ChatGPT is often marcom professionals used as a general search engine. And as a kind of question and answer that has answers to everything. Do we need a new online strategy for organization X? Let’s ask that question 1-on-1 to ChatGPT.
The way of prompting also shows many similarities with how we Google. This is also called the horse carriage syndrome: new techniques are seen in the light of old techniques. The output also seems to be judged on that basis. The first result is often accepted – after all, you have no interaction with Google.
Furthermore, ChatGPT was used to devise and execute (steps of) content creation, but also to (re)write prompts.
For the smaller, concrete tasks of content creation, ChatGPT is more contextualized and feedback-driven. Prompts are also more likely to be structured better here.
In the more complex tasks, such as strategy and creative concepts, we see that ‘big’ questions are asked to ChatGPT. A question is not split into substeps and there is little interaction. AI is fed little with feedback. In these types of assignments, the ‘conversation’ with AI also stops relatively quickly. The answer is easily accepted or the prompter quickly gives up.
Understanding how an AI model works
What works now, how do you prompt well? The biggest key to this is understanding how an AI model works. What data is it based on? And why does the model sometimes take a certain path when prompted? By better understanding how the model works , you can also better determine the content of your prompt.
The bad news: you won’t learn how the model works and namibia email list 137403 contact leads how it responds to prompts in your field in one day. The good news: we can help you get more out of an LLM like ChatGPT. Based on our research, we have put together a list of tips for you that you can apply immediately to prompt more effectively.
5 Tips for Effective Prompting
Pay attention to the structure of your prompt. Make sure you provide enough context to get AI on the right track right away. It helps to shape your prompt using a framework, such as PREPARE or CRISPE. Unfortunately, we have not yet found evidence of which framework works well (in which situation).
It helps to give examples. Are you having ChatGPT the impact of color on customer behavior: lessons from bol, starbucks & mcdonald’s write a blog? Then give a few examples of blogs that are similar in structure and writing style to where you want to go. A kind of ‘priming’ of AI.
Use prompt chaining : not one big question at a time. ao lists Take AI through the thinking and doing steps as you would go through a big task yourself.
Writing style matters ! AI delivers the best output when you engage in a conversational style, as you would with a colleague. This also means being polite: research has shown that the output AI delivers is much worse when you use rude language. And also: emphasizing what something means to you or your work delivers better output .
Give the output some more control by using parameters, such as adding ‘Temperature’ or ‘top P’. This already gives you an indication of how conservative and how likely the output is that ChatGPT comes up with. For example, if you enter ‘temperature = 0’ in the prompt for writing a blog, a blog with a conservative character will roll out.