Artificial intelligence (AI) has been a hot topic in digital. Actually, it’s been a topic since its development and study started at Dartmouth College in 1956.
But in the early 2020s (which feels both a long time ago and yesterday), generative AI became a heavy player in the discussion of web strategy.
And before we get really into things, let’s use a glossary for what’s what, because a lot is being said, and it helps to lay out the differences.
Glossary of terms
There are better, more comprehensive glossaries, but this covers enough for our purposes.
- Conversational AI is used by developers to build conversational interfaces or chatbots. You’ll see these on websites to help you find information, access help articles, or contact customer service (if you’re lucky).
- Emotional AI (affective computing) analyzes the emotional state of the user, such as voice or sensors, to personalize actions to fit the mood. No thanks.
- Generative AI is a technology that lets you create data and content from a prompt. DaVinci AI Images are an example. Give it a prompt, get an image. John Oliver married a head of cabbage about it. It’s pretty funny.
- Knowledge-based AI is a technology that captures the knowledge of human exports to support decision-making and problem-solving. They’re testing this in doctor’s offices for patient diagnoses and visit summaries.
- Rule-based AI is what chatbots follow. It uses pre-set rules and scripts to deliver information back to users. They’re usually considered task-oriented or declarative chatbots. That’s what we’ll talk about today.
There are a lot of technologies that go into AI, and many can be helpful!
If you help them first.
AI works best when trained on solid data and by someone who understands your voice, tone, and site structure. Setting an AI loose on your poorly structured website, without human oversight, is a recipe for a poor user experience and frustrated users.
How do chatbots work?
Chatbots for websites use keyword recognition and user patterns to determine what information to send you.
For example, if a user asks for “help desk,” the chatbot may send you a link to the Contact Us page of the website. The established rules of the chatbot link language rules like “help” or “call” or “contact” to the Contact Us page as the answer.
Chatbots are popular. So popular, in fact, that Gartner analyst Uma Challa predicts that generative AI will lead to a 20–30% reduction in human customer service and support agents by 2026.
While most of these chat experiences feel like you’re having a conversation with someone, some only go so far. Pre-written replies provide friendly answers and may prompt an opportunity for the user to ask for more information.
How does it understand? Through training and using multiple language tools, including Natural Language Processing (NLP). NLP is an algorithm that analyzes text and speech to help chatbots learn how to process and answer. Along with Machine Learning (ML), AI can use past interactions and store data to continue to evolve.
The more data it receives, the more it learns and, ideally, the more helpful it becomes.
If it can deliver helpful information at all.
A study on chatbots in the information technology (IT) sector from TeamDynamix showed that 48% of respondents say that chat technology does not accurately solve issues. In addition, chatbots fail because they:
- Fail to understand the user query (61%)
- Give incorrect or inaccurate answers (45%)
- Don’t understand natural language (43%)
In addition, 65% of respondents said that the chatbot/AI they’re currently using doesn’t take action and only offers answers. And 38% said the chatbot/AI is time-consuming to manage, and 23% said it’s difficult to train.
This isn’t a new trend. In 2018, the Global Consumer Customer Service Report examined customer engagements with chatbots and found:
- Only half of all respondents said they’d turn to a chatbot for a customer service need
- 25% would reach out to the company via email or social media
- Nearly half of respondents still prefer human customer agents to chatbots
AI and chatbots aren’t just robot employees you don’t have to pay: They still require monetary investment, time, training, and patience.
Your chatbot isn’t the answer to your problems
If your website is poorly structured, your chatbot is a poor addition.
UX content strategist and designer Jane Ruffino has kept an eye on big data and emerging technologies and notes that AI’s promises still require human work.
It (AI) will prove marginally useful to a lot of people, and extremely useful to a few, and will, like all the data-based hype cycles before it, require everyone to clean up their freaking data before it can deliver a single benefit, which is exactly the hard, boring, expensive work they’re trying to avoid by using magical hype solutions in the first place.
A chatbot on your website may be a great idea. It can be helpful for users if you have valuable content that is appropriately titled and easy for the chatbot to understand and serve up.
“If” is the operative word here.
Here’s a more palatable (no pun intended) example. Imagine asking a waiter for a sandwich. And he brings you:
- Three oranges
- A cup of coffee
- Three cups of water
- A sorbet of some kind
- A…kiwi…brownie?!
Art courtesy of Pixlr AI image generator. Yes, we get the irony.
Where’s my sandwich?
Well, if your desserts are labeled “Food,” your sorbet is labeled “Food,” and the orange is labeled “Food,” and the sandwich is labeled “Food,” then he really wasn’t sure what to get for you.
There’s no nice way to say this: If your
- website structure isn’t clear;
- content is incorrect, jargoned, or disjointed; or
- site lacks any clear actions for people to take
…then your AI chatbot isn’t doing anything but adding more unnecessary and potentially confusing noise to your website interface.
All AI is useless without good content to supply it. Large language models (LLMs) rely on input, and if the input sucks, so does the output.
So how do you solve this?
You (still and always) need a content strategist
Content strategy, especially web information architecture, is like moving houses.
You need to audit what you have. Those decorations in the attic that haven’t seen the light of day since 1988? You can probably get rid of those. Same thing with your 10-year-old PDF documents.
When you move houses, you label your boxes.
When you move websites, you label your navigation.
But if you don’t have labels, or you label them poorly, you’re not going to know where stuff’s supposed to go.
Have fun finding what rooms these boxes go in with… crop circles?… as the labels. Photo from Pixlr AI image generator. The irony continues.
That’s where content strategists come in handy, and then some. Beyond labeling your stuff, a content strategist helps you:
- Audit existing content and refresh what needs updating
- Set language, voice, and tone standards for your content across the site
- Build content models and taxonomy to make content easier to maintain and use across platforms and channels
- Construct a usable navigation that meets your audience’s needs first
- Organize content in that navigation in a sensible structure that’s easy to find from Google search to web search to self-navigation.
- Find opportunities for new content needs based on user research, keyword analysis, and search engine opportunities.
- Test content with real people, identifying where to improve navigation, content, or expected experiences.
Content strategists also work with structured content. Think of structured content like this:
- What pieces of information make up a Google Location profile? Address, phone number, hours of business, and website link.
- What pieces of information make up an article on how to fix your sink? Introduction, tools you need, and steps to follow.
- What pieces of information make up a biography page about a person? Early life, career, family, and personal life.
Structured content has meaning and assigns meaning. It helps untangle the mess of 1s and 0s that make up the Internet so that content can be created, found, understood, and acted upon.
Content strategist Carrie Hane says:
The structure you give to your entities gives the people creating content a guide for what needs to be captured and recorded for your situation. It frees up mental space to be creative within the constraints of the structure.
Structured content doesn’t only make sense of information for people but systems. That includes AI.
Besides all the “things” content strategists organize and document, content strategists are the folks in your corner asking: “Why?”
When does a chatbot work?
According to many experts, chatbots work great for fast, always-on customer service.
If I want to reset my password or have a quick question about shipping, a chatbot is a great first line of defense. It never gets tired. It knows all the answers and the use cases and can get through the process with low variability, high consistency and immediate responses.
So, if a website visitor is looking for content, instructions on a process, a form, or information about managing their account, a chatbot can direct them.
AI is excellent at taking orders. It aims to please and resolve with results, answers, and something that matches your query. But it won’t dig deeper for meaning. And it certainly won’t build a foundation that makes a healthy website grow.
Robots still need babysitters. The chatbot’s user activity, search terms, answers, and conversations should be monitored by a human who can track its successes and failures and, if needed, correct course.
So before you try to make any AI work for you, you must ask why and understand the meaning and purpose of what you’re publishing and pushing into the world.
If it doesn’t have a purpose, neither will your AI.
Your customer experience should come first
At the end of the day, you should base your AI decisions on customer experience, not cutting corners.
Plenty of articles have been published recently about the anthropomorphic nature of chatbots. The more human they sound, the more confusing and frustrating it can be for the actual human on the other side of the screen. Tufts University professional Daniel Dennet has called these machines “counterfeit people.”
Sure, finding a how-to article or instructions for updating a password may be low emotional barriers, but consider services or information that someone may need at times of high stress. Customers deserve human interaction if they need to:
- Find health care nearby for yourself or a loved one
- Get help for a utility bill you can’t pay or need to pay
- Access government assistance in a time of need
- Fix a major banking issue
- Lodge a complaint about an experience
- Try to do anything when feeling ill or under the weather, mentally or physically
Moz has gathered quite a few AI overview fails from Google, which is taking up more screen real estate and becoming the default top search result for users everywhere. And the endless recycling of content that AI uses to provide output means inaccuracies can occur, sometimes detrimentally, like when it lied about an airline refund for a bereaved customer.
There’s a branch of content design known as trauma-informed design, which focuses on minimizing harm, anxiety, undue stress, and other adverse effects through thoughtful language and design practices.
For example, someone who needs to unsubscribe from a service due to a life change, financial strain, or other emotional reason.
Do you work for a credit card company? Maybe a husband is calling to cancel his wife’s card because she died. Do you work for a grocery shopping app? Maybe someone is putting together an online pickup or delivery order for a friend whose child just died.
We don’t know who is accessing our content or what they are experiencing. But we can design in a way that offers them hope.
Read more about trauma-informed principles.
As content strategists, we also collaborate to understand and correct things like unconscious bias, and yes, bias is being revealed across the AI industry. We recommend David Dylan Thomas’ book on this topic.
So what does this mean?
It means you need people to do work for people. Content strategists take content types, voice and tone, language, style, design elements, user research, and more into consideration to create content that is:
- Readable and more easily translatable
- Accessible for all tools and web preferences
- Inclusive to all people, reducing harm and stress
- Clear and concise to eliminate confusion
- Consistent across voice and tone to strengthen brand and user connection
Humans should write for humans. We understand emotions, empathy, and desire for clarity.
But humans also have to set the proper parameters for AI.
Conclusion
AI chatbots can be useful with all those ifs. If a human manages and maintains it. If you measure and monitor the interactions, keywords, and success. If you have a well-structured website that makes it easy for the chatbot to access useful content.
These tools, these machines, can be great helpers in performing tasks. But they can’t do it all.