Hallucinating content: AI’s impact on search, Google Ads and shopping

If you’re anything like us, you may have mixed feelings about AI. Our generation of marketers has been set up and then let down by several decades of Steve Jobs impersonators promising us the next big thing in tech: Google Glass, cryptocurrency, 3D movies, Google+, self-driving cars, Elon Musk’s Hyperloop project, NFTs… the list goes on. Google alone has hundreds of failed moonshots, as shown by the dedicated graveyard Killed By Google. And our SEO team is still waiting for voice search to become the ‘next big thing in SEO’, as everyone in the community insisted it would.

But as users find more and more inventive applications for generative AI, we have to admit: we’re excited about the impact these developments will have on search and Google Ads.

Crowd-sourced AI tools

The latest innovation from ChatGPT’s parent company OpenAI allows users to make and market their own specialised AI chatbots that fulfil almost any function you can dream of (as long as you pay them for access, of course). Grimoire builds a website for you from a sentence, Gif-PT makes animations frame by frame for you, SEO Optimizer does what it says on the tin… you get the idea.

What’s interesting about this development is that it’s being crowd-sourced. OpenAI aren’t making these bots, the community is. And in doing so they’re unlocking the potential of AI as an assistant to our everyday tasks. And it’s not just limited to OpenAI – communities are popping up and creating image generation libraries that go from cartoonish, to painterly, to photorealistic. You can use AI to mimic your voice, a famous politician’s voice, or even make music from your favourite artists.

From a marketing perspective, we tip our hats to OpenAI. They’ve made a complex topic accessible through a chatbot and then handed their marketing over to the masses and are crowdsourcing an endless supply of content that regularly goes viral. And now with custom bots, they’re turning that process up to 11. ‘Power users’ of the platform are incentivised to put their personal modified chatbot out there to gain popularity, and the public must join the paid tier to access it. For a company based on research and development, they’ve taken to marketing their product remarkably capably.

Across the AI world, other companies are making their own strides but with different marketing approaches. Google, with their 92% share of online searches and relative monopoly on online advertising, have their own chatbot Bard – which is free. It’s also less advanced than ChatGPT. But that doesn’t bother Google, because they don’t need AI to be a product; their users are the product. Google’s focus is on bringing AI to the places where they can gather valuable information from you.

When your search phrases become AI prompts…

An upcoming feature that’s rolling out in the USA first is a function to search Shopping results with a description of the product you’re looking for. Currently, you search with keywords like “colourful patterned puffer jacket” and Google matches that to the title of products that ecommerce sites send it via product feeds (as well as all the available metadata those products have). You, the user, then hope the row of advertised products matches what you were looking for. In practice, this often involves a lot of luck. But now, you’ll be able to generate 4 images with those keywords, pick the one that most closely resembles your vision, and Google will match that to products visually.

Immediately, we can tell this is going to be useful for a lot of apparel. Matching words (your search) to words (product schema data) is not the best way to find something you’ll subjectively like based on appearance. So, using AI to magically ‘hallucinate’ an ideal product has the potential to revolutionise online shopping. But there are going to be obvious limitations, and this development raises the following questions:

• Not every search is limited to aesthetics – the phrase Google uses to demo this feature is purely visual but people might want to add sizing info, cost constraints, or even non-obvious materials. Will this interfere with the image the AI generates? Should this ‘meta’ data be provided by the user after they’ve visualised their ‘dream jacket’ via AI?
• Subjectivity and perfectionism go hand in hand. Can a user tweak and amend their chosen image? What if ‘colourful’ was too vague, and the user wants to specify a colour of lining, or which colours a gradient contains? We know from other AI use cases that amends are possible – often very easy – but will Google allow them here?
• Which leads us to costs: all this generation and prompting costs money. Tiny amounts, admittedly, but it can add up. OpenAI states that you can now generate small images for roughly $0.01 (depending on the prompt and the size of the resulting image). Users won’t be paying those few cents, but advertisers will, along with a healthy margin depending all manner of factors – and of course, pure greed.

Google’s demo of the AI Shopping search function in action:

For marketers, this is likely to mean higher costs in Google Ads for Shopping campaigns. Google may absorb the cost initially but their track record on extracting profit from advertisers is exactly why they are under investigation by the DoJ in the US, and got hit with an antitrust decision in the EU for their monopoly on Shopping ads in their own search engine.

But for the public, it might be the dawn of a new age. It may seem like a bit of a gimmick initially, but even taking that first step in connecting what the user is thinking of to the world of online images is significant. And for Google, it marks a fundamental shift in the foundation of their business. They change their algorithm constantly, including updates of 200+ factors they take into account in order to rank your business in Search results.

On the other end of that equation is that it has always been a case of keywords matching to results. And now you may enter a longer string of keywords to produce an image that’s then matched via similar image search techniques to your results. But once again, we have thoughts:

• Let’s not get carried away. Your keyword prompt will definitely be used to match to results too, but it will be through the filter of the image search. If a site matches 100% on keywords but the images look nothing like your generated image, it won’t appear… hopefully!
• Similar image search has already been around for years. We’re used to connecting a prompt to an image to similar images, which is the basis of this change.  But is it any good? Try it for yourself by right-clicking any image you like – Chrome, Firefox, Safari, and Edge all support the use of Google’s Lens tool that does this. If you’re on an Android phone, your default camera app also likely has this function. It’s even been tied to Google Shopping results for some time now. We may see more transparency to this feature – not any kind of explanation of how Google does it, but named updates to the function.
• What does it mean for SEO? Like a lot of modern SEO problems, it means SEO will bleed into other disciplines even more. User Experience (UX) has been at the forefront of modern SEO recently, and now even product photography is likely to get the same treatment. Expect reams of articles to be written on how to light your shots, frame them, how many product images Google expects, and so on.

Time will tell how exactly these innovations change the landscape of search, advertising, and marketing more broadly – and AI is only going to develop further. Google, Bing and Meta are all currently working on innovations to keep an eye on, but their progress won’t quite match that of open-source projects and public communities, as the big corporations that control the online marketing landscape have to move slower by necessity. It could be that we see a tussle emerge between user-generated innovation and progress made by the big tech firms – with Big Tech having to play catch-up. Either way, it’s fascinating time to be working in SEO, PPC and online marketing. Bring on 2024.