What is ChatGPT?
According to ChatGPT itself: “ChatGPT is an AI-powered language generation model developed by OpenAI. It is capable of generating human-like text based on a prompt or context given to it. It uses deep learning techniques to understand the context and generate a coherent response. ChatGPT has been trained on a massive corpus of text from the internet and can generate text on a wide range of topics, including answering questions, providing explanations, and generating creative writing. It has been used in various applications, including chatbots, language translation, and question answering.”
This article will explore some potential use cases for this exciting new technology in market research, along with some possible pitfalls as well.
Let’s investigate some of the potential applications of ChatGPT in quantitative, qualitative, and desk research.
ChatGPT for Quantitative Research
Questionnaires: One potential use-case for ChatGPT is in the development of questionnaires. Providing the model with a list of research objectives allows it to produce a series of appropriate research questions, written concisely, clearly, and in a neutral voice. Depending on how comprehensive the objectives are, the model can generate a diverse range of questions to suit the needs of the study, all in a matter of seconds.
This isn’t to say that all the questions produced will necessarily make it to the final draft, but they can form a quick starting point to developing a questionnaire, saving precious time in the research process. Chat GPT could also be used to refine a set of questions that may have been unclear or superfluous. By using its language generation capabilities to rephrase questions, it could improve the clarity and specificity of the questions, making them more likely to elicit desired information from respondents.
Conducting surveys: An exciting opportunity to use this technology might be in creating chat-style or conversational surveys. This style of survey exists already and can help with participant engagement with the survey, particularly with younger audiences. Implementing the natural language processing and generation capabilities of ChatGPT could make the survey experience feel less robotic and more engaging for users. In addition, it could allow for dynamic questioning, in which questions are generated based on previous answers, resulting in a more personalized and efficient survey.
Data Analysis: There are a number of ways in which ChatGPT could theoretically assist in quantitative data analysis. Complex numerical data can be converted into insightful and easily accessible text to give an overview of key findings in a data set. It could also be used to draw comparisons between different data sets, for instance between several years of a tracking study. ChatGPT could also identify trends in the data and make projections based on historical patterns, providing insights into future patterns and trends.
Report Writing: The language generation capabilities of ChatGPT make it particularly enticing as a tool to aid in generating text for quantitative (and qualitative) reports. For instance, the model could condense research findings into an executive summary, using clear and concise writing. Another area in report writing that ChatGPT could assist is in creating complex visualisations. Whilst potentially out of the scope of a typical market research project, ChatGPT could be used to write code that can be used in unison with a data visualisation platform such as Matplotlib, to generate charts or more complex visualisations for a report.
ChatGPT for Qualitative Research
Topic Guides: In a similar manner to questionnaire design, mentioned above, ChatGPT has the potential to assist in topic guide generation. By providing the model with a list of research objectives, it can produce a structured topic guide that can be used as a framework for the final topic guide. Again, similarly to questionnaire design, the advantage of this is that researchers could go from a list of research objectives to a working topic guide in a matter of seconds, saving time and effort for other tasks.
Live-chat interviews: There is some potential for ChatGPT to be used to conduct real time interviews, similar to a depth interview. ChatGPT could potentially be programmed to follow the structure of a topic guide, asking questions and clarifying responses, while also incorporating follow-up questions and encouraging elaboration. To be able to respond to speech input from respondents, it would also need to be paired with a speech-to-text platform like Amazon Alexa, using the Alexa Voice Service API.
Analysis: A key strength of ChatGPT is its ability to seemingly ‘understand’ human text inputs. This makes it particularly enticing as a tool for qualitative analysis. One potential area where it could be used is in summarising open ended responses into a shorter output, or similarly, extracting the key information from focus group discussions. There is also potential for it to efficiently code qualitative text responses into a number of categories. This is one of the more time consuming aspects of qualitative research so the opportunity to significantly speed up this process holds considerable value. Finally, using its language processing capabilities, ChatGPT may be able to efficiently conduct sentiment analysis on large sets of text, extracting the key themes from the data. This again would be a huge advantage to researchers, as this is a time consuming and specialised process.
ChatGPT for Desk Research
The ability for ChatGPT to assimilate and condense information from a vast array of sources, makes it theoretically useful for desk research. In a similar manner to some of the topics discussed above, key information can be extracted, summaries can be written, and specific datapoints can be highlighted, saving the researcher valuable time and effort.
A theme that you may see emerging from all the findings in this article is that ChatGPT will likely be used as a time-saving technology for researchers, improving the efficiency of research and allowing their efforts to be focused elsewhere.
Potential Drawbacks and Limitations
Following on directly from the desk research section above, there are currently some limitations to how ChatGPT could be used for desk research. In its current state, it does not have access to the internet, but rather, it has been trained using a vast quantity of data. The most recent data that was used to train the model was from 2021, so any events or information published after this date are not included in the outputs from the model. As you can imagine, this has significant drawbacks to the use of ChatGPT for desk research. Additionally, whilst ChatGPT is capable of citing sources of information for desk research, the sources would be limited to what it has been trained on.
Another potential limitation of using ChatGPT in market research is the technical learning curve that may be required. A good understanding of machine learning, natural language processing, and AI development may be necessary. Many of the use cases mentioned above would require setting up an API in order for ChatGPT to perform tasks. For instance, to start analysing quantitative data as mentioned above, the data would first need to be set up in a text format that is accessible for ChatGPT, such as a CSV file. To actually give ChatGPT access to this data, you would need to set up an API request using the OpenAI API. Within the API request you would be able to specify the analysis to perform and any other actions to take on the data. The benefits received vs the time and resources put into setting up this sort of analysis, may not be considered worthwhile.
Something to bear in mind is that for a lot of the use cases above, ChatGPT would require ‘training’ through a process called fine-tuning. For instance, to use ChatGPT to conduct chat-style surveys or live-chat interviews, its language model would need to be adapted by training it on a smaller, specialised data set. This might include survey questions, follow-up questions, and response options. Such a data set would need to be large enough for the model to learn and recognise patterns and subtleties within the data. This would be followed by progressive rounds of refinement and fine-tuning. It’s also worth mentioning here that the quality of the output from ChatGPT can only be as good as the data it receives, so ensuring high quality data input would be critical.
Whilst the outputs from ChatGPT are generally accurate, there are limitations to the power of AI generated information, so human checking will always be essential. This is often challenging, as the model is designed to create authentic ‘human’ sounding outputs, which can make it difficult to spot inaccuracies.
The spread of false information is already a pressing issue in the world today, so researchers will need to be cautious in how they use ChatGPT for their purposes.
There is also the issue of copyright that needs to be taken into account, as the way in which ChatGPT generates its outputs could be considered stealing. Citations or sources are not inherently provided by the platform, unless specifically requested by the user.
It should be noted that ChatGPT is not a platform that is set up to run market research projects in isolation and would need to be used in conjunction with external software. For instance, if you were running a chat-style survey, a third party chatbot platform like Microsoft Bot Framework, or virtual assistant like Amazon Alexa would be required.
There are also ethical considerations to take into account if using real participant data with AI. It would be the responsibility of the researcher to ensure that the use of ChatGPT complies with all applicable privacy laws and regulations. This may include obtaining the necessary consent from data subjects, conducting data protection impact assessments, and implementing appropriate technical and organisational measures to protect personal data. There is also the consideration of client consent that should be taken into account.
In conclusion, this AI-powered language generation model holds a vast amount of potential to boost productivity within market research and beyond. It shows great promise in speeding up many of the more tedious aspects of research as well as opening up new opportunities to change the ways in which research is conducted.
On the surface, ChatGPT might appear to be equipped to undertake a market research project all by itself, however many of the use cases discussed here would require significant human input to set up and run. Whilst generative AI can help with some tasks and bring new insights, but it can’t replace human creativity and expertise. It’s important to critically evaluate AI results and involve human experts in decision-making.
This technology seems poised to revolutionise the way research is conducted, and it will be exciting to follow the different approaches that researchers employ to take their research to the next level.