Generative AI can help with hard problems like SEO and Songwriting, but is not yet ready to replace experts in each
Like many of you I’ve been watching Generative AI like ChatGPT, Bard and Bing Chat with great interest. On the one hand, challenges with the SEO expertise of generative AI have been well documented. On the other hand, I think it’s amazing that these large language models can help SEOs with tasks like article creation, keyword research, title tag and meta description creation and templates for outreach. These once time consuming and menial tasks can be more efficient now and SEOs should have more time for the creativity and strategic thinking necessary to beat well optimized competitors and get to the top of the search results.
Because how I think of SEO is not necessarily who can create the most keyword-relevant, technically sound piece of content and get authoritative links to it, but rather who can see that piece of content in the first position and then still think, “I am going to make my piece so good that people will fight each other to link to it first and Google search engineers would be embarrassed not to rank it ahead of what’s currently there.”
This is a hard problem that AI isn’t ready for yet. Maybe at some point as the technology evolves we’ll be able to tell ChatGPT, Bing Chat or Google Bard to do something really hard like “write a Pulitzer prize winning poem” and it will understand what that means and do it, but it is far from that today:
This is the actual Pullitzer prize winner this year for context:
How are you going to create that with generative AI? You don’t have to teach poetry to know the difference between the two poems in terms of quality.
A similar hard problem is songwriting. Sure, a lot of hit songs are formulaic and generative AI should be able to train on a dataset of the top 40 songs in the last 50 years and come up with something but will it be good?
You be the judge of that. I asked chatGPT, Bing Chat and Google Bard to write a hit song, and only Google Bard attempted it. All three language models gave me lyrics, but only Google Bard gave me lyrics, chords and a suggested tempo.
Initially the prompt was simply “Write a hit song,” which it gave to me as though that were the easiest thing in the world:
After seeing what it gave me I decided to make the problem a little harder, saying instead “Write a hit song about search engine optimization.”
My songwriting teacher at the Old Town School of Folk Music in Chicago would probably have responded differently, asking wisely if such a task is even possible given that hit songs are usually about relatable topics like love or fun or honor and not the arcane process of optimizing content to appear high up in organic search results. Google had no such qualms, but instead simply provided what I asked it for. First the lyrics:
Then the chords and the tempo:
If you ask ChatGPT or Bing Chat to give you chords or tempo for the songs they write, in my experience they push back and send you to a professional:
But because Google gave me those three components I could use them to actually perform the song, which you can find here:
As a song, is it a hit? Probably not, but it’s catchy and easy to play given that it’s four simple chords throughout the entire song with no guitar solos or key changes.
I’m an amateur musician who just plays local gigs so maybe Bono or Lizzo could turn it into a hit, but that kind of proves the point.
Generative AI technology on its own isn’t yet good at solving hard problems like songwriting or good SEO without humans helping. Use the technology to make the article that you want ranked first better and do it in less time, but don’t expect to have it write articles in competitive niches and rank sustainably. For that it still needs the help of a good SEO. And likely will for a while.