Issue #20: Hacking Google with AI, Prompt Engineering, Generative AI ads

Good morning.

Would you like to “hack” Google’s search results with AI?

I’ll refrain from giving you the exact instructions, as I don’t want to see people abuse it.

But it’s possible to do so.

Example: For the second time in just one month, Google's search engine has allowed an AI-generated fake of a famous artist's work to rise to the top of its search results.

It’s already happened to the American realist artist Edward Hopper (Google fixed the issue).

Who knows if Google (and Bing) algorithms will ever be able to detect an AI “fake” for things like this. I doubt it.

It does open up a lot of possibilities for black or gray hat marketers to potentially game the search algorithms.

On the topic of AI’s impact on search and SEO, this article by Jon Cooper is essential reading:

Even if you’re not directly working with SEO, you should read it.

Now, about Deepfakes. The 2024 election in America will be… interesting.

Politicians elsewhere have already used Deepfakes to win voters. But what Generative AI is capable of now, in 2023, is beyond what was possible even in 2022.

And it’s not going to degrade in quality.

Now, back to marketing (the reason we’re here).

You could easily create a Deepfake of a company spokesperson—or even of your CEO or founder.

This can become personalization at scale, too. Every customer who spends a certain amount gets a “personal” video from the founder.

Generating video and audio content (even images) is incredibly easy and fast. No need for photo shoots, spending hours recording videos or podcast episodes.

One of the best tools on the market right now is HeyGen.

I suggest you give it a try to see what’s possible—and make some pretty good video content for your marketing campaigns, sales or how-to videos.

The good news is according to one study, about one-third of people can’t tell the difference between a human or AI. Other studies show it’s a 50/50, which means people are just guessing. So you’ll get some mileage out of your company Deepfakes.

In today’s issue:

• The truth about Prompt Engineering (what it is, what it’s NOT).

• How to evaluate the performance of ML/AI Models.

• Generative AI for ads (forget Web3 and Crypto).

• AI and machine learning applied to customer experience.

• Google making AI-powered advertising a thing for everyone.

• Immersive "AI room" experiences for better customer experiences.

Let’s dive in.

BIONIC JUICE

You’ve probably seen the job of a “Prompt Engineer” thrown around online—and how some companies are paying north of $335,000 per year for the role.

While this is a newly evolving skill in the job market, it’s an intricate skill that requires finesse and strategic thinking, and it’s starting to see high demand.

And now companies are paying big bucks for those who can do it.

So, what is a prompt engineer?

There are two versions of this:

What some online marketers say it is.

And what it actually is.

Some marketers will say that being a prompt engineer is crafting “good prompts” that are often very long (2,000+ words) and complicated. I’ve seen sales pages for this and it’s BS.

That’s not what it is. It’s a marketers version of a buzzword and it’s wrong.

Long prompts can work—but it’s not prompt engineering.

A legit prompt engineer uses a combination of code and language to get the exact results they need from generative AI applications.

Sounds simple, right?

Not quite.

It involves:

• Designing, testing, and optimizing prompts (often on a code level, interacting via API).

• Using advanced prompt engineering techniques (few-shot learning, ReAct, chain-of-thought, tree-of-thought, etc) that enable emotion analysis, information extraction, chatting with your data, and more.

• With tools like LangChain, DUST, OpenAI Python client, and many more.

In other words, learning it and doing it covers:

Mastering prompt engineering is a nuanced skill, and no matter how good your English is, it requires an in-depth knowledge of neural networks, natural language processing, and analytical thinking.

It’s not writing a long, complicated prompt for getting emails generated.

It’s complex and takes serious brain power (and patience).

Now, as a marketer or copywriter, you don’t need real prompt engineering to get good outputs. It can be insanely useful to understand the tech—but you do NOT have to.

I’ve spent a few years learning the tech side (including Python) but it’s not required of you to do so.

One of the best mental models for you to adopt is:

Imagine ChatGPT is a junior marketer or copywriter. Then imagine what kind of project brief you’d need to give him or her so they could do a good job. You need details, information, context, definitions, examples, and so on.

Give ChatGPT what you’d give them.

And ask ChatGPT to refine the output (tell it what to correct, do differently, and cut out).

That’s basically every PrOmPt CoUrSe out there in a nutshell.

True prompt engineering is more than formulating fancy GPT talk. It also involves measuring your current AI model’s performance.

Lucky for you, Sara A. Metwalli breaks it down in an article on performance evaluation. It’s a one-stop solution to fine-tuning your ML/AI models.

She breaks down how to:

• Split datasets.

• Define metrics.

• Validate and tune.

• Iterate and refine.

But it's not just about learning the theory behind these metrics. You'll also gain practical knowledge on how to apply those insights specifically for marketing.

With various evaluation techniques to aid any struggling messaging strategist, validation testing combined with OpenAI’s vast toolkit allows users to decode data to generate performance-optimization strategies in minutes.

So, don’t bother buying some flimsy internet marketer “course” on prompt engineering. You can learn about it for free here. 

LATEST DISCOVERIES

Looking for AI ad inspiration?

Check out Coca-Cola’s magical new ad. It’s packed with creativity.

Anyone who’s ever stepped foot in the MoMa or Louvre will instantly recognize the references to Hokusai, Picasso, Andy Warhol and Van Gogh, enchantingly set to classical music.

It might be one of their best commercials yet.

And it was made with the help of StableDiffusion.

It comes after Coca-Cola’s global creative strategy exec, Pratik Thaker, urged marketers to focus on AI over Web3 for lasting audience impact—a serious blow to the Metaverse and crypto movement.

He made the announcement at Ad Age’s Web3 Marketing Summit, no less. And, to his credit, his prediction was correct.

Instead of showering customers with digital collectibles, Coca-Cola is using AI to assist and engage with tangible, valuable applications.

Their latest campaign, Create Real Magic, featured a free-to-use image generation bot, where artists were invited to create digital assets with iconic Coca-Cola motifs.

Each submission was entered into a contest, from which 30 winners were selected to attend a three-day marketing workshop at Coca-Cola's headquarters.

The real cherry-cola? Their work was featured on prime digital billboards in Times Square and London's Piccadilly Circus.

Talk about a career launch! (and free creative ideas—smart play Coca-Cola).

They’re not the only beverage giant embracing generative technology with great success.

Ogilvy Canada took the Swedish vodka label Absolut to the Hood.

Their clever use of AI mixers (shaken, not stirred) delivered a sense of identity, creating a high-end interpersonal marketing effort that only Ogilvy can serve.

The Mix Your Hood campaign saw the creation of over 50 custom cocktails for a variety of Canadian neighborhoods, designed to celebrate the nation’s rich diversity and individuality.

The commercial was recently crowned The Drum’s Ad of the Day.

This is proof that personalization is where AI truly thrives.

Take a look at Walmart. They’ve already been using algorithms to track their in-store stock to reduce the agony of scouring the shelves.

And they aren’t stopping there.

Every crumb of data, from the time of purchase to the average order size will be used to create a cheaper, faster, and easier digital grocery store haul for you.

They intend to continue leveraging AI—as they should. And their partnership with Pactum AI is a significant upgrade to the virtual shopping experience.

You don’t have to be a big brand to do any of this, either.

Ads, images, video, capturing and mining data—you can do it all with just a few tools, even if you’re a small business or marketer with a small team of one or a few.

One way is by using first-party data analysis. Marketers can use a customer behavior perspective to get a competitive advantage.

Omnichannel retail platform Zitcha calls it a Data and AI Maturity model.

They created the graph to demonstrate how easy it is to get the upper hand in your market to multiply and maintain your consumer base—at an infinitely scalable rate.

All it takes is some clean data stacked with customer satisfaction.

Customer behavior preference analysis allows brands to offer personalized recommendations, improve customer satisfaction, and build loyalty.

It’s that simple:

Satisfy, multiply, and maintain your consumer base.

And Machine Learning, even at a basic level, can help you do that.

It’s not just about “crunching numbers and data.”

It’s about applying these tools in your particular business context.

Upwork, for example, is recruiting Machine Learning experts en-masse for an upcoming AI-powered matching system.

Specifics are currently few and far between. But according to the ad, they aim to deploy the tool within the next three months.

They’ll probably use a similar model to track potential recruiters’ requirements to source the best possible freelancer for each project.

But when it comes to a platform like Upwork, with hundreds of niches and skill sets, it’ll require a step further than simple data to create a truly intuitive product.

My prediction? They’re going to need Multimodal AI.

That would allow hyper-specific, super-intelligent search functions to pair talent with demand.

For those needing a quick turnaround, average response time and delivery rate data could recommend the best marketer for your project niche.

Alternatively, an image-supported search function would allow writers or app designers to browse by artists, niche, style, or education.

What an exciting prospect:

An AI that generates a proof of concept, then matches clients with artists who can bring that vision to life.

We already know AI-generated visuals are gaining serious traction—especially in digital advertising.

And of course, Google dropped a brand-new AI-powered advertising project for YouTube.

Word on the street is that they’re developing a Stable Diffusion-like tool to help creators design and launch the perfect commercial for their audience.

The app will generate potential topics, titles, descriptions, and even visuals for each video idea, streamlining that creative process.

An amplified recommendation tool will likewise bring that crucial personalized viewing experience, connecting creators with audiences craving their next fix.

This development is just another piece of the puzzle Google is strategically piecing together.

This AI-recommendation algorithm is poised to reshape the web experience for each and every user.

As personalization reaches new heights, it's important to stay vigilant.

The potential risks and challenges are inevitable as AI evolves and many “godfathers of AI” seem to be chiming in on their concerns.

Leading start-up founder Emad Mostaque recently warned that without proper controls, artificial intelligence will only get "crazier and crazier".

Considering Stability AI’s recent partnership with Google, I’d wager they know what they’re talking about. Their founder's warning serves as a timely reminder that as we embrace the power of AI, we must also prioritize responsible development and implementation.

I also see the doom and gloom we have been seeing as a smokescreen for much larger world issues, but I digress.

Remember, knowledge is power.

Stay informed, stay tuned, stay focused, and let's navigate the exciting yet complex world of AI-driven marketing together.

Funding:

$28M: Course5 Intelligence, an analytics company, secured a huge $28 million to enhance its suite of AI-powered market intelligence. They employ that all-important AI and Data Maturity model to fuel hyper-effective research operations.

$19M: Revolutionizing the AI landscape, Union.ai receives $19 million in funding to launch Union Cloud, an innovative platform that converges and simplifies AI, data, and analytics. It reduces time to market and offers significant cost savings.

$6M: McAfee has a new competitor in town. Lumeus.ai unifies connectivity and control from campus to the cloud with AI, empowering digital security teams with enhanced control and seamless connectivity and security measures.

BIONIC TOOLKIT

Most companies and marketers I speak to are concerned about AI being used for all kinds of fraud (customer, social engineering, hacking your company, and so on).

And, in a world about to be ruled by human-brain-mimicking robots, you’re allowed to be nervous.

It’s all too easy to fire up ElevenLabs and synthesize the voice of your accountant, systems administrator or even your grandma and ask for a quick cash injection over the phone.

Likewise, ChatGPT’s coding ability has transformed a whole generation of duplicitous deviants into fully-fledged hackers.

That’s why you should look to safeguard your business.

🤖 Facia is a cutting-edge platform specialized for the AI age.

Advanced facial recognition technology ensures the authenticity of individuals, providing businesses with a robust defense against identity theft and unauthorized access.

Likewise, the platform can effortlessly verify the identity of customers, employees, and users in real-time, preventing fraudsters from infiltrating your systems and compromising your customers’ precious accounts, and your reputation.

At the same time, there’s a huge risk and danger with this. While facial recognition has some place and use, for some types of companies—largely, and for society, it’s a bad, bad idea. Regulation around this is virtually non-existent, unfortunately. I hope that changes.

Ask yourself:

Do you want to walk around anywhere outside and have your face scanned and identified in real time—so that whoever is looking can keep track of you at all times? The potential for abuse and control is truly unparalleled.

This kind of technology does not need to become ubiquitous or even used much—not even by the TSA for flying (you can opt-out of their facial recognition trial run and I suggest you do).

What’s a better use of your face? When live streaming went mainstream, everyone got on board.

From Twitch to Facebook Live, platforms worldwide rolled out tools to connect with audiences in real time.

But, for those wanting to host a video conference on their own site, live video is an expensive investment.

Between equipment, software, coding knowledge, and network support—it’s hard to jump into hosting a video conference with the board or stream your latest launch for your social media following.

 🤖 KITT saw the gap and is helping those that want more personal integration in the world of live video experiences.

This powerful tool is part of the LiveKit suite, designed to create video conferencing, live streaming, robotics, and even Metaverse experiences.

It’s perfect for all means of communication, be it breakout rooms mid-meeting or a digital classroom for your latest workshop.

A cloud dashboard, status updates, developers documentation, and top-notch security tools are included with every account, so you can monitor and please the crowd, all at once.

Best of all, KITT and LiveKit are free to use.

Speaking of immersion:

Imagine you're a marketing agency promoting a fitness apparel brand.

Wouldn’t it be super if you could simulate a real-life shopping experience in a sleek, stylish showroom, taking those customers out of their web browser and into a captivating virtual room?

 🤖 With Rooms.xyz, you have the power to bring that stretchy pants and dry-tech muscle shirt experience to life.

Picture this: your browser is transformed into a state-of-the-art fitness studio, complete with cutting-edge equipment, motivational posters, and energetic music.

Users can explore the room, interact with virtual trainers, and even try on the clothes.

As visitors immerse themselves in this dynamic environment, they get a taste of the brand's vibe, values, and product offerings.

It's an engaging and memorable experience that leaves a lasting impression and boosts conversion.

By creating immersive rooms that resonate with your target audience, you generate excitement, build brand loyalty, and make sales.

THE LAST BYTE

Sooner or later, the Hype Cycle will shift and put us squarely in the backlash phase, the Trough of Disillusionment.

I think we’re maybe ~2 months away from that becoming the dominant narrative.

In 2023, if you want to insult someone, just call them an AI.

As marketers and copywriters, there’s a fine line between a pretty amazing ad made with AI and a silly ad that doesn't move the needle at all.

You only get 1 or 2 of the “Wow, you made this with AI??” before it becomes so common, no one cares.

People are still somewhat amazed at what AI can produce. But it won’t last.

Plus, as with everything hyped, there are lies, half-truths, and overblown capabilities.

Here’s my advice, that has been the same for years:

• Focus on using tools that make sense for your workflow.

• Integrate AI and Machine Learning into what you’re doing at key points.

• Don’t chase shiny tools. There are thousands of them and almost all of them do the same thing.

• Become familiar enough with the tech so that you understand it and can make the best use of it.

• Practice the very human skills of instinct, discernment, judgment, perspective, and a sense of taste.

• We’re all tastemakers now.

• It’s never been easier to dominate your market with your talking points, your perspective, and your taste.

• Stop chasing silly prompts. Make AI and Machine Learning tools work for you.

And hit me up with any questions (including what you’d like me to cover in-depth).

See you next week,
Sam Woods
Editor