGPT will begin forgetting your format or what the specific variables you want and require you to start over again.
Quite true, there's definitely a limit to what ChatGPT can consume. I do like using variables in case I want to pick and choose the inputs from which I'd like OpenAI to respond. Let's hope, someday soon, Chat will increase it's token limits and we can throw everything into one prompt!
@@LawtonSolutions Here's hoping. Right there with you. I like to give it reminders though to help it maintain it's memory longer. I just hate when it suddenly decides it wasn't to do things differently and you have to argue with it to try to go back to the original way that was more efficient, and sometimes still refuses.
@@tacticaldesigns8794 Yeah, it can be fussy like that..lol
Amazing. Question: Is there anyway to drop a link in to have it reference? Or do I always have to copy/paste from a source? Like you mentioned a LinkedIn profile for example. Can I just tell it to check their page? Want to make sure I'm understanding ChatGPTs limitations (if any!). Thanks again for all your excellent content!
The quick answer is... it depends. LinkedIn URLs don't typically work as they are considered private. However, most websites (that don't require a login) can be referenced. I like to separately call out the website while I'm priming ChatGPT and then ask Chat to specifically process the information in the link prior to moving to the next step.
Another great, easy to follow, step-by-step video! Thanks so much!
Do you think it’s possible to pre-train an LLM to understand a body of company info relating to policy etc, plus their brand values/behaviours and so on, plus the writing/speaking style of, say the CEO, and have it respond to questions which are based on the pre-trained input? This is an idea I’m working on at the moment and I’d be interested in your take on it.
Thanks for the kind words. Yes, I absolutely think you can prime Chat GPT or most of the major LLMs (Claude, Bard, Bing) to do exactly that. Follow the same pattern in the video. Create variables or informational sources for each of the categories: Company information - history, product/services, number of employees, mission and vision; brand values - might be tricky to find but some companies share it; operational policies - Again might be difficult to find; CEO communication style: speeches, letters, profile. Then, once you have all of that information primed, ask ChatGPT to respond to the specific questions. Let me know if you have further questions...
@@LawtonSolutions just to be clear, that this would be a process I would be testing with the client’s permission in order to assist them. I’m not looking to deep-fake anything. That would be very much against my ethics. 👍🏻
@@ocuix Yes, of course! I assumed you were talking about a publicly traded company, of which most of that information would be available in annual statements and such.
awesome. thanks. Been at this a while and you have great organization. DO you run into memory issues after a while? can you tell me more about how you deal with that? i feel like after I go for a while in a chat, especially in the pitiful situation when you keep going back and forth with prompts that aren't quite getting what you want... What are all the ways to make sure there is persistent memory? What to do when you encroach on the 4096 token limit. limit of the memory? (in ChatGP...without fine-tuning your own model, that is) Cheers
Appreciate the kind words!
The quick answer is yes, memory issue will eventually surface if you keep piling on more and more variables. I wish I had a concrete value for the limit... I've tested it a few times and it's difficult to see when Chat GPT's memory gets maxed out as it will continue to provide solid responses with slight shifts towards your newest inputs.
One of the ways I navigate this issue is to create different chats for various input elements. For instance, in the video I defined the categories of the persona in a different string. Then I moved those persona categories (and their definition) into this Chat. I try to keep the chat focussed on one theme at a time.
In case I run into issues with very long data sets, I'll head over to Anthropic Claude. I'll ask Claude to summarize the lengthy material and then drop that summary into OpenAI. Anthropic can do 100,000 tokens so it's great for large data sets.
I hope that helps. My guess is OpenAI will grow beyond 4096 tokens in future versions. Though I'm hearing version 5 is several months away...
@@LawtonSolutions no doubt that the ceiling will break, and yes, Claude is a recent alternative if you have access. I keep copies in other docs, too, to "reboot" personas, for instance... like when they don't sound like themselves anymore, and you ask them a basic question and they give you the whiny standard "I'm just an LLM with... blah blah limitations"... I've always wanted a good 'control-z' for chatGPT, too, but my workarounds (when they get serious) tend toward API Calls form Python programs I write, the advantage being the ability to actually read and write data files to a hard drive or database, rather than to hope it remembers the [variables] primed prior. Ahh-mazing stuff - Gotta love it, right!?
There's an interesting line between editing a chat string (yet again) to produce a new result based on slightly different parameters ...and punting to start a new one with some of the key elements you liked from the previous. My biggest frustration is not know exactly when I've crossed that line.
Definitely agree in wishing OpenAI would incorporate an undo button (control-z) because I also feel like I've lost valuable memory each time I edit an input (once I'm deep into a lengthy string).
I continue to hear people talk about using python in the Playground. [Sure wish I knew how to code.] So I think you're on the right path until a bigger set of tokens is possible with Chat.
Where have you read that it can keep in memory so many steps? Are you sure about this ? Is there any technical documentation for this ?
Thanks for the thoughtful questions and for your patience. There are, of course, still token limitations on the dialogue. I experimented with this approach a couple of months ago and began testing it independently (primarily as a way to create flexibility throughout a conversation). As I used it more, however, I noticed far better results from GPT (especially for long-form content). If you're interested in digging deeper, here's a nice research paper specific to process supervision to improve performance:
openai.com/research/improving-mathematical-reasoning-with-process-supervision
Keep in mind, the approach I describe is more about enabling variability so the user can make changes to elements in the dialogue without needing to start over. If you use fewer words per dialogue, you can get consistent results for longer chat strings. Unfortunately, there remains limits to how much detail you can provide in total. Hoping the token limits increase soon!
Nice content. Can you teach prompt engineering as well? I would like to learn how to create useful and professional prompts
Thank you so much for the kind words. I am working on a few new videos. One will be discussing prompt templates within Open AI and using better prompts in a Chat. I hope to turn it around soon. These things take me way too long... So thanks in advance for your patience.
@@LawtonSolutions We're looking forward too it.
This is both creepy as shit and really really cool 😂😂😂 thanks
Edit: I think it just feels more creepy to me for whatever reason when it’s not. An actual therapist or something would understand all this stuff normally it’s just funny to me
Ya... I feel the same way (me furiously checking everything I've ever made public on the internet)... ha!
All of these to be taken out, none of this to be allowed.
Thank you for your thoughts.