“The emergence of GPT tools has revealed a plethora of ways in which Large Language Models can support not just insight but more efficient, or effective, research within the marketing function,” said consultant Neil Perkin in Econsultancy’s most recent Digital Shift presentation.
A recent report by AI at Wharton and GBK Collective threw a spotlight onto how much the usage of generative AI by marketing and sales has grown in the past year – from 20% of the marketing/sales function using it weekly in 2023 to 62% in 2024.
Among 800 senior business leaders surveyed for the report, 62% are reportedly using generative AI for data analyses and analytics, while 57% are using it for customer research or competitive intelligence. Fifty-five percent of leaders are also using generative AI for predictive customer behaviour.
In the opening chapter of Digital Shift Q4 2024, Perkin gave a run-down of some of the practical current applications of generative AI in marketing research and insight, and what specialist tools can be used to support these use cases.
Custom GPTs are a no-code and more specialised iteration of ChatGPT – sometimes likened to apps – that carry out a particular function or functions. In January, OpenAI launched the GPT Store to enable browsing and download of custom GPTs; originally available to paying ChatGPT+ subscribers only, the store was opened up to free users in May, although GPT creation is still locked to paid subscribers.
Summary and analysis of long documents and papers is a valuable function of large language models like ChatGPT, and Perkin highlighted the availability of a range of custom GPTs that specialise in this function, such as AskYourPDF Research Assistant.
“These research tools offer up a variety of capabilities, including … the ability to summarise YouTube videos without watching them; or analyse research papers or Excel files, or a number of PDFs simultaneously,” said Perkin.
Marketers can use tools like these to analyse industry reports or even consumer behaviour studies.
“A marketer might upload a PDF of a recent market research report and use the tool to ask questions like: ‘What are the top emerging trends in this industry?’ Or, ‘What are the main challenges faced by customers?’ That helps in swiftly gathering the key insights to shape campaigns or strategies.”
An important caveat issued by Perkin when using LLMs to gather insights from data is to avoid uploading any proprietary data to public GPTs; enterprises can create closed, restricted GPTs if desired that can only be used by members of the organisation.
Beyond GPTs, generative AI has been used to power and underpin new tools that can assist marketers with research and gaining insight. Perkin highlighted three standout examples in this area:
Launched by ex-DDB strategist Piotr Bombol, Adaily offers a database of thousands of marketing and advertising campaigns that marketers can use to find examples and answer specific questions, such as, ‘What’s a great example of reinventing category norms through advertising?’
The tool has daily, monthly, and yearly price tiers that enable short or long-term uses of the database as needed.
Described as “Your springboard in the creative process”, Springboards is a tool designed to aid agencies in developing creative ideas by riffing on concepts for campaigns, taglines, manifestos, and more.
“The founders make the point that many existing genAI tools are designed to try and give you the answer, whereas theirs is about helping you to get to the best answer – by inspiring different thinking, by sparking new creative conversations … and also exploring other unexplored options,” Perkin explained.
NotebookLM is a tool developed by Google and powered by Google’s Gemini 1.5 that the company bills as a personalised research assistant. It offers summarisation of a range of sources from PDFs and websites to videos and slides; connections between topics; and answers to questions with citations.
Perkin listed a large number of different marketing use cases for NotebookLM, including content curation, distilling complex research, campaign planning, competitor research, and team collaboration – such as via creating quick summaries of meetings or brainstorms, helping the team to stay informed and on the same page.
Tom Roach, VP Brand Planning at Jellyfish Global, wrote in Marketing Week that “‘Share of model’ is the new marketing measure for the AI era”. Derived from similar concepts like share of market and share of voice, Roach explained that share of model measures “the number of mentions of a brand by one or multiple LLMs, as a proportion of total mentions of brands in the same category.”
It can also provide insight into brand positioning by “generat[ing] clusters of positive and negative associations with brands and categories” that can be compared with competitors and tracked over time.
Perkin cautioned marketers to “avoid the temptation to try and boost brand visibility in models by low-quality content spamming” – as has frequently been done in search to the overall detriment of search result quality.
Columnist Mark Ritson, a minor shareholder in synthetic research startup Evidenza, has written about the advantages of synthetic data for Marketing Week, calling it “as good as real” earlier this year. He has highlighted that it offers data “where there was none”, in a small fraction of the normal time taken and for a fraction of the usual cost.
A note of caution about using off-the-shelf LLMs for synthetic data comes from Kantar’s Ashok Kalidas and Jane Ostler, who point out that GPT-4 displayed a strong positive bias compared to actual human respondents, while not appearing to capture subgroup trends; GPT-4’s qualitative responses also lacked variety. With that said, these tests were run more than a year ago, and LLMs have advanced in various ways since then.
Experts have also warned about synthetic data’s avoidance of outliers: because synthetic research replicates averages, it does not account for outliers or fringe behaviours – which as Helen Edwards points out in her book ‘From Marginal to Mainstream’, “can unleash ‘consumer-driven disruption’, promoting new innovation, new routes to market, new winners and losers – and new growth.”
Perkin concluded that, “Synthetic research can be used to inform your thinking – about an audience, about user experience, or a strategy.” Use cases include creating synthetic personas that can simulate responses to qualitative interviews; and generating synthetic reviews from customer personas based on key brand attributes, which can then be analysed for sentiment and perception trends.