Want to use AI to transform education marketing? Stop fixating on generative AI.
Learn so so much more about this topic at the ContentEd AI bootcamp taking place over two half-day sessions 5-6 June 2024.
Learn how others in higher ed are already embracing AI in content strategy, marketing and recruitment.
Hear from sector leading voices on the use of AI in content marketing.
Hear from higher ed’s first ever Chief AI Officer
Get hands-on with our workshop to show you how to build your own custom no-code AI assistants. Create yours live!
Register for just £179 for an individual or £499 for a team pass.
AI has been a breakthrough topic in the world of education and marketing (and just about everywhere else) in recent years. And while some of us may remain skeptical of its value, reliability, and ethical impact, it’s undeniable that AI and machine learning has already changed so much in how we engage with the world around us, and will continue to do so.
However, with so much of the conversation focused firmly on generative AI - with tools like chatGPT, Dall-E and midjourney dominating - are we missing huge opportunities for education marketing?
In this post I’m going to talk about the applications of AI, data science and machine learning, and share some of the work that we’ve been doing in this space at Pickle Jar. We’ll explore the potential for AI far beyond casual desk research and content creation. And - not to dismiss generative AI - we’ll also show how with some careful planning we can use it for content production in a truly bespoke way for our institutions.
So how are we using it for ourselves and for our education sector clients?
Designing a chatbot that gives whole-institution access to audience research and insights
This one excites me so much. Universities often hold insights and research into their different audiences. But those findings exist in many different formats: research reports, survey results, slide decks, user story banks, customer service enquiry data, engagement analytics, and much more. Some are PDFs. Some are .csv files. Some sit on a web dashboard. Some in raw interview transcripts as .mp4 and .txt files. It’s often a rich - but wildly disconnected - picture. And typically those insights sit in someone’s drive, often only accessed and used by a handful of people. Their value is rarely fully realised.
Right now we’re leading a project for a UK university in which we’re looking across all of their target audiences, filling insight gaps by conducting original audience research, and then designing a chatbot enquiry system that gives all staff around the university access to the research findings (the knowledge base) in a highly user friendly way.
Want to know what content alumni from Singapore prefer to receive from the university? The chatbot will tell you by interrogating a real and accurate knowledge-base based on actual alumni research data. Want to know what information parents want ahead of graduation ceremonies? You got it. Want to know which channel is the best place to reach major funders. Our chatbot can probably tell you! Want to know what time of day to send that all-department email? The bot can tell you, based on actual data specific to that university.
Cool huh?
Conducting qualitative research interviews
At Pickle Jar we love qualitative audience research. And we spend a lot of time doing it. Conducting individual interviews isn’t just time-consuming for us. It can also be expensive for the client. And sometimes interviewees don’t always give an honest reply because they’re afraid to speak their real truth to a human that they believe represents the university.
In a bid to make our audience research more cost effective for our clients, to increase sample sizes for qualitative interviews, and to experiment in prompting authentic replies, we are using an AI assistant interviewer.
We train the AI on the questions and insights that we want it to gather and the outcomes we hope to receive from the interview. We suggest questions, but it uses those as a conversation guide not as a defined script. It’s then matched with an interviewee, who is told that they are being interviewed by an AI. The conversation takes place over a text exchange.
If you worry that this experience would be stunted or awkward, think again. Our tests so far show that the AI is able to engage in a very natural conversation. The AI gets curious and builds their next questions on the responses given by the interviewee. And we are currently also exploring and testing the hypothesis that this experience leads to a more authentic response. How many times, after all, have you been willing to express your real feelings to a chatbot in a way that you never would to a human being? I do it all the time (sorry chatbots).
Making sense of extensive stakeholder consultation
No matter how brilliant the human brain is, it’s not adept at processing large amounts of data at the same time. A big part of our work when developing brand or content strategies is stakeholder consultation. And when we’re working on something like a whole-institution content or brand strategy, stakeholder consultation can be extensive, sometimes involving hundreds of voices.
Even with smaller research pools, it results in an overwhelming amount of transcripts and notes to be processed and for us to make sense of them all. That work is painstaking and time-consuming, often involving the need to map key points from a transcript across into a spreadsheet, and then meticulously tagging and categorising every insight. And don’t even get me started on how we then synthesise those insights from there.
So, here we’re using AI in two ways. First we will often record stakeholder interviews and then run them through an AI transcription tool to produce a pretty-accurate script of the conversation. The one we use hits about 95% accuracy most of the time these days and usually the transcribing mistakes don’t alter the meaning of the content (and can be corrected). Then we load all of those transcripts into another AI tool and start to use that tool to identify patterns, trends, priorities, correlations, disagreements, key points and much much more. We find that AI is able to process the data in many more ways, shapes and detail than our lovely human brains ever could. And it’s faster too.
Building AI knowledge bases for deeper desk research
The ability to upload lengthy data and have AI process, summarise and report on it for you is probably one of the greatest uses I and my colleagues make of it right now. So, in the previous example we talked about using it to process stakeholder consultations. But sometimes a project also requires us to read through lots of hefty research papers, policy reports, and more.
One client currently has me wading deep in research about teacher CPD, teacher salaries, and all kinds of other related topics in order to help me produce a case for support for a new initiative. And I need to understand it from a global perspective. As you might expect, there’s some great data and research reports on this topic. But those reports are often huge, with many over 100 pages! To find the answer to something reasonably simple could involve me having to trawl through thousands of pages of detailed and dense PDFs.
So, again, we now create a knowledge base of all of the research reports, and use chatbots to help us find the answers (and specific references) to the questions that we are seeking to answer. The ability to combine the data from multiple sources and make sense of those patterns, correlations and differences, is a total dream too!
Processing large user story banks to inform content and information architecture
When a thorough job has been done of audience research for content marketing and content design projects, we can end up with large user story banks. I have one in my drive at the moment relating to the student recruitment and enrolment journey. It contains 342 user stories. Now, that number isn’t insurmountable in terms of the capacity for the human brain to process it. Just. But if we start to factor in other audiences and their needs and motivations, suddenly we’re working with huge datasets to influence information and content architecture and design.
So, we’re also using AI assistants to help us understand patterns in user story banks so as to identify common themes. Such pattern recognition informs information architecture, user journey design, and taxonomy design too.
Working in an AI partnership to make sense of a user story bank means that we can identify patterns in audience type, audience needs, and audience motivations - or a combination of all three at the same time. And in doing so we can design user journeys that really meet the user exactly where they are at irrespective of whether we approach it from the perspective of audience type, content need, or audience motivation.
Designing a generative AI co-pilot to create content based on audience insight, brand, style and tone
Now, I’m writing this blog post for real. I promise. I’m sitting here with my cuppa, and I’m loving the moment to write something a little more complete than I’ve had the space to produce in a while. But I don’t always want or need to generate every word from scratch. I’m truly open to partnering with AI to produce content, and to repurpose and reshape content for us and for our clients. And I recognise that it needs some serious training to get it right.
Recently I built a custom AI content marketing co-pilot for Pickle Jar. I’ve trained the AI on our target audiences (their interests, needs and motivations), on our style and tone, and on the types of things that Pickle Jar loves to talk about and has expertise in.
Our content marketing co-pilot is able to suggest blog post topics for us, and it can then write the post. We’ll always only ever use the content created as a rough first draft, but it certainly speeds up the whole process and helps us to explore new content ideas in a totally new way. In the images below you can see the assistant generating ideas for a new blog post for me to write.
Creating bespoke course combinations using a content model and natural language processing
The final example that I can tentatively talk about here is a new project that I’ve been commissioned to work on. It relates to a new learning platform. In that platform students will have access to a vast library of modules, but we want them to be able to curate modules together in such a way that they form totally bespoke courses. That’s where AI comes in. By treating module transcripts as a knowledge base, and allowing the students to tell the AI what they’d love to learn, the AI can then create recommendations from the knowledge base to then suggest a “collection” of modules that form those bespoke courses. This one is at concept stage right now, but we have the content model mapped out, we’re working on the taxonomies and the content strategy, and we already know how we’ll deploy machine learning to make the bespoke curated course experience a total reality!
A final word on the ethical implications and how we’re working responsibly and with caution
I’d be amiss in a post that explores the opportunities of AI for education content marketing to not reflect on the ethical implications of this, and the downsides. AI takes patience and training (like most humans, let’s face it). And we lean towards uses of it right now where we have tight control over the data being used and the way in which it is used. In another words, we create the knowledge base and then we also fence it. In doing so we’re able to have much tighter controls over things like biases built into the data, or the risk of AI hallucinations. It also helps from a data privacy perspective. And when we’re dealing with person-specific data (like interview scripts, or survey responses), that data is all anonymised before entering any knowledge base. When we recommend its use for anything that involves personal data or a human-AI interaction, transparency and consent are paramount.
It’s an ongoing learning process for us all, but hopefully this post has given you a few thoughts for other ways to embrace AI that aren’t all just about chatGPT and generative AI.
Want our help?
We can design a bespoke project just for you, or provide inspiring keynote talks or workshops on the use and potential of AI for education marketing, content strategy and advancement. Contact us and we’d be happy to discuss ideas with you and make suggestions.