Is your business struggling to manage an influx of support tickets? Maybe you’ve decided to turn towards automation and AI-based tools to supplement your human support team.
Unfortunately, all the responses your AI tool generates come across as “robotic”. You tried it out for a bit, but your customers complained. The problem might not be the tool itself. It might be that no one on your team bothered to humanize the content before sending it off.
I’ve worked for a number of companies that got excited about the utility of ChatGPT in customer service. They thought it would streamline operations and eliminate the need to spend hours per day typing up replies to customers. Unfortunately, none of them considered the human touch that would be missing from the generated content – to their detriment.
Let’s talk about the benefits of humanizing AI content specifically for customer service, and then explore how to do it yourself (which isn’t as hard as you might fear).
What Benefit Does Humanized AI Content Have for Customer Service?
The problems with using AI in customer service often have little to do with the AI itself. Rather, it’s the lack of personalization and humanity present within generative text that leaves customers soured.
If buyers feel like they’ve received lesser service while the company benefitted at their expense, they’ll pack up their business and take it elsewhere. This is why ‘humanizing’ AI content is so important in customer service – even minor interactions impact your reputation.
Adjusting AI content to make it feel more human allows your business to:
- Maintain the customer service voice and branding you’ve worked to develop. AI content, especially generative text, is often impersonal and generic. You want everything customers get to sound like it comes from your business.
- Use nuance and context to shape your approach. The gravity of a particular situation or circumstance might be lost on AI. It might miss contextual cues and small details. You’ll want to ensure that your responses acknowledge customer emotions and speak to their specific situations.
- Improve user engagement. If AI text feels more authentically human, the customer can connect with it beyond its purely functional purpose. Many customers approach AI with an understandably jaded mindset. If you can imbue AI with a personalized touch, you can use it more frequently without incurring the wrath of your customers.
What Does It Mean To Humanize AI Content?
Chances are if you’ve spent any time on the internet in the past few years, you’ve been exposed to plenty of AI-generated content. You can probably immediately recognize it – whether it’s a piece of text, or those uncanny TikTok videos and YouTube ads.
Let’s take a look at two written replies to a customer inquiry. The customer wants to know how to add courier options to a shipping platform called Sprightly.
One was typed up by my (super) human colleague in a few minutes, and the other spit out by ChatGPT in seconds. Can you tell the difference?
And here’s the prompt I fed into ChatGPT to generate that second example. Note that I did have to provide some additional information about Sprightly’s policies for it to generate a more accurate response:
The major concern with using generative text to reply to customers is that it lacks nuance and context. Many interactions involve unique situations, or information specific to your niche and products. AI won’t be able to account for any of that, at least not by default.
On a more subjective level, humanizing AI for customer service means making content sound more natural, engaging, relatable, and personal. Embedded human emotion, personalization, and sentiment help to balance the fact that AI is restricted to the data it’s trained on. Human edits maintain authenticity, and elevate generic responses.
3 Quick Tips To Humanize AI for Customer Service
Is your business looking to create a canned response using a tool like ChatGPT, and leverage it for customer service? Or do you want to shift a large portion of support over to AI chatbots?
Before you can begin humanizing generated content, you’ll need to determine where and when you’ll use AI in customer service. Once you’ve done that, here’s our advice for turning bland default responses into valid replies.
1. Use First-Person Pronouns and Address the Customer Directly
Depending on the AI tool you use, it will often fail to address the customer directly. After all, you’re feeding it a prompt in the form of a question or command.
While AI can spit out first-person plural text, what you get won’t directly speak to the customer. It’s directed either to you, or an imaginary third party.
You want to avoid creating impersonal distance between your AI-generated text and the customer. First-person pronouns convey a sense of personal experience and relevance. Addressing the customer directly adds a layer of personalization that feels as if a genuine conversation is taking place.
In each of the examples throughout this post, I’ll show the customer inquiry, the AI prompt used, and the raw text it provides. Finally, I’ll assess how we can humanize the result for customer service.
Our first inquiry:
For this example, I wanted ChatGPT to provide me with a return policy for a retail clothing store. Our imaginary store hasn’t determined a concrete policy, but has already started to receive return requests.
Here’s the prompt, and the generated text result:
Humanizing the AI Content
In this example, I want to reframe the return policy so it focuses on the product the customer wants to return (a cardigan). That means singling out this portion from the generated text, adding personalization by integrating first-person pronouns, and speaking directly to the customer.
More specifically, I’m going to remove “At [Store Name]” and replace it with “I’m sorry to hear you weren’t happy with the sizing of your cardigan.” Based on my knowledge of the product, I can validate the complaint and explain why the issue has occurred.
Next, instead of diving into the return options formally, I’ll ask the customer which option works best for them, offering up both as potential solutions. Then, I’ll paste back in the return choices.
Here’s a revised version of the parts I edited:
Hey Tara,
I’m really sorry to hear that you weren’t happy with the sizing of our ‘Cherry Pie’ cardigan.
Sizing on this particular product varies quite a bit, as each piece is handmade by our supplier in Turkey.
Is the product unworn, and are the tags still attached? If so, I’d be happy to arrange a return for you. Please find our two return options below – you can let me know which one works best for you!
Most companies will already have their own return policies. So they don’t need ChatGPT (or another AI tool) to create one for them.
However, let’s say your company uses a help desk like Groove that has built-in AI features. AI drafts can pull from your internal documentation (like a return policy) to create replies.
You’ll still need to humanize the content. Most policies aren’t formatted for brevity or for use in email responses. Taking the time to edit the response for each inquiry makes it more personal and accessible. But a built-in AI tool saves a lot of time.
2. Eliminate or Rewrite Overused AI Words and Robotic Language
Here’s an example of a customer who isn’t being quite as nice or clear in the reason behind their email. They reference “frustration”, but don’t mention what aspect of the service caused it. AI will never account for this sort of nuance, because it doesn’t know your product.
Instead, it will use very broad and generic language to automate a response to a cancellation request. This text could be applicable to any organization with no defining characteristics. It’s polite, but it still feels robotic.
Here’s our customer complaint:
And our prompt and default generated text:
Humanizing the AI Content
In this example, there is a lot we could change to humanize it. However, our focus right now is on rewriting for overused words and robotic language. Generative text has a tendency to repeat itself, use similar words frequently, and regurgitate the same phrases.
Also, I’m going to adjust the very bland: “Sorry to hear you’re considering canceling your subscription, but I’d be happy to assist you with the process.”
The cancellation steps need to be changed as well, so that they’re relevant to our business. And the robotic “we’d love to hear your feedback” needs to be replaced with genuine outreach to understand where we went wrong.
Was it a technical issue that we can fix ASAP? If we can work to resolve the customer’s problem, we still have a chance to win them back. AI focuses on providing information – humans are needed to solve problems.
Considering that this customer is already upset, I want to balance a conversational tone with appropriate formality. And express empathy, which this AI content lacks entirely.
Here’s the revised version:
Hello Nick,
Coincidentally, also Nick replying here! I totally understand if you’re having some frustration with the platform, as I know any new software can have a steep learning curve. Was there a specific problem you’re experiencing that I can address prior to cancellation?
Our goal is to help all our customers, so we’d love to know where we went wrong. If we’re not able to resolve the issue for you, then no worries – here’s how to cancel.
And then I can add back in the AI-generated cancellation steps. Those are nice and clear, plus repetition of words and phrases is expected in instructions.
3. Account for ‘Perplexity’ and ‘Burstiness’ To Make AI Content More Dynamic
Perplexity and burstiness are used to measure the competence and complexity of language models. We can also use them to make content feel more human.
- Perplexity refers to how likely the model is to predict the next word in a sequence of words. A higher perplexity score indicates that the content is less predictable and more varied.
- Burstiness refers to the tendency of words to appear in “bursts” throughout the text, rather than being evenly distributed. A higher burstiness percentage indicates that a piece of content has more varied content structure and word usage – indicative of human-written text.
In layman’s terms, this is about distinguishing AI and human text based on how we use words and how often we use them.
Imagine that you read an article, and you can immediately identify that it’s AI. It uses a narrow vocabulary, and often repeats keywords in the same spots. Conversely, if content flows more naturally, and is more creative, people will assume it’s human.
For our example, a customer is having a problem trying to log into their account. It’s a common question in SaaS or retail:
Here’s the ChatGPT AI prompt, and the result:
Humanizing the AI Content
The basics in this content are fine. But it needs to be a little more varied to come across as less robotic. Even if customers don’t suspect that a response like this is AI generated, they’ll probably think you just copied and pasted a reply with zero attention to the urgency of their concern.
The primary way to adjust burstiness and perplexity is to vary the length and style of sentences. Adjust the text as though you’re having an authentic conversation with the customer.
The goal is to ebb and flow like a human conversation, and break the robotic monotonous style of AI writing. Here’s a revised reply that accounts for perplexity and burstiness:
Hey Martin,
We’re here to help! Don’t worry – this could very well be a temporary technical issue, and nothing to do with your account.
Is there a specific error message you’re seeing? And have you tried these quick fix steps?
- Resetting your password – just hit the “Forgot Password?” link on the login page.
- Clearing your browser’s cache – let me know if you need any guidance on how to do this (and what browser you’re using).
- Trying the website on a different browser (like Firefox instead of Chrome) or device (like a smartphone instead of desktop).
- Updating the browser to the most recent version. We’ve had some customers report issues with older versions of Chrome and Microsoft Edge.
If you’ve already tried all of these, please let me know your login credentials and the error message, and I’ll investigate further with our IT department.
Building AI Into Your Customer Service Workflow
At the end of the day, a lot of AI tools just aren’t designed for customer service. They really have to be adapted to work within this context.
As for the few support-specific AI products that are out there, we’ve found that they’re quite expensive.
For the best of both worlds, you’ll want relevant AI features incorporated within your customer service dashboard. That way you can make use of AI functionality that’s relevant for support, and avoid the need to switch between tools.
Groove, our help desk for small businesses, provides a shared inbox that centralizes all your customer communications. Better yet, it enables your team to leverage a handful of powerful AI features right within that inbox.
You can craft a detailed response and use “AI Edit” to make the content more formal or casual. “AI Overview” provides sentiment analysis of the customer’s emotions, and “AI Drafts” provide detailed answers pulled directly from your own email responses or self-service knowledge base.
Sign up for a free trial of Groove today, and deliver on personalized content without sacrificing productivity!
