When the internet came to the car business, it ushered in a radical shift to our business model; so radical, in fact, that many ignored it for a time. Some looked at it as a passing fad and waited for the “internet thing” to be over so we could get back to business as usual. Well, the dot-com bubble may have burst, but the “internet thing” continued to become one of the most prominent drivers of dealership awareness and marketing today. Enter AI…
AI, or Artificial Intelligence, has sparked conversations from the showroom to your dinner table since the introduction of ChatGPT in November 2022. With over 100 million monthly users as of January 2023, it quickly became the fastest-growing application in history, trailed by similar technologies, including Google’s competitive offering, “Bard.” At this point, you’ve all heard about ChatGPT and AI, and you may have even started using it yourself. But is AI a boom or a bubble? When we look back in 30 years, will the winners be the early adopters, or will those early adopters suffer when the bubble bursts and end up at the back of the pack?
While there’s promise in this new technology, there are also countless misconceptions prompting us to apply it where it has no business. But AI is here to stay, so your understanding of the pros and cons will decide which side of history you’re on. So, let’s explore the utilities and the pitfalls of AI as it applies to your dealership and automotive marketing as a whole.
Come with Me If You Want to Live
First off, this article is not for or against AI. This is about arming you with the knowledge to avoid the cons and capitalize on the pros by digging deeper into how AI works and focusing on the parts that matter to you.
It’s Not Really AI: ChatGPT and its peers are not “true AI” (at least not in the concerning way we’ve seen it presented in the movies). There’s no consciousness or understanding going on here. It’s not Skynet, and if you run ChatGPT on the nuclear launch computers, it can’t wipe out humanity. It isn’t anywhere near that advanced.
A Glorified Word-Predictor: Take out your phone and start typing a text. See how it predicts the word or phrase it thinks you’ll use next? That’s actually how ChatGPT formulates its responses to you. It starts crafting a response to your prompt, and it builds upon its own output as it generates it. Why does that matter?
It Hallucinates: Seriously, this isn’t an exaggeration to get your attention. OpenAI, the maker of ChatGPT, says the following on its ChatGPT-4 page, “GPT-4 still has many known limitations…such as social biases, hallucinations, and adversarial prompts.” In this case, a hallucination is a term referring to what happens when the text prediction goofs but tries to keep crafting the response in order for it to read coherently. The technology will keep going in the wrong direction, potentially giving you a blatantly false response. According to studies, approximately 20% of ChatGPT outputs contain false information. So…hasta la vista, factual accuracy.
Outdated Information: A major criticism of ChatGPT and its peers is that the public doesn’t get to know where the information comes from. ChatGPT, for instance, is “trained” on a fixed database. In short, this is a combination of a big collection of things pulled from the internet and also from a private database — whatever that means. When you give it a prompt, it interprets what you’re asking and then forms its answer from that fixed database. The problem? The database was last updated in September 2021, so asking about anything newer will produce flawed or speculative information. Got any cars newer than 2021?
Safe When Muzzled: Increasingly, the problems with hallucinations, biases and false information are leading to OpenAI, Google and others muzzling their AI tools. This means they’re putting increased restrictions on the answers the tools can give you in pursuit of safer and less lawsuit-inducing answers, making the technology less valuable to the average user.
OK, so now we know how AI forms answers and why it can get things wrong. Let’s look more practically at how to use it and where your people should be employing it to improve productivity.
Prompts: There’s No Fate but What We Make for Ourselves
A prompt is a question or instruction you type into a chatbot, and it determines your result. If you type, “Tell me a story,” it’ll generate a random story. Type, “Tell me a story about buying a car,” and it’ll spin you a fictitious yarn about Sam buying a classic car. Type, “Write me an article comparing the pros and cons of leasing vs. buying a car,” and you’ll get something beginning to look like it might belong in your marketing. Read it carefully, though, and you’ll notice it’s a bit…hollow, without any specificity to your store, and not really ready for use. Now, add some instructions to rewrite that article with specific points, and the response will get better and better. The more detailed the prompt, the better the output! Just check the facts, especially if you asked for recent info.
“Prompt Engineer” is already becoming a job of its own in large corporations, and for good reason. With enough feeder information and instruction, ChatGPT and its peers can produce fantastic outputs even when you need something complex and accurate. The usefulness of today’s AI all hinges on whether the cost (time) to create your prompt outweighs the cost of what it replaces. Here’s a constructive example…
Most service department pages are generic and bland. The website vendor just gives you a boilerplate page, and that’s it. Go to ChatGPT and type, “Write me 10 FAQs about a dealership service department.” It’ll bust out 10 pretty good Q&As. Don’t like #7? Just ask ChatGPT to replace #7, and it will. Very quickly, you can have something easy to fact-check and useful to a consumer. You still need to format it nicely on your service page, but many website vendors will do that part for you if you hand them the complete content. This is an example of a use case where a simple prompt and minimal adjustment can go a long way.
Generative AI is a great tool, but there’s no free lunch. Simple projects and repeatable tasks are great applications for AI, but more complicated, one-off projects require just as much time invested with or without AI. Now that you have a better understanding of how AI works and how best to use it, it’s time to start looking at some uses beyond FAQs.
A lot of our dealers rely on us as a consultative sounding board. One of them approached me recently and said, “I want to eliminate either a role or a vendor. What should I be doing with ChatGPT that I don’t need a person or a vendor for?” I’m all for axing snake oil vendors or employees who don’t want to work, so this question is certainly a good one. Unfortunately, the technology isn’t mature enough to replace anyone or anything in automotive wholesale. What it can do is create efficiencies that should make some vendors cheaper or give your people fewer reasons to say, “I don’t have time.” Let’s start with some ideas of what AI can do before we get into costly myths you need to look out for.
Vehicle Descriptions: What’s more compelling for a consumer on one of your VDPs? A block of text with every feature, including power windows and AM/FM radio? Or a well-written paragraph highlighting the most compelling equipment in a human-like tone? The paragraph is much better, and with AI, you don’t need your used car manager to slave away at it — sometimes with questionable spelling and grammar. This is an application for a prompt template. Write your template to include which features are desirable and which aren’t, then paste the vehicle’s details to the end of the prompt. Your used car manager can now create custom vehicle descriptions quickly and repeatably.
How To: Say a customer asks post-sale about pairing Bluetooth or connecting their car to the OEM smartphone app. If your salesperson has to write out email instructions, they’re going to burn 30 minutes on a sale they already made. Just ask ChatGPT to do it instead. A great way to get started on this is to go to ChatGPT and type “How to write a prompt for ChatGPT.” However, it is essential to make sure someone knowledgeable double-checks the instructions. Asking ChatGPT will save time in creating how-tos like this, but it often makes mistakes.
Email Summaries: Do you have a sharp marketing manager whose emails look like novels? Paste the emails into ChatGPT and ask what they are getting at. For single emails, AI can often do an excellent job extracting the key points for you. It also works with articles, but be cautious. I copied this article into ChatGPT, and the summary missed a lot. It showed a bias toward generative AI, highlighting the pros and barely mentioning the cons while getting the order all wrong. The same can happen with more complicated emails or email threads. Just know that the more important the information, the less you may want to lean on AI to give you the key points.
Review Responses: Ideally, you should reply to every good and bad review online to show future customers you pay attention. This can be a time-consuming task to do well, and a lot of dealers let this fall to the backburner as a result. Here, we have another use for a prompt template. Build one for positive reviews and one for negative. Include some guidelines about contact information to include and how apologetic to be. Then, paste the customer’s review at the end of the prompt and let ChatGPT write you a response. This is an efficient way to write complete review replies that offer great optics without taking all day. Just make sure you — sing the chorus with me — check the work before you use it.
Hopefully, this gives you some ideas about what AI can do for you, especially for repetitive, time-consuming tasks. It’s a tool in the toolkit, but we can’t talk about what it can do without talking about what it can’t. There is no shortage of people shoehorning AI into applications where it has no business. Worse, we’re seeing a lot of vendors using this new shortcut to decrease production costs at the expense of whatever quality they had been providing and still charging full rate. This isn’t a generalization either, as we’ve run into verified cases of it.
Maslow’s Hammer says, “If all you have is a hammer, everything looks like a nail.” If your go-to tool is ChatGPT, you will try to shoehorn it into every problem you have. Let’s look at some myths — or at least gross oversimplifications — about what AI can do and some of the poor practices to watch out for.
Throughout the use cases above, there is a common warning that goes back to AI’s limitations. Hallucinations, biases, outdated information and more can affect the response. The more complex the desired end product, the more expertise and time it takes to engineer a prompt to get the job done, and some jobs just aren’t suited to AI at all. At a minimum, AI outputs need human checking and massaging to be fit for publication.
Probably the most blatantly false myth about AI use in automotive is about SEO — which is already a hard-to-understand topic. To be effective, SEO content requires thousands of words providing useful information that’s geographically specific and unique to your store, with baked-in references and links to other resources on your website. Google specifically rewards content that provides unique and clear value to the intended user (car shopper) and does not reward content that is hollow, short or flawed.
At Wikimotive, we offer a free, no-obligation SEO content and organic ranking audit for any dealer who asks for one. I recently conducted one of these audits for a dealer who has been using the same vendor for his SEO for a couple of years. The older blogs from the vendor were short but unique, with a human flair to them. The more recent content, however, felt…hollow. I copied a recent blog into ZeroGPT, an AI content detector, and lo and behold, the blog was entirely AI-generated with no human modification. The vendor’s shortcut was so obvious that on my first try I successfully wrote a two-sentence prompt that produced a near-identical blog, including many identical phrases! I then learned that the vendor was still charging this dealer the same price for the content despite only putting about three minutes of work into each blog. Watch your vendors, especially if they’re producing written content.
Rise of the Machines
Outside of ChatGPT and SEO, you may also be considering AI-based chat tools or image generators. However, a snazzy demo won’t account for the edge cases, and the same concerns apply.
Chat tools are a good application for AI when done right, but we haven’t encountered one that’s convincingly human. Put yourself in the customer’s shoes when the chatbot hallucinates, fails to answer their question, doesn’t address their need or hands things off to a human operator who can. You’ve just lost a lead.
As for AI image generators, they’re great for creating fun images based on whatever you can dream up — but not for vehicle images. A Jeep Wrangler with a JL generation front end and a JK generation back end? AI will unwittingly produce this kind of oddity, and if you don’t pay attention, you’ll be hearing from OEM compliance before you can say “better prompt.”
Simply put…inspect what you expect.
When the AI bubble pops, the technology it leaves behind won’t disappear from our lives or businesses. To quote The Terminator, “This is the world now. Logged on, plugged in, all the time.” So, use it wisely, remembering that every form of AI is prone to flaws, inaccuracies and unintended consequences. Police it, scrutinize the output and find the cost-saving uses. Just don’t let anyone convince you that every problem is a nail.
Do you want to learn more? Scan this QR code in this article for a 25-minute educational video by the author.