SEO for AI — What Does SEO Mean Now That We’re All Using AIs?

Trying to keep track of all the AI answers about my brand on different AI chat platforms got my head spinning. I had to sit down and think about how to actually start thinking about SEO for AI.

While the world is going crazy producing more content with AI in the hopes of improving their brands’ SEO scores, many have failed to realize that there will be no user facing “Search Engines” to Optimize.

Search Engines were great until we all learned that we don’t have to waste time visiting links in search results, but can get directly to the answer we are looking for by just posing the question to a Large Language Model. In effect

“Search engine UI’s have turned into AI backends.”

It will be our little AI friends sipping through the search results and figuring out what information to surface from the links to the UI.

What does that mean for my brand, product or company?

It means that anyone curious about your product, company or brand is likely going to be asking ChatGPT, Bing AI or Google Bard questions about it. Our new AI overlords will start having the highest influence on the perception of your brand and determine the sentiment on first contact with a potential new customer.

Wait, aren’t those AI answers based on generative models and full of hallucinations?

Correct! Read a background story on hallucinations about personal data on Hackernoon: AI and Personal Data: Does GPT-3 Know Anything About Me? or have a listen to an interview on American Public Media Marketplace radio on the topic - Marketplace Radio show: What does an AI chatbot know about you?

Now let’s look at the same problem from a product/brand/company point of view. Here’s the breakdown on answers coming from AI’s:

The first generation of popular LLM’s like GPT-3 came with a lower parameter count and thus have answers just barely getting things right. Only well known and established brands would hope to see outputs reflecting the truth, while the rest would see statistical garbage.

The second wave of LLM’s came with a larger parameter count (ChatGPT, Google Bard) and can encode enough to provide useful outputs for a larger set of brands, companies and products. This is what the majority of AI users are seeing right now through a service like ChatGPT.

And lastly we have LLM’s just interpreting search results to automate the step of searching through the links. Microsoft Bing AI and Perplexity.ai have taken that route for example and Google just announced Bard is switching as well.

The major difference between 1,2 and 3 is that the first two get answers from inside the models while the latter combines internet search results into an answer with the the help of an LLM. I thinks the last approach is the most likely to succeed given the obvious shortcomings of LLM’s having trouble encode links and references.

How NOT to do SEO for AI?

Most SEO salesmen seem to be focusing on suggesting you to use generative models to generate more content to boost your old-school SEO score. If you google “seo for ai”, you’ll find a list of sponsored adds for services such as jasper.ai, jemsu.com that all help you generate more content. That is likely going to be a dead end as everyone will be using AI to produce more content and you should expect the search engines to downrank volume soon.

Do AI answers actually get updated?

For older models such as GPT-3 there isn’t much you can do. There is no economic incentive for AI companies to update the parameters and I would bet that they won’t. It’s more of a test in brand recognition to see if your brand was important enough to be captured by the first evolution of LLMs.

For larger parameter count models such as GPT-3.5, GPT-4, Bard etc. we should expect them to be continuously updated at some interval that makes economic sense. Updating won’t happen often and most of the tuning will happen on filtering inputs and outputs of the models, but actual weight updates that influence the outputs to a larger extent are likely happening. Here your best chances are to make sure Wikipedia, Quora and any other heavily weighted information sources that are used for LLM training have the details and sentiment right about your brand. More on that in future articles.

Lastly for what I think is the future of search engines — the AI bots that interpret search results. Here you can also expect economic incentives to limit how many links the bots can cover to come up with a good answer. Even when content by search engines is cached and doesn’t actually require visiting sites — the token counts will set the limit to how much data is crunched through to generate the output. For economic reasons I think for the next year or so we’ll probably see no more than 5–10 first search results interpreted to capture the essence of your product.

For now any SEO work you do to make sure the content that captures your product best is surfaced at the top of the search results will not be wasted. But in a year’s time however all bets are off.

Ads inside AI answers

I expect all the big AI companies to execute on the new advertising opportunity in LLM answers and start offering paid services to influence model outputs or present ads intermixed with AI results. Google has started first iterations of mixing the results already. Google presenting LLM results in Search

This will likely also change what is the most sensible way to influence model outputs and the honest answer about this future is that no-one knows! If anyone tells you otherwise, they are probably not being honest.

Tracking AI answers about your brand

Those of us who want to prepare should start with tracking — make sure you know what all the models say about your products, brands and companies, their positives, negatives and the competition and keep track of those responses over time. This way you’ll know what the initial sentiment will be as search moves to AI and know if anything substantial regarding the AI answers changes.

The list of possible things you might want to track should include:

• is the brand well known by a model?
• what is the brand mainly known for?
• what are the most positive aspects associated with the brand?
• what are the most negative aspects associated with the brand?
• who are the competitors to the brand?
  etc.

How to actually do SEO for AI?

Since there are quite many models out there to keep prompting manually for changes regarding your brand and I did not find any existing solutions to track multiple models, then I had to write my own. I called it SEO for AI and decided to make it public as well as others might want to prep for the future of the changing SEO landscape. Give it a spin at

https://seofor.ai

This is just the first step and besides tracking, I don’t think anyone knows what the next step should be. People telling you otherwise are likely fooling themselves in this crazy race to redefine search. This space is evolving fast and something major has probably changed by the time I finish writing this article so I welcome you to come on this journey with me. Start tracking your brands in AI answers either manually or with the help of my tool and let’s figure out how to prepare for this future together!

Feel free to reach out on Twitter and let me know what you think in the comments below!