How AI Analytics Is Reshaping Marketing Personalisation
If "AI in marketing" still conjures up images of chatbots and image generators, it's worth zooming in on something less flashy but arguably more useful: AI-powered analytics that read, sort, and personalise content at a scale no human team could match. Four recent studies on content tagging, social media monitoring, image personalisation, and the state of AI research in marketing more broadly show what this actually looks like in practice, and where its limits are.
AI That Reads and Organises Your Content For You
Imagine a large news organisation publishing hundreds of articles a week, every one of which needs to be tagged with topics so it can be found, recommended, and analysed later. Traditionally, that's a job for editors and it is slow, inconsistent, and impossible to keep up with at volume.
Salminen and colleagues (2019) tested whether an AI analytics tool could do this instead. Trained on more than 20,000 articles from Al Jazeera and nearly 800 possible topic tags, the best-performing model correctly tagged content with an accuracy comparable to a solid human editor, and confidently labelled 99.6% of previously untagged articles on its own. Even more interesting: without any extra training, the same tool was applied to the organisation's YouTube videos, which is a completely different format with much shorter text to work from and still classified 96.1% of them. When human reviewers checked a sample, the AI's tags agreed with human judgement about as often as two human editors agreed with each other.
Why it matters: if your content covers similar ground across your website, blog, and video channels, an AI tagging tool can likely save your team significant manual sorting time, though it works best when your channels genuinely overlap in subject matter, not when they're wildly different.
AI That Knows Which Comments Actually Need a Reply
Most social media monitoring tools work by flagging negative comments, which is an assumption being that unhappy customers are the priority. Vermeer and colleagues (2019) show this misses the point. Looking at roughly 60,000 Facebook comments and 11,000 tweets across sixteen brands, they found that most of the comments genuinely worth a brand's attention weren't negative at all. Many were simple questions, suggestions, or enthusiastic praise that still deserved a reply.
Instead of asking "is this negative?", the researchers trained an AI system to ask a better question: "does this need a response at all?" That reframing paid off. The AI-based approach correctly identified response-worthy comments about 60% of the time, more than double the accuracy of standard sentiment-tracking tools (15–28%). It also worked reasonably well when trained on Facebook data and then applied to Twitter, which is useful, since most brands operate across several platforms at once.
Why it matters: if your social listening setup is built purely around flagging negativity, you're probably missing most of the comments that are positive, neutral, curious and are actually worth a human reply.
AI That Matches Images to Personality Types
This is the most futuristic-sounding of the four studies, but the logic is fairly intuitive. Matz and colleagues (2019) asked: instead of designing one image to appeal to "the average customer," could you predict which images appeal most to which personality types, and use that to personalise creative at scale?
Using over a thousand professional stock photos, the researchers had AI analyse features like colour, composition, and whether people appear in the image, then matched those patterns to the personalities of the people who liked each photo. The system could predict, with reasonable accuracy, which images would appeal most to extroverts versus introverts, or to people high versus low in openness. In follow-up experiments, showing people images that matched their own personality rather than just generically "appealing" images measurably increased how much they liked the image, how favourably they viewed the brand, and how likely they said they were to buy.
Why it matters, and a caveat: the effect was real but small shifting someone from a poor to a good personality match increased purchase intent by only about one-tenth of a standard deviation. The authors themselves note this kind of personalisation is most worthwhile for businesses running large-volume campaigns, where small gains add up across thousands or millions of impressions probably not (yet) a priority for a small local business running a handful of ads.
What the Research Says About AI Analytics Current Limits
Ma and Sun (2020) step back from any single application and ask a broader question: how good is AI analytics, really, at helping marketers understand customers and not just predict their behaviour?
Their honest answer: AI analytics tools are genuinely strong at predicting what will happen, which article will perform well, which comment needs a reply, which image will resonate. They're much weaker at explaining why, and at reliably tracking how an individual customer's preferences shift and evolve over time. In plain terms: treat AI analytics as a powerful pattern-spotting assistant, not a substitute for understanding your customers.
Recommendations for Marketing Professionals and Entrepreneurs
• Start with the highest-volume, most repetitive task first. Content tagging and comment triage are the easiest wins. They're exactly the kind of high-volume, rule-based work AI analytics tools handle well today.
• Don't rely on sentiment alone for social listening. If your current tool only flags negative mentions, you're likely missing most of the content genuinely worth responding to. Ask your platform or agency whether it can flag relevance, not just tone.
• Test cross-channel tools before assuming they'll transfer. An AI model trained on your blog content may or may not work well on your Instagram captions or YouTube descriptions. It depends on how similar the subject matter really is. Pilot before you scale.
• Treat personalisation gains as incremental, not magic. Image or content personalisation based on AI predictions can lift engagement and purchase intent, but the effects tend to be modest per customer. It's a volume game worth investing in once you have scale, less urgent for smaller campaigns.
• Keep a human in the loop for judgement calls. These tools are best at surfacing patterns and narrowing down what deserves attention, but not at fully replacing editorial or customer-service judgement, especially on anything sensitive or high-stakes.
• Ask what a tool can't do before you buy it. Any AI analytics vendor should be able to tell you plainly where their tool struggles with causation, with genuinely novel content, or with tracking individuals over time. If they can't, that's a red flag.
References
Ma, L., & Sun, B. (2020). Machine learning and AI in marketing – Connecting computing power to human insights. International Journal of Research in Marketing, 37(1), 481–504. https://doi.org/10.1016/j.ijresmar.2020.04.005
Matz, S. C., Segalin, C., Stillwell, D., Müller, S. R., & Bos, M. W. (2019). Predicting the personal appeal of marketing images using computational methods. Journal of Consumer Psychology, 29(3), 370–390. https://doi.org/10.1002/jcpy.1092
Salminen, J., Yoganathan, V., Corporan, J., Jansen, B. J., & Jung, S.-G. (2019). Machine learning approach to auto-tagging online content for content marketing efficiency: A comparative analysis between methods and content type. Journal of Business Research, 101, 203–217. https://doi.org/10.1016/j.jbusres.2019.04.018
Vermeer, S. A. M., Araujo, T., Bernritter, S. F., & van Noort, G. (2019). Seeing the wood for the trees: How machine learning can help firms in identifying relevant electronic word-of-mouth in social media. International Journal of Research in Marketing, 36(3), 492–508. https://doi.org/10.1016/j.ijresmar.2019.01.010
