"Us" versus "AI in Search": Niche Down, Be an Expert, and Buckle Up!

bykppowell

133 followers

Updated 7 months ago

A few months back I set a Google Alert for "AI use in content creation." Today several interesting articles popped up. I thought I'd share a few of them.

First, from Wired, "Google's AI Overviews Will Always Be Broken. That's How AI Works." In this article, Wired discusses the debut of Google’s AI Overview function, which has (for the moment) been deemed a bit of a mess.

In the weeks following its release, Google’s AI Overviews - a feature that utilizes a large language model (LLM) akin to OpenAI’s ChatGPT - has been criticized for disseminating inaccurate information, the Wired article notes. The article goes on to explain that, despite the remarkable text fluency of LLMs, they can propagate falsehoods or mistakes, particularly when condensing online data. Liz Reid, the head of search at Google, conceded the need for enhancements after the system "pulled information from a satirical article" that "advised people to eat rocks" and also drew from a Reddit comment suggesting people put glue on pizza.

Google aims to improve the detection of illogical queries and decrease its dependence on content generated by users, writes Wired. However, managing LLMs is difficult because no one has yet solved the problem of enabling LLMs to distinguish between TRUE and FALSE and the internet - on which LLM algorithms are trained and from which they draw information - is replete with inaccurate and unreliable content ... or content that is not meant to be taken literally.

My take on this: We are at the beginning of this AI journey ... not the end ... not even the middle. As amazing as generative AI technology is, it still has significant limitations, and those limitations are worth keeping in mind ... both as users and creators.

Second, is a Zapier article, "Google AI Overviews: 4 early learnings from generative AI in Google Search," which addresses AI Overviews impact on search and SEO.

Experts believe, the Zapier article asserts, that while AI Overviews may impact superficial content it won’t degrade search results or end SEO. They predict a shift from broad, shallow content to more specific, lower-funnel topics tied to brands or products. AI’s limitations in replicating nuanced, experience-based or data-driven content will push content creators to focus on E-E-A-T (Experience, Expertise, Authoritativeness, and Trustworthiness). This shift will lead businesses to produce advanced, specific content requiring real expertise, which is challenging for AI and content farms to replicate accurately.

My take on this: Niche down and be an expert. This may not get the whole SEO job done, but regurgitating broad and general content is now doubly competitive, a space likely already claimed by others with greater authority and definitely ALSO now claimed by generative AI chatbots. But the plus side of this is if you can achieve expertise in a poorly served topic area you may not have to compete head-to-head with AI regardless of its integration into search.

I'll throw in a third article from Georgetown University, "OpenAI v. Scarlett Johansson? Georgetown Law Professor Answers Legal Questions on AI-Generated Content." I am not so much interested in Scarlett Johansson - though her case does make me think of the Black Mirror episode, "Joan is Awful." (The best episode of Season 6, in my opinion.)

I am interested in the bit about OpenAI's licensure deals with the Atlantic, Vox Media, News Corp and other media companies. These deals allow OpenAI - the creator of ChatGPT - "to use these companies’ content to train its AI models and answer user queries based on their content." As Kristelia García, the Georgetown University law professor interviewed for the article, says of OpenAI's licensure deals, "OpenAI gets to operate without fear of crippling copyright litigation from the companies it has struck licensing deals with. That could give it a very real competitive advantage over other companies that don’t have such licenses." Like Google.

My take on this: It's my understanding that, to date, LLMs like ChatGPT, Gemini, and CoPilot have been using the internet and its content to train models and answer user queries, much as WE might use pictures from Pixabay - FREELY! Limited only by their underlying architecture and model updates, they have all had access to more or less the same information.

The courts seem disposed to establish protection for content creators' rights vis-a-vis LLMs. It seems to me that as creators' rights to their content and its use in this context are codified and commercialized LLMs will not only have to compete based on the robustness of their underlying architecture, but also based on the information they are able to access for their knowledge base.

I cannot help but wonder how licensure of content will alter the generative AI landscape - perhaps, further exacerbating differences in generative AI models - and how this might trickle down to solitary users and creators like us.

Something to think about, anyway.