A guide to using AI chatbots
By now, you’ve no doubt had a play with ChatGPT, the AI-driven chatbot which can debug and write code, give you a list of ideas for your next party theme, explain quantum physics to you in a pirate voice or write a rejection letter as if it were your personal HR division (among other things).
Across the content industry there have been mixed reactions to ChatGPT and its competitors. As expected, some are focused on the risks, while others champion the benefits. Is AI coming for our jobs? Will it make our jobs easier? Will it fuel the spread of disinformation? Will it lead to widespread plagiarism? Can we just ignore the whole thing entirely?
In answer to the latter question, Bill Gates recently described the development of AI as “fundamental as the creation of the microprocessor, the personal computer, the Internet, and the mobile phone”, while LinkedIn Co-founder, Reid Hoffman, has declared that ignoring AI is “like ignoring blogging in the late 1990s, or social media circa 2004, or mobile in 2007.”
Whatever you think of it, AI is here to stay, and it looks set to transform the way we live, work and communicate. So let’s unpack some of the ways we might be able to utilise AI chatbots as content professionals.
A new era: Who’s who and what’s what
There are a lot of developments happening in the AI space and keeping up with it all can feel like trying to learn people’s names at a Fitzroy house party. So here’s a rundown of the main players as it currently stands (mid-2023).
Chat GPT
First is ChatGPT, the free, publicly available and most well-known large language model AI chatbot that made headlines when it was quietly released in November last year by a company called OpenAI. Why? Because it was the most sophisticated, useful AI chatbot the world had ever seen. Originally intended to be used for research purposes to help build their next model, it went viral, reaching 100 million users in two months (something TikTok took 9 months to achieve). We’ll cover off more of what it and other platforms can do later in this article.
GPT-4
If ChatGPT is your everyday passenger vehicle, GPT-4 is the luxury, subscription model. Probably the most astonishing thing about GPT-4 is that it’s multi-modal, meaning it can process images, not just text. It can’t generate images (like DALL-E or Midjourney), but it can look at an image (photograph, illustration, sketch) and understand it. This is a huge breakthrough. However, at the moment the image capability of GPT-4 has only been given to one partner - a charity that uses images and AI to help those with vision problems (more about that later).
Apart from this incredible capability, Open AI claims GPT-4 is more advanced than ChatGPT in other ways too, such as longer and more precise answers, more attention to context, and its capacity to collaborate creatively on projects including music, video games, screen plays and technical writing.
They’ve even claimed it can learn style and tone of voice, which could be really useful if you’re having to write volumes of copy or you have a number of team members working on a project where consistent voice is important.
Currently, the biggest issue with GPT-4 is that it can have ‘hallucinations’ where it makes things up and can be completely wrong with its predictions. But OpenAI says it’s constantly working on the issues, which is reassuring, but certainly something to be mindful of regardless.
ChatGPT Plugins
OpenAI recently launched plugins, a bunch of software applications which can harness the power of ChatGPT. For example, Expedia has added ChatGPT features to its mobile app to provide an interactive experience for users when trip planning. Plugins will see the capability of ChatGPT broaden as various software companies take advantage of the chat functionality to increase the usability of their tools even further.
Perhaps the most significant plugin, at least for content professionals, is the browser plug-in. Currently available in beta mode in the paid version of ChatGPT, it connects the chatbot to the internet, essentially making it a powerful search engine. No longer is the chatbot limited by the data set it was originally based on, but it can trawl the web for data as well.
Although new for ChatGPT, search engine capability has been available through competing chatbots, Microsoft Bing and Google’s Bard, for some time.
Bing Chat
Bing Chat is a Windows web search engine chatbot, currently powered by GPT-4 and available in Windows Edge, Microsoft’s browser. Until recently, its main drawcard was being connected to the internet, but with ChatGPT launching its browser plugin and Google integrating AI into its search function, Bing isn’t quite the stand out it has been.
Having said that, ChatGPT’s browsing function harvests its content from the web using the Bing search API, and BingChat runs on GPT-4, so they’re heavily reliant on each other’s technology. (Windows has invested billions in OpenAI so far.)
Google Bard + AI Search
Google recently launched its next generation language model, PaLM 2 to power Bard, which Google claims has “improved multilingual, reasoning and coding capabilities”. It’s also opening up Bard to 180 countries, including Australia. Like Bing and GPT-4’s browser extension, Bard is connected to the internet, and there’s a "Google It" button next to responses when you use Bard that takes you to Google Search.
Like other chatbots, Bard can generate text, translate languages, write different kinds of creative content and answer your questions. It can also write and debug code in a variety of programming languages.
Something it can do which is pretty clever and unique is analyse websites. You can ask it to analyse a website’s usability and accessibility, SEO, readability and more. The findings and recommendations aren’t detailed and not as helpful as a specialist piece of software or a consultant, but it is a good starting point. I’ll expand more on this when I get to ways of using the technology below.
At the same time, Google is building a generative AI feature into its search engine. This will mean Google will package up search results for users, essentially doing the hard work of collating research for you, while still offering Google Ads and regular search results. The new Google Search will stop short of being a chatbot though, leaving that to Bard.
Now that we’ve surveyed the main players, let’s shift our gaze to how we might be able to wisely harness the power of this technology as content professionals.
Using AI as a content professional: Quantity over quality
AI Chatbots are intoxicatingly quick and sophisticated, but it’s important to note they shouldn’t be used without a human to oversee, refine and build out what they produce.
The beauty of the technology is the speed and quantity of responses a chatbot can deliver, not necessarily the quality. Ask for a large quantity of results and use what it throws up as a starting point for your own thinking rather than taking what it generates without critical thought.
One way to see it is that chatbots are a helpful creative collaborator, rather than a straight out content generator.
Need a catchy heading or some sharp UX copy? If you’re struggling to generate enough ideas, or are stuck and need to kickstart your thinking, a chatbot can come up with a huge range of options in a few seconds. From there you might come up with something completely original or hone what’s been suggested, but the AI is able to give you a leg up.
For the purposes of this article, I asked GPT-4 to generate 10 error message options that played on the theme of boating or sailing. It delivered, in a matter of seconds the list you see below, which, to be frank, isn’t half bad.
- "Lost at Sea! We can't find the page you're looking for. Time to adjust your compass."
- "Bermuda Triangle! The page you are searching for seems to have mysteriously disappeared."
- "Shipwrecked! This page appears to have been swallowed by the digital deep."
- "Off the Map! The page you're seeking isn't marked on our charts."
- "Marooned! We can't seem to dock at the page you're seeking."
- "Abandoned Ship! The page you're after has been left high and dry."
- "Tossed Overboard! The page you're looking for is not on board."
- "Dead Calm! Your page couldn't be found. It's like there's not even a ripple in the water."
- "Walk the Plank! Yarrr, the page you be seekin' be gone to Davy Jones' Locker."
- "Anchor's Away! But unfortunately, your page isn't anchored here."
What I found most interesting about asking GPT-4 to do this task is it’s been programmed to maintain clarity (at least in theory). At the end of the list is this note:
These error messages playfully incorporate boating and sailing themes while maintaining clarity and communicating the issue at hand.
I also asked it to generate some error messages in a wry cheeky tone, and it gave this wise warning:
When using a wry, cheeky tone, it's important to strike a balance between humour and clear communication to ensure the error message is still informative and helpful to users.
Of course these error messages could still be misunderstood and you’d have to analyse them through a UX and accessibility lens, and potentially do some testing, but for a quick brainstorm of some punny ideas, it’s pretty good.
Be specific
Try and make your prompts specific, as a chatbot is only as good as the information you give it. And again, focus on the quantity of responses, because 99% of them will be awful.
I recently asked GPT-4 to generate 20 ideas for the name of an app which helps people make a fire plan in preparation for a bushfire. Here's what it generated in a matter of seconds:
- Blaze Blueprint
- Ember Escape
- PyroPrep
- FlamePlan
- FireSafe Strategy
- Inferno Inoculator
- SparkSecure Planner
- Cindersafe Coach
- Conflagration Compass
- FireRoute Framework
- Wildfire Wisdom
- Blaze Breakout Plan
- SmokeSignal Strategizer
- FireFortify Formulator
- Ember Evacuation Expert
- Combustion Codex
- Inferno Initiative
- FireShield Framework
- Scorching Security Planner
- BurnBattle Blueprint
The majority of the results are pretty bad. But narrow down your prompt, and you get a much better set of results.
New prompt: "Come up with 10 possible names for an app which helps people develop a fire plan, so they survive a bushfire. Come up with names that have two words, one of which is fire."
- FireGuardian
- EmberAssist
- FlameForesight
- FireEscape
- SafetySpark
- FireBlueprint
- TorchTactics
- FireShield
- FlameFortitude
Aside from not consistently following my instructions to use the word fire, it's done a much better job with a narrower set of instructions.
Creating outlines
If you’re writing an article or presentation, AI can help you generate a structure or outline to kickstart your thinking. Sometimes having a rough structure or outline is a helpful prompt to get writing, even if the structure or titles end up changing.
You can then ask it to expand on particular points in the outline if you want more to work with. A word of warning though, to always have original thoughts (this should go without saying!) and to cross-check the outline to ensure it’s relevant and accurate. A chatbot is only going to provide you with bread and butter responses, not thought-provoking, challenging and cutting-edge ideas. It won’t push the boat out; you have to. And the more personalised, specific and timely you are, the more engaging your content will be (all of which even a very clever chatbot will struggle with).
Research
Content professionals often need to research audiences, user behaviour and trends. AI can be used to uncover this information by analysing large amounts of data you provide it with, or if it has access to the internet (such as Bard, Bing or GPT-4) it can aggregate information from the web on a topic or trend.
To test its research capabilities out, I asked Bard to outline the car purchasing trends of millennials and here is what it delivered. Of course it’s important to check what a chatbot tells you, and you’d ideally source direct research findings to verify its claims and undertake proper user research, but a chatbot can provide some helpful high-level research at the start of a project to guide further research and thinking.
Car purchasing trends of millennials (according to Google Bard):
- They do their research online. Millennials are more likely to do their research online before they even step foot in a dealership. They use a variety of resources, including online reviews, manufacturer websites, and social media, to learn about different cars and find the best deals.
- They are more likely to buy used cars. Millennials are more budget-conscious than previous generations, so they are more likely to buy used cars. They also tend to be more environmentally conscious, and they see used cars as a more sustainable option.
- They are more interested in fuel-efficient cars. Millennials are concerned about the environment, so they are more likely to buy fuel-efficient cars. They are also more likely to live in urban areas, where fuel efficiency is more important.
- They want cars that are connected. Millennials are tech-savvy, and they want cars that are connected to the internet. They want cars that have features like built-in navigation, smartphone integration, and remote start.
- They want cars that are personalised. Millennials want cars that reflect their personal style. They are more likely to buy cars that are unique or customised.
Overall, millennials are a different breed of car buyers than previous generations. They are more informed, more budget-conscious, and more interested in sustainability. As they continue to become the dominant car-buying generation, these trends are likely to continue to shape the automotive industry.
Website and content analysis
As a content professional, it’s not uncommon to need to analyse content for its accessibility, readability or usability.
I wanted to see if an AI chatbot could help with these tasks, and asked Google Bard to analyse my photography website for its readability, accessibility and SEO.
Particularly when analysing for SEO, I found it picked up a hugely inaccurate Title Tag and meta-description, but offered some good general principles and recommendations. It doesn’t come close to a detailed, tailored report from a professional, so the lesson, as always, is: use with caution, and cross-check with your own knowledge of best practice, other specialised accessibility checkers and subject matter professionals.
Summarising + condensing text
It’s possible to ask AI chatbots to summarise, condense and clean-up copy, such as a press release, article or web page. In fact, the paid version of Chat GPT, GPT-4 has a Linkreader plugin which can be used to read and summarise the content of all kinds of links, including webpages, PDFs, images and more. GPT-4 also has a Video Insights plugin which can generate transcripts and metadata, which could save a lot of time.
A word of caution though: if you decide to give a chatbot something to summarise, it’s important to ensure the most important points are covered and make sense, as AI relies on the structure of the content itself to determine importance and could overlook nuance or detail a human would include.
Proofreading + re-formatting
AI Chatbots can proofread text and make changes across a document such as removing all contractions, adding in contractions, changing text to British English, or applying a particular format to the content (bullet points, table etc). This could be a huge time-saver ahead of human proofreading and checks.
Playing with voice + tone
A fun way to engage with a chatbot is to ask it to write in a particular voice or style. It’s a low stakes way to play around with tone and voice, particularly if you’re writing in a voice far from your own. Even if you don’t go on to use what it generates, it can help get your brain in the zone and get your synapses firing so you go on to write well in that voice and style.
A creative thesaurus
We’ve all found ourselves snagged trying to think of another way to say something. Chatbots can be a helpful de-snagging device. As an experiment, I asked GPT-4 to give me 10 playful alternatives to the word “sneakers”.
Here’s what it came up with:
- Street Stompers
- Pavement Pounders
- Sidewalk Striders
- Urban Gliders
- Tarmac Trotters
- Foot Fliers
- Kickin' Kicks
- Stride Sliders
- Groove Groovers
- Bounce Boosters
Once again, the responses are of varying quality (tarmac trotters?!) but even just reading through a list like that could help get your brain buzzing to come up with a creative phrase for your next piece of content.
Improving accessibility
Open AI has partnered with an organisation called Be My Eyes, which connects volunteers with people needing help with everyday tasks, like identifying products or navigating an airport. Open AI has given them access to GPT-4 technology to create ‘virtual volunteers’. The AI can not only recognise what’s in the image, but have a conversation about it. This will be game changing for people who have low vision or are blind.
As far as content accessibility goes, the image recognition function will be useful for generating meaningful and accurate alt text for screen readers once it’s available to more users.
What you put in is what you get out
One thing is clear when it comes to AI chatbots, of any kind: their answers are only as good as the prompts and content you give them. They can’t strategise like you can. The more specific and detailed you can be with your instructions, the better the results will be. Try to provide chatbots with details like word length, tone of voice, format and any other information you can to give it the best chance of meaningful, unique copy.
Summary
There's no doubt AI chatbots are changing the way content is created, modified and consumed. But however powerful and increasingly sophisticated they are, they equally bring the potential for inaccuracy and bias. They also lack the nuance, context sensitivity, creativity and quality control of a human, but that doesn't make them redundant.
If seen as a creative collaborator and a time saver, with a focus on quantity of responses not quality, AI chatbots can be harnessed as a time-saving tool in the content professional's toolkit, rather than a replacement or a usurper.
Let's work together
Get in touch