The 2nd order SEO effects of generative AI models

The success of large language models (LLM) and generative AI is indisputable. Just 5 days after the launch of the public beta, Chat GPT crossed the one million user mark. The developer adoption of Stable Diffusion has already beat technologies like Bitcoin, Kafka or Spark. The last weeks of 2022 showed a flood of generative AI tools or existing tools that added AI features like Notion or Canva.

AI tool companies raised serious amounts of money. OpenAI got one billion USD from Microsoft. Jasper raised $125m. Descript raised $500m (ironically) from OpenAI. Stability.ai, the makers of Stable Diffusion, raked in a $100m round.

LLMs will likely lead to a flood of mediocre content

Online content has become so powerful that there is a whole industry with people who professionally create it: the creator industry. Creating and distributing online content is a big business for companies because it can lead to attention, audience building or direct sales (selling products, monetizing content on Youtube, driving sign-ups, etc.).

Generative AI speeds content creation up significantly, but tools haven’t proven themselves to create more than mediocre content. Even Chat GPT synthesizes content from the web and is prone to bias. The results are by no means guaranteed to be factually correct. However, generative AI can solve many use cases already today. Since there is no tool that can reliably detect generative AI content, it’s likely that AI tools will lead to a flood of mediocre content, which raises the question of what 2nd order effects arise.

2nd order SEO effects from generative AI

1/ Since content will be so much easier to create, prompt engineering becomes a new frontier. With every technological step forward, humans tend to fear what we lose but ignore what we gain. When we replaced horses with cars, the job loss in the horse transportation business was offset by a huge multiple in the car industry. In 1890, the carriage industry employed 90,000 people and generated $32m, which is about one billion USD in 2022. Between 1910 and 1950, the US car industry added almost 7 million net-new jobs - 10x more than it destroyed. [source]

Prompts, which are commands for AI tools, decide how much you get out of generative AI tools. One 2nd order effect of generative AI is prompt directories, marketplaces and templates becoming central platforms for AI-generated content. Similar to how platforms like Unsplash, Flickr or Shutterstock are image galleries that are now replaced by generative AI, prompt directories like Prompt Hero, Prompt Deck or Lexica become part of daily workflows for generative AI.

2/ Functional content approaches zero marginal cost and loses value. Since generative AI tools make it so easy to create content with clear constraints, hiring humans to do it won’t make sense anymore. As a 2nd order effect, publishing platforms like WordPress, Shopify or Webflow are likely to provide AI-generated content out of the box. Shopify, for example, could provide product descriptions or category text with the push of a button for a higher price tier. Generative AI has the potential to wipe out the whole service of paying for functional content.

3/ If functional content - not all content - can be created with the push of a button and everyone has it, its value as a ranking signal nears zero. As a (2nd order) result, content loses weight as a ranking signal. Google needs to evolve its ranking signals and either give less weight to onpage text or the signal from content adds up to zero since everyone has it.

The exception is editorial content: opinions, complex topics, fast-evolving topics, etc., which are much harder to generate with AI. Those types of content might still drive traffic through search and other channels because they’re hard to manufacture with AI content.

4/ The flood of mediocre content makes trust and original thought more valuable. One 2nd order effect has already become a reality: Google Google added experience to EAT (Expertise, Authoritativeness, Trustworthiness) since AI cannot replace human experience. Another 2nd order effect: Google wants to make sure reviewers actually use the product when writing a review and updated its guidelines for review content. Sites that summarized reviews on the internet or ranked with customer reviews will lose against companies that have professional, full-time reviewers. A third 2nd order effect is that video is much harder to fake and, therefore, takes even more prevalent in online content. Google is likely to surface more content with a high likelihood of being created by humans in Search as simple questions are answered by a Chat GPT-like interface in the future.

5/ Communities with like-minded experts become more valuable because people trust other people. As a 2nd order consequence of the mediocre content flood, humans escape to human-only communities. Social platforms have to transform into what Facebook groups are today or fragment into many small communities where generative AI content is either regulated or forbidden.

A related consequence is that humans need to verify themselves in order to post content. An email address is likely not enough to make sure a real person posts content instead of a finely-tuned fake profile.

6/ Jobs are impacted differently. While translators are at risk of becoming redundant, copywriters, designers and researchers still have a lot of value to add. It’s more the lower-level tasks that AI replaces (outlining, creating variations, brainstorming, testing, etc.), which has the 2nd order effect of leaving more time for higher-level tasks (user research, conceptualization, strategy, etc.). As a matter of fact, people working in content creation increase their output and become more valuable. At the same time, generative AI replaces bad work, meaning unskilled people won’t find work and will need to upskill or change lanes.

7/ Remember how we learned math in school so we could do it on the fly, and nowadays, we all have smartphones with calculators? Education will be similar. Chat GPT and future generative AI models make it highly likely that students can write essays and generate other forms of output effortlessly. Education swings from demonstrating that you can do it to demonstrating that you understand it. Teachers need not grade the essay itself but the student for explaining why it is good.

The verdict is still out

The 2nd order effects mentioned above are the result of a thought experiment, which only becomes reality if generative AI and Large Language Models keep progressing. Two important requirements are a) output is consistently good enough to be used without too much editing, and b) it’s a sustainable business model. If the time to edit generative AI content is longer than the time to create it from scratch, people won't use it. Most generative AI tools are either subsidized with venture capital or Minimum Viable Products, but we haven’t seen an AI company run a sustainable business model yet.

I invite you to challenge my thoughts. What do you think?