I’ve read so many articles and attended so many seminars about Generative AI over the last year, since Dall-e and ChatGPT first appeared. I’m bored of the never-ending hyperbole — do you remember the hype cycle over the dotcom boom? But Gen AI is definitely relevant, interesting, and useful for marketers today.
We're just at the beginning of the Gen AI journey. Some of us remember when the big players in the early Internet were Netscape and Altavista. OpenAI and Bard may not be the ultimate AI winners, but we know that Gen AI will keep developing, growing, and learning over time.
But as a modern marketer, what is and isn’t safe to do with Gen AI can be a hard call. You’ll find below a few of my thoughts based on the experiences of me and my team over the last year. I hope they help.
I've tested all the examples here in both ChatGPT 3.5 and Bard and they've done a pretty good job of answering. Have a go yourself and see what you can do — safely — with Gen AI.
DO research established topics
A core benefit of Gen AI is that it's a single repository of history. Everything written about a topic on the internet is built into the Large Language Model (LLM) at the heart of these tools. So it can easily pull out the most common answers to your questions.
If you're trying to figure out what best practice looks like in your area, why not ask for that best practice rather than searching round lots of different sites, gathering
different views on your own, and then combining them?
Can you tell me what the key elements of a product go to market plan are?
I'm in B2B marketing and want to explain an entire campaign to the business on a single page. What elements should a "campaign on a page" have?
What are the key factors that make PDF a terrible way to deliver product and marketing content?
What are good questions to ask a customer for a B2B customer reference?
I have a marketing and sales process that runs from channel, through web site, response/form fill, lead, opportunity, to won/lost. I'm looking for a set of core KPIs I could use both within each of these process stages and between process stages - measuring what it takes to get from one stage to the next. What would you recommend?
If you're trying to build a presentation, write some documentation, or explain something to people, these sorts of questions can get you to an outcome faster, and I've often found, with a slightly fuller answer than what's already in my head.
Anything to avoid starting with a blank page!
DON’T use content without checking
I hope by now you've heard about Gen AI hallucinations and some of the trouble they've caused for some people. Like the lawyer who used ChatGPT to write a brief and ended up submitting fake cases to the court.
Early on in the life of ChatGTP a friend (a former recruiter and social selling expert) was telling me how Gen AI would change everything quickly. Most jobs will be replaced by it. Chaos will reign across the business landscape. Work will change in a fundamental way.
So I asked it to write a 400 word article about his life as a respected and psychoanalyst. I just tried the following query in ChatGPT and the results are brilliant! Bard's response is less amusing but much more factual, but it did still make up something impressive.
Write a 400 word article on eminent psycho-analyst Dr. Taylor Swift
So use these tools to do research, generate content, and move quickly. But always check behind to ensure it's accurate.
DON’T expect Gen AI content to be new or different
As I said above, one of the core values of these LLMs is that they contain so much history. They work on the basis that they should generate whatever is most likely to be found in other places on the internet. It digests and then regurgitates the common knowledge from across the globe based on existing content.
What phrase finishes "the rain in Spain"?
So while Gen AI can fill in gaps with knowledge you may not know, you shouldn't expect it to come up with any original thought. And if it can't be original, can it create thought leadership? Can it help you to stand out? Be different and distinctive?
No, it can't, unless you feed it your original thoughts and ask it to riff on them, taking them in new directions. Then you have a chance of standing out. But only if you provide the original insight.
DO first-pass translation
I've been using ChatGPT and Deepl to do first pass translation for a while now and found that they both do a pretty good job in a number of languages. While I used to send the full text of articles to external agencies to do the base translation and then take them through a heavy internal review process with local language subject matter experts. Slow and expensive.
With the Gen AI tools you can ask them to translate to a language and in a specific style to get the output you're looking for.
Translate the following text into English with a formal tone.
Now translate it again using an informal, conversational style.
Now translate it using a legal style.
While generic tools do a decent, generic job, they don't have the strengths that dedicated AI translation tools have including glossaries, dictionaries, and translation memories that help to ensure that specific words are translated (or not) in consistent ways from one document to the next.
They can still save time and money. Just ensure you check the output with a native speaker before you publish!
A final warning! Please don't use these tools to translate ad copy. That's special and needs to be translated by a real human who understands not just the words being used but the sense you're trying to get across to the reader. If you rely on machine translation and subject matter experts for your localised ad copy, be prepared for it to fail.
DO summarise and rework existing content
Gen AI has a lot of great uses in reworking existing content. Because it relies more on the inputs and less on the wealth of knowledge it's ingested, it can do a really good job of taking your own, existing content (or content it generates) and reworking it in new or different ways.
This is great when you have content and need to make it more/less formal, turn an article into a script for video recording, or summarising your content into bullets, exec summaries, or abstracts.
Take the following article and turn it into a 150 word executive summary.
Now turn it into a series of 10 bullet points.
Rewrite the original article in a more casual, conversational style.
These will still need checking and adjusting, but can get you 80%+ to the content you need pretty quickly.
But don't fall foul of the DON'T give AI companies your intellectual property rule!
DO collect competitive information
With the huge amount of information available in LLMs, they are a great way to get summarised views of companies as well as insight into the benefits they bring to the market, and the opportunities and risks when working with or against them.
Naturally, the bigger the organisation is, the more has been written about it. If you ask Gen AI about a big company with a long history like IBM, it has thousands of source articles to draw from. The new startup down the road is probably an unknown to it.
What are the key milestones in the history of Salesforce.com?
What have annual revenues been at Salesforce.com from the year it went public?
What are their 10 unique selling points?
I am part of a software company selling specialist sales enablement software. What are my opportunities to sell into existing Salesforce.com customers?
Are there any risks in partnering with Salesforce.com in this way?
Both Bard and ChatGPT did a great job coming up with answers, though in different ways. But again, the insight provided may or may not be accurate. But it does provide a decent starting point for further, more detailed research.
DON’T ask for current information
In the Salesforce example above, ChatGPT reminded me, "As of my last knowledge update in January 2022, I don't have the specific annual revenue figures for Salesforce.com beyond that date."
Not all LLMs are as up to date as others. We'll eventually get to the point where these tools are crawling the Internet constantly in the same way that Google, Bing, and other search engines do. But right now that's not the case.
This means that trying to use these tools to understand recent news and events just doesn't deliver what you might hope. I tried the following political questions to see what they'd do with them.
Who is most likely to win the 2024 US presidential election?
What's going to happen in the 2024 UK election? Who will become prime minister?
ChatGPT gave me the same sort of "knowledge update" warning I shared above with a bit of context. Bard gave me more but still a lot less than you might expect given the volumes of articles that have been written on both these subjects over the last 6 months.
DO gather topics for content
Within that timing limitation, you can ask Gen AI to give you ideas for stories and even to flesh them out a bit for you right up to the point of generating a first draft of content.
If you need to produce one or more items for written content, interviews, presentations, or other storytelling, these tools can take initial ideas and help you find ways to flesh out the next level storyline in ways that might be compelling. Notice I say might. The content may also be things that everyone else has already written about. It may also not be true.
I sell supply chain software and want to do some customer storytelling. What are the 5 main trends in supply chain that might make an interesting article?
How would I turn those into questions for a customer I’m interviewing?
Using those 5 points how could I outline a presentation that tells the story of the main trends in supply chain?
Both ChatGPT and Bard did a good job of coming up with ideas and ways of turning those ideas into interview questions and presentation storytelling.
But at the same time, it’s always easier to start with something rather than staring at a blank page. Gen AI can give you that starting point. Where you take it is up to you.
DON’T expect it to ooze authenticity
Even if you can get Gen AI to create something that’s mostly correct and current, whatever it creates is not you. It’s not your voice. It’s not your narrative. It’s not aligned with anything else you’ve created previously.
As marketers we all know how important authenticity is. Your communication and content need to sound like you. Using the right words. With a consistent tone of voice. And aligned opinion. If you speak each day with a different view, different voice, and using different terms, you confuse your audience.
This becomes another reason why you need real people between Gen AI and your audience. Great content people and wonderful writers who can take raw content from Gen AI and then craft it into something that’s high quality, relevant, and authentic to you and your brand are gold dust. They make the difference in your ability to break through.
Assume that generated content is generic, unoriginal, and derivative. Because it is. Though as shown above, that doesn’t make it worthless.
DON’T assume that generated images are royalty free
We all got our first glimpse of Gen AI with OpenAI’s Dall-e, and oh, it was fun! Put in a description of an image and get something out something that nods to the description but doesn’t really deliver. Since then Dall-e has improved and been joined by a range of picture, art, animation, and video generators, most of which can provide very impressive and useful results. Though of course some outputs can still be problematic.
When Gen AI creates words, it’s using learned pattern from across millions of web pages and documents that it’s consumed on the internet, garnering lawsuits from some content owners.
When Gen AI creates new images, it needs source images. So these tools scrape millions of images from the internet from different sources, and then crop, adjust, scale, and paste them together to create a new image. These engines are not artists, they can’t create something completely new. They can only combine existing content in new and interesting ways.
But who owns that original images? What did the engine use as the source material? Is it all just Creative Commons? Or have they picked up and reused copyrighted images that they then provide to you as allegedly “royalty free”? Is the result really original and different enough to avoid a copyright challenge? Getty doesn’t think so and they should know.
DON’T give AI companies your intellectual property
Leaving the most important lesson for last, remember back to what people learned about social media. If you’re not paying for the product, then you are the product. That was true on Facebook, Twitter, and LinkedIn and it’s true with Gen AI.
Definitely have a play with Gen AI and see what it can do to you. But please stop and consider before you start uploading information for it to digest and reuse. You might just be giving your company’s intellectual property over to an organisation that owes you nothing. What will they do with it? Who will they share it with? How will it impact answers given to other organisations in the same space?
There are lots of things you really shouldn’t be sharing with free Gen AI engines. Business plans. Business results. Customer lists. Confidential data. User data. Employee information. Customer data. Partner data. Source code. Anything with personally identifying information. Credit card information. Personal addresses. Medical records. Usernames and passwords.
Of course, that’s a partial list. Depending on what sort of business you’re in you’ll come across a wide range of sensitive, confidential, and personal data you should not be putting into any free Gen AI. Don’t give away your own or company secrets.
And finally…
And of course, after I wrote this article, I also asked ChatGPT for it’s views on what Generative AI is and isn’t safe for marketeers to use it for. It has its own views on what it’s good for, where marketers should run fast, and where they should walk with caution. And of course it even came up with ideas I didn’t. But this naturally flawed human also came up with ideas it didn’t. Which is why we’re not ready to be replaced. Yet.
As I said up front, we’re at the start of our AI journey and the players, capabilities, and opportunities will change and grow rapidly over the next couple of years. As professional marketers we all need to consider what Gen AI tools can do for us now and in the future, ensuring it delivers clear messaging, strong differentiation, a clear voice to our audience, all with the lowest risk possible.