Over the past weeks a couple of clients of mine came to me with requests regarding generative AI. They had done a GEO audit on their website. I had to look up what GEO is. It stands for Generative Engine Optimization. If that doesn’t make you hurl, you probably want to read my take on online marketers first. With a hefty dose of skepticism I went into those conversations. What the online marketers sold and suggested, made me angry. It was a waste of time and money, just like some SEO scans are.
A good designer is skilled in typography, colors, rhythm, breathing room for content and visually striking designs. A good copywriter understands the importance of focused pages, unique content, and good site structure. Lastly, a good web developer is aware of optimization, usability, accessibility, semantics and performance. This is why websites I work on with my dear partners, usually score green across the board on automated website performance and SEO tests, without having done any optimization. Good professionals know what they’re doing and they create websites that perform and rank. They also inform clients on the importance of a good content strategy going forward.
Do we need anything else for a website to perform well in organic search engines? No. We don’t. Do we need anything else for a website to perform well in “conversations” people have with a statistical word parrot? No. We don’t!
So what did the GEO audits deemed critical to change:
Explicitly add generative AI companies to
robots.txtwhen that file is already configured to allow everything.Create a markdown duplicate of the existing XML sitemap and call it
llms.txt.Add
JSON-ldfor every single different type of block on every page when most of those types are only permitted once per page.
It’s easy to summarize this entire audit as complete bullshit. We don’t need to allow individual companies to scrape our website, when we already allow all bots. We definitely don’t need a duplicate of our sitemap in a format that is currently a proposal and not even being used by any Large Language Model (LLM). And we really don’t want a duplication of schema objects that will invalidate our pages and break current rich snippets in search engines.
Why do we even have to optimize our sites for LLMs when they’re supposed to be really good at parsing text?
One of LLMs best strengths is supposed to be parsing text. They’re even better than the search engine bots that have been scraping and indexing our websites for decades now. Why do we even have to optimize our sites for LLM’s more than we already do for search engines?
For the purpose of this article I asked ChatGPT to scan the homepage of one of my clients’ sites to inform me where I could possibly add more JSON-ld snippets. Funnily enough it only took ChatGPT a few seconds to parse the homepage to figure out what types of content it contained. It seemed pretty good at this task. This leads me back to my initial question: why do we have to optimize our sites for LLMs when they’re already really good at parsing text? ChatGPT then gave me a list of bogus suggestions (similar to the ones found in the GEO audit) that would actually break the current JSON-ld implementation.
In my 20+ years of doing this I’ve learned that some online marketing agencies sell bullshit. They keep clients on their retainers by scaring them into thinking their website will stop performing when they don’t constantly run nonsense tests and bore webdevopers with nonsense requests. The products they make themselves are often of abysmal quality, as they usually click together a website with WordPress and Elementor. This week they made my clients lose time and money. They hurt our industry and make me look bad. And I’m sick of it.
If you have any doubt or questions about proper web optimization ask your designer, copywriter and web developer first. Just make sure they aren’t completely victimized of brainrot yet because of their over reliance on LLMs.
What the f*ck are we doing?