Going Against the Grain - Issue #22
GenAI has changed the marketing world since last edition on August 2021
It feels like a decade ago in the digital marketing world thinking back to August 2021… it was coming to the end of summer and 2021 Dutch Grand Prix had Max Verstappen win his home race. It was 15 months before OpenAI would release it’s ChatGPT which grew to 1 million users in 5 days after launching in November 2022 and now over 200 million weekly users. It seems everyday there is a new tool selling itself to marketers but is nothing more than a glorified wrapper on ChatGPT…
AI is not new it had been used within companies like Google for years such as Machine Learning used to help searchers correct their spelling since 2001. It was Google researchers that invented and open sourced the neural network design in 2017 with “Attention is all you need” but it was OpenAI that commercialized the research and introduced the first GPT model in 2018. If you ask ChatGPT you get a much more defensive answer on the history of the GPT model.
In the past AI was mostly invisible to marketers but has been heavily used by Google powering their Dynamic Search Ads & Responsive search ads since 2018. While Google had created the framework that allowed ChatGPT to exist it was OpenAI that made it consumer friendly and Google seemed to be left in the dust even after it tried to catchup with BERT, LaMDA, PaLM and now Gemini, consumers are obsessed with ChatGPT. AI was considered something too technical or geeky for most consumers but ChatGPT seemed to break down all those barriers with a friendly web interface.
There were also marketers like Christoph Cemper who were quick to jump on the bandwagon and build custom templates to make creation of content via ChatGPT even easier and more scalable such as AIPRM. No longer did someone need to spend hours creating prompts or hiring overpriced AI prompt creators to write a custom prompt (prompts are used to guide ChatGPT on expected output) you could crowdsource the perfect ready to use prompt with the click of a button and re-use the most suitable ones again. What makes ChatGPT interesting for it’s users is that it uses human friendly output and doesn’t require technical background to use it.
Microsoft was one of the early beneficiaries of ChatGPT integration due to it’s extensive funding of OpenAI which seems to have caught Google off-guard. What broke the model was that ChatGPT was initially free which broke the model as most AI/ML technologies were limited to researchers or enterprise businesses. But now fairly advanced content creation was in the hands of anyone with access to the web and no restrictions were put in place on what could be created.
One aspect still causing unsettling conversations was the way that OpenAI and several other GenAI models sourced the data to build it’s models by scraping copyright content. There is also the ethics and copyright/ownership aspects on the content output from these tools?
Still being asked is who should be compensated for the source of ChatGPTs inspiration for the content creation and are they entitled to any royalties for it’s continued use? Some other GenAI tools have slowly started to licensed material from publishers to improve the accuracy of the data into the platform but more publishers are opting to block GenAI crawlers and to this day most big legacy publishers are blocking OpenAI crawlers and several still have pending legal action for copyright infringement. Not to be forgotten about Stable Diffusion has been sued by Getty Images for using 12 million images to build it’s image generation model which was exposed when their watermark appeared on GenAI outputs.
Google has not had an easy run trying to catchup to ChatGPT/Bing as it’s first launch Bard AI wiped $100 billion off it’s share price and the newly renamed Gemini platform has been hastily embedded in organic search results causing some disruption for digital marketers and website owners. But what has caused more issues has been the flood of misinformation such as telling consumers to put glue on pizza or eat one rock a day there is not enough transparency on what is showing in the AI overviews on Google or Bing. While many of the AI overviews containing misinformation have been linked back to The Onion articles or troll posts on Reddit but has raised safety concerns about no oversight on the output of GenAI tools.
Companies were quick to start to ban ChatGPT after incidents such as Samsung employees accidentally leaking sensitive data into ChatGPT. Microsoft was quick to launch an enterprise friendly Private ChatGPT via Azure OpenAI that allowed data feed into the GenAI tool not be used to further train the public model. But while many corporations have banned ChatGPT Glassdoor’s survey in Jan 2023 showed 62% of professionals have used ChatGPT or similar AI tools. So it seems the cat is out of the bag but still too many companies don’t have responsible AI usage in place or have created Azure instances for staff to safely use ChatGPT for tasks. Microsoft Copilot has become the de facto standard enterprise answer to corporate ChatGPT usage but there are limitations to what it can do without custom development of new GenAI models.
There has been little information shared for website owners from Google on how users are interacting with GenAI results and there is no easy way refresh the data that Google is choosing to show to searchers. Microsoft Bing did some early efforts to show website owners the search performance of web & chat but never broke out chat as it's owner filter. So it’s kinda useless as you don’t know what % is purely web and what % is just GenAI results.
There is a slow rollout of Google Ads into the AI overviews but it’s not clear what metrics will be made available to advertisers. So Google is making it very hard for marketers and website owners to understand how AI overviews helps or hinders their business going forward.
There are a handful of tools such as ZipTie that are tracking the AI overviews visibility across several verticals that show the average is now around 16.48% but if you break that out by vertical it’s all over the shop
ecommerce 2.4%
beauty 1.2%
food 6.99%
health 57.94%
finance 23.08%
publishers 21.6%
automotive 10.6%
jobs 20.4%
real estate 2.9%
What seems to be clear is that top of the funnel need type terms such as “How to Rank #1 in Google” are the ones that are seeing more AI overviews but they are also the ones that typically have a lower commercial intent and less Google Ads showing anyway. So for a lot of businesses that don’t capture much top of the funnel traffic the AI overviews won’t have an impact on their website traffic anytime soon.
It feels that tech companies are forcing GenAI into their tools without satisfactory safeguards in place and trying to move as fast as they can without concern for impacts on customers or society. In an interview with Bloomberg, Klarna CEO and founder Sebastian Siemiatkowski said that he hasn't hired any new humans in a year amid the company's push to embrace AI which is more an effort to increase profits that make life better for their employees.
It feels that adding AI labels to any Saas platform feature makes it cool now but a lot of what people think is GenAI is actually just some smart ML algorithms. There is also questions of ownership and usage of client data that needs to be considered but I do foresee ChatGPT type technology becoming utilised more but there are a few factors that keep bouncing around in my head.
Yes it is allowing people to create more content faster & cheaper but is it better?
Is it allowing people to create content that no human writer would ever bother to create in the first place?
Is it a race to the bottom that GenAI creates content that is then crawled and feed back into the model which creates new content based on it’s own output? See below for some inspiration on what happens when you feed it back.