Why do people hate AI personalized messages?
And what should you do about that?
It started with some weird sounding comments on LinkedIn posts. Then I saw them everywhere in my feed. Then on collaborative articles. But then they were in my inbox. At that point I just had to ask one such messenger - Do you even understand the term personalization?
As AI becomes increasingly capable of generating content that at the first glance sounds ‘personalized’ and ‘meaningful’, what happens when each and every person on this planet starts sounding the same in their messages? How do you think the readers of such messages respond when they see the same format, the same vocabulary, the same sentence structure being used by tens of other people who reach out to them?
Well received AI generated communications
Don’t get me wrong. I am not against using AI for generating content, comments or cold emails. But only when done for the right purpose and using the right techniques.
One such use case is automated customer support - when AI-generated messages are clearly identified as such and perform simple tasks effectively, users generally respond positively, appreciating the efficiency and availability that cannot be feasible if we only rely on human interactions. It is certainly better to get a terse one liner emotionless AI response that tells you how long it would take to deliver your order than to wait for minutes on customer care calls that are usually being handled by underpaid and overworked humans. It’s even better if the AI can add just a slight bit of twist that makes the reader smile - yet not creeped out.
When the use of AI becomes debatable
In other circumstances such as reaching out to a stranger on LinkedIn, hoping to create a long term professional relationship, attempting to mimic human-like communication with AI might make the receiver of such messages more skeptical or uncomfortable, especially if they suspect whether the sender genuinely read through their profile or sent an AI generated message without spending any time or effort in actually getting to know them as a person.
While there’s nothing wrong with using AI to polish your draft messages, ensuring professional tone and an error free language, people are reacting in varied ways to such AI polished content. To someone who might not be using AI for communication, AI polished and AI generated content might look exactly the same. People who are used to spending time getting to know other people and value a genuine connection would not be too excited about messages that seem to lack the effort that goes into hand crafted messages. The worst part is - even if the sender is using AI to just refine the language - it can still look like the same message that a ton of other senders are using.
In addition to that - there's a growing concern about the use of AI in creating deepfakes or highly personalized phishing attempts, which can lead to a general distrust of anything that looks or smells like AI infused.
AI generated content today still lacks creativity that out of the box thinking brings into a write up. I have talked to many writers over the past months and invariably all of them think that AI can still not generate good poetry. Sure, it's still better than what many people can attempt, but as far as connecting with the audience goes - AI generated poetry doesn’t go too far.
And that’s a key insight for personalization as well - connecting with your audience matters more than any clever use of words or emojis. For that you need to understand the audience, not just repeat phrases from their LinkedIn profile or state how impressed you are with their experience at XYZ company (for which they haven’t added any descriptions of their work either). [Btw this is what LinkedIn’s personalization feature is doing right now.]
In other cases, some individuals are more open to human contact in general and despise AI-driven interactions from personal bias. This can vary based on factors such as age, technological familiarity, and personal values.
How to get AI generated communication right?
Human generated or AI generated content. The readers are always humans. Biased, emotional, less knowledgeable than know-it-all LLMs today, and moody humans - although even LLMs reportedly can have bad days now ;) Until the time comes when the receivers of the messages are AI too, the same factors affect any kind of communication targeted at humans.
Transparency is the first key to get this right. Users tend to respond more positively when it's clear that they're interacting with AI. This transparency helps set appropriate expectations and can mitigate feelings of deception. Knowing that the response is from an AI powered bot even makes the conversation fun and exciting.
Accuracy is the next key. AI-generated content that is highly relevant and accurate to the user's needs is more likely to be well-received, regardless of its origin. A bot blurting out wrong details about an order or an exaggerated estimated price or offering customer services that do not even exist, is worse than no customer support communication at all.
Context is the final one. AI today is actually pretty good at recognizing and responding to emotional cues; users may find interactions more emotionally satisfying - sometimes even more than human listeners. However, if the AI's attempts at emotional connection feel forced or inauthentic, it can have the opposite effect. There’s nothing cute about an AI telling me that I am not going to get my food delivered in time before an important client meeting - with a text littered with emojis.
Ethics, privacy and security are the overarching theme. With recent research clearly indicating LLMs capability to scheme against the users and developers, manipulate them, it is all the more important to draw a line where to use AI and what all information about the end users/customers is actually shared with such systems.
Setting Realistic Expectations for AI in Personalized Communication
As AI continues to advance, its capabilities in generating personalized communications are constantly improving. However, it's still important to maintain realistic expectations about what AI can and cannot do in this domain.
What AI Can Do Well
Data-Driven Personalization
AI excels at analyzing vast amounts of user data to identify patterns and preferences.
It can dynamically insert personalized elements like names, past purchase history, or browsing behavior into messages even without a template in place.
AI can segment audiences with decent precision at scale, ensuring messages reach the most relevant recipients.
Content Recommendations
AI algorithms can suggest relevant products, articles, or services based on a user's past behavior and preferences.
These recommendations can be seamlessly integrated into personalized emails or messages.
Demographic Adjustment
Advanced AI models can adjust the language and tone of messages based on the recipient's past interactions, preferences, or demographic information.
This includes adapting to different levels of formality or even matching the recipient's communication style (although this would require a much higher level of caution).
Multilingual Capabilities
AI-powered translation and localization can help create personalized messages in multiple languages, expanding reach to diverse audiences.
Sentiment Analysis
AI can analyze the sentiment of previous interactions and adjust the tone of future communications accordingly.
Current Limitations and Challenges
Emotional Empathy
While AI has made strides in recognizing emotions, it still struggles with conveying genuine empathy or handling emotionally complex situations with the same nuance an emotionally intelligent human would.
Contextual Understanding
AI may miss subtle contextual cues that a human would easily grasp, potentially leading to inappropriate or tone-deaf messages in certain situations.
Creativity and Original Thinking
While AI can combine existing ideas in novel ways, truly original, creative thinking remains a primarily human domain.
Handling Unexpected Scenarios
AI systems may struggle when faced with unique or unexpected situations that fall outside their training data. This is rare now that AI is basically trained on humongous data, but still possible depending on the use case.
Maintaining Brand Voice Consistently
Ensuring that AI-generated content consistently aligns with a brand's unique voice and values can be challenging, especially for brands with complex or nuanced messaging. AI is easy to manipulate with language in user facing systems.
Ethical Considerations
As AI becomes more sophisticated in personalization with time, concerns about privacy, data usage, and the ethics of hyper-personalized persuasion and manipulation arise as indicated in this recent research.
Avoiding Over-Personalization
There's a fine line between helpful personalization and creepy over-familiarity. AI systems might not always gauge this boundary correctly.
Realistic Expectations for AI-Driven Personalization
Given these capabilities and limitations, here's what we can realistically expect from AI in personalized email and messaging:
Increased Efficiency and Scale
AI will continue to enhance our ability to create and send personalized messages at scale, significantly reducing the manual effort required.
More Relevant Content
Expect AI to get better at delivering highly relevant content to each recipient, improving engagement rates and user satisfaction. But this still relies on having access to such relevant data about the recipients.
Enhanced Customer Segmentation
Look for increasingly sophisticated audience segmentation, allowing for more targeted and effective messaging campaigns.
Predictive Personalization
AI will get better at anticipating user needs and preferences, potentially offering suggestions or content before the user even realizes they want it.
Human-AI Collaboration
The most effective personalization strategies will likely involve a combination of both AI and human oversight.
So for the next AI based personalization feature that you develop, make sure you keep your expectations realistic and use the strengths of current AI capabilities wisely.
More importantly, do not send out rephrases of the same AI generated ‘personalized’ Christmas and New Year Cards to everyone at your office. Add a personal touch if possible.
Keeping in mind how much of a herculean task it would be for you, I would wait until the next year to publish my next article. See you in 2025!
But if you do plan on taking on a similar project for your product or organization, feel free to reach out on LinkedIn and book some time with me!
That's very true: "Transparency is the first key to get this right. Users tend to respond more positively when it's clear that they're interacting with AI."