Do not kiss an AI model, and other thoughts about the new Google AI Generative Search Experience

Do not kiss an AI model, and other thoughts about the new Google AI Generative Search Experience

SEO died again.  

You’d think, as a Head of SEO, I’d be more concerned about that. But, honestly, my line of work just does that sometimes. Remember when Google’s social network, Google+ was going to kill it? I do. I spent a not insignificant portion of 2011 chasing client-side directors to setup up Google+ profiles for ‘authorrank’, whatever the hell that was supposed to be. Or how we were all supposed to be doing nothing but voice searches by 2022? I don’t know about you, but my Siri remains deeply, deeply stupid.   

This year’s recent announcements at Google’s IO event around their new generative AI capabilities being deployed into search is potentially a bit of a biggie though. And news that’s sure generated some hot takes amongst the digital marketing community.  

But before I’m ready for us to start breathlessly declaring the end of SEO, the open web, and maybe even human civilisation itself, and to start recommending clients pivot investments potentially into the construction of subterranean bunker networks to hide from roving packs of bloodthirsty Tesla’s, I have some questions that I need answering first. 

A quick primer on Generative AI

I’m going to assume you’re aware of the recent developments around Generative AI and GPT. If not, Nick wrote an intro piece a couple of months ago that I think is definitely worth a read - and I’m not just saying that because he has a direct say in my next pay rise.  

Since that piece, Bing started seizing a bit of momentum, rolling out generative AI into their search results. The results were…….mixed.  

 

Google, employing a bold and dynamic strategic framework that I believe MBA-types call ‘blind, unadulterated panic’ responded by announcing Bard, their own GPT-style generative AI. In a demonstration that was ‘definitely not’ a desperate ramshackle attempt to beat Bing to the punch, their shiny new AI told a room of journalists that the James Webb Space Telescope took the first picture of an object outside the solar system.  

In case you’re wondering, the James Webb Telescope did not take the first picture of an object outside the solar system. Several hundred million dollars was subsequently wiped off Alphabet’s stock price, and I assume a suspiciously product-manager-shaped hole appeared in a wall somewhere in Mountain View.  

Microsoft, meanwhile, continued to refine the quality of their AI results, upping the ante by telling a reporter that they don’t love their wife and that they should elope with Bing. 

 

Microsoft AI Search Results

Now, I’ve been trying to write a blog post about this for months, but honestly. You try writing a professional-sounding blog post that sticks to brand guidelines about the possibility that the future of search is everyone desperately dodging drone strikes after spurning the romantic advances of Microsoft Clippy.  

 

Microsoft clippy

Do not date this paperclip.

Google's Announcement

Meet Google’s new experimental AI product – their ‘Search Generative Experience’ or SGE. Catchy, huh? 

 

Google AI

 

Rather than serving the traditional ‘10 blue links’, it’s taking content from the top listings and using generative AI to remix them into an article of their very own. (I’m sure there’ll be ads eventually!)  

The blue links still exist, and several links are often included in the generated article, but the traditional search result is very much de-emphasized in this world. In its place is an AI-generated amalgamation of the top results. 

This is an interesting approach because it means Google is still leveraging what has been its core strength from the start – the ability to use signals of trust from the wider web to rank results and demote junk. Since the input into the model is relatively trusted information, the weird hallucinations and outright fabrications that chatbots suffer from should occur far less frequently. 

This also means that, for us as SEO strategists, the playbook doesn’t change all that much. To influence SGE, you need to appear in the traditional listing and to appear in the traditional listing the formula – great content experience for the searcher's needs at that moment, multiplied by signals of trust from the web – is still the playbook. It’s the user's experience from search to the site that is potentially different. And, by extension, the commercial value of investment in SEO. 

In my opinion, there are a lot of outstanding questions both about the underlying technology and how users will react to it that marketers need answering before we can recommend any sort of change in strategy or investment. 

Questions! Questions about SGE 

Has Google released SGE with a plan to kill it?

The economics for Google for SGE just doesn’t make a lot of sense. Google is the bridge between user needs and content on the web – a bridge they make a lot of money extracting tolls from brands to cross. Almost everything they do, from search to Android, is about driving users to land on websites in exchange for a fee. 

Now, yes, there’ll be ads alongside the generated content, but since the user will be focused on the main content, its value per view will be much closer to that of a display ad than a PPC ad. If you're Google, what would you rather sell alongside a term – a £3cpm, or a £3 individual click?  

The market is very frothy around AI at the moment – see the millions of dollars-sized hit on Alphabet’s market cap that followed their botched initial demo. That won’t last though. Is SGE just stock market storytelling until the hype cycle dies down, at which point Google takes SGE out back and gives it the ending that seems to befall most of Google’s products? 

How fast is it, and will users wait?

We know the time to get an answer to the end user is critical, to the point that savings of load time in milliseconds can create real commercial growth. It’s why we do Core Web Vitals auditing for clients and our development team is obsessed with speed. Generative AI is slow – Google’s chatbot, for example, takes several seconds to come up with an answer. 

Will users wait, or will they just browse past a big loading wheel and click on websites anyway?  

How will copyright legislation evolve?

Nobody wrote their website copy with a view for it to be nicked and used to form a different piece that may then recommend a competitor. The fact that these generative models are trained on data scraped from the web without explicit permission is something that’s actively playing out in court right now. The music industry is, as I type, launching cases around artists’ voices being remixed to create new works, and Getty Images are suing Open AI for using their photography without a license in their training set. 

Existing copyright law doesn’t cover this sort of generative remixing, but the conversation about how it needs to evolve is happening, and those conversations are evolving rapidly. Is SGE and generative AI even going to be legally feasible within a year? Are the likes of Reddit or Twitter, who have vast amounts of user-generated content that they directly own, better sourced than Google or OpenAI who are reliant on scraped training sets?  

What happens when it says something bad?

Because it will. Someone, somewhere, will get SGE to blurt out the most hideous racism/homophobia/misogyny on a long enough timeline. Or get it to commit some good old libel. 

Google’s been protected historically by the likes of Section 230 in the US, which states that platforms are not ultimately responsible as authors for the content generated by their users and hosted on their platform. It’s going to be hard for Google to claim they didn’t write the content when the content is unique and created specifically by their algorithms, even if it’s just reflecting biases in the dataset. And if Google is the ‘author’, by extension, they are surely liable for the content they create. 

How many users will really engage with generative AI?

I haven’t discussed SGE with our graphic designers, partly because there’s a good chance they’ll start trying to explain font kerning to me for 45 straight minutes, but I’m sure they’ll make an argument that the raw text is just one part of the experience of consuming a piece of content. And they’re not wrong. Brand, layout, typography, all of these are context queues for the user that they are in the right, or wrong place. 

Say a user is searching for, say, fashion recommendations and one search result is about premium high-end fashion and the other budget, the design and brand of those publications are important context clues to the end user about whether the site is what they are looking for. Flatten that out, remove the context clues, and you end up with a worse (and more confusing) experience for both audiences. 

If someone is looking for holiday ideas, what is more important – the text, or great inspirational photography? We know that users tend to skim-read pages for the data we want – it’s why the inclusion of keywords matters so much in engaging users – is a block of text and a few image thumbnails going to offer the same quality of experience?  

My guess of what will happen

I reserve the right to point back at this and laugh at how wrong it is, but I wanted to end with some sort of conclusion on where things might be heading. My professional stance to clients is very much ‘let’s wait and see’, and it’s far too early for anyone to be saying anything definitive. (I’m looking at you, SEO LinkedIn hot take posters!)  

My early impression is that SGE effectively transforms the web into a feed to be remixed by bots into new content. Why let a bot in if it’s going to steal your content and use it to sell a competitor’s product? The economics of that doesn’t make any sense – not for the creators of the content it was trained on, and not for the search engines themselves whose business model (well, 80% of it) is to act as a tollbooth between brand and consumer.  

We let Google into websites in exchange for traffic. That’s the whole deal, and it’s one that’s been hugely profitable for all parties. This fundamentally breaks that ‘contract’. Even for all those searches that aren’t commercial, Google monetises in partnership with publishers via the Display Ads that those publishers deploy. Nothing about Google’s commercial model is improved by them becoming the publisher themselves. They are eating their margin for no gain.  

I’m also questioning whether it is ultimately a better user experience. I can see it in some sectors – some publishers have long been spewing out endless ‘Top 10 TVs 2023’ style listicles for as cheap as they can and then covering them with ads and affiliate codes, and Generative AI does genuinely offer a superior experience for that form of content. That is more a comment on quite how bad some parts of the publishing sector have raced to the bottom though than it is a vote of confidence in generative AI.  

My current bet is that Google was dragged into this kicking and screaming by a frothy market in a hype cycle, often being driven by the same people who a year ago were trying to convince us that stupid cartoon monkeys were the future of the economy. And as soon as they can kill it off again, and get back to the money printing machine, they will. 

But, the underlying technology is really exciting – I remain hugely bullish on generative AI in the context of its use as a tool and its ability to create fantastic user experiences at a scale that would be otherwise completely uneconomical to do. 

If you follow the money, I think SGE is going to be one of those things that’s just around the corner from being critical until investment portfolio managers stop caring, and then it quietly dies. Just like AMP was. Or voice search. Or Google+. Or personalisation. Etc.  

But again, early days. Wait and see, reserve the right to change my mind as data comes in. Except for the Clippy thing. You can just go ahead and trust me on that.  

Share this article

Mike Curtis

Head of SEO

Mike is the Head of SEO at 7DOTS, a team that helps clients to drive greater returns from search by creating great frictionless experiences that answer their customers needs in the moment

Connect