From the Archive: What Happens to Your Brand When People Start Searching in GenAI

While I'm on vacation, I'm running a couple of my old favorites

If you watched NBC in 1997, then you probably saw one of my favorite marketing campaigns of all time. The ad made the point that while NBC was airing reruns, “If you haven’t seen it, it’s new to you.”¹ Here’s my version of that. I’m on vacation this week and next, so I’ll be rerunning a couple of my favorite posts from when I was starting out and had fewer subscribers.

This one is about optimizing your brand for AI search (AIO).² There’s another one about optimizing for B2B search here. This topic is even more relevant now than it was 10 months ago, so worth a read if you missed it the first time.

Introduction

I don’t know about you, but a huge percentage of my searches have moved from traditional search engines to ChatGPT. I can ask questions that are very specific to the question I am trying to answer. Here are some recent searches:

• I’m having a family reunion in Ocean City, NJ. What’s a good place for dinner that is family friendly?

• The recipe calls for mirin. I don’t have any, and I don’t have any sake either. What do I substitute?

• I want to get a pair of new socks that I can wear with shorts[1]³

That last one is a big issue for retailers and brands! How do you ensure your brand will show up at the top of the list with an accurate description? The answer to that could be worth a lot.

What is AIO and why does it matter?

Companies trying to optimize their position among traditional search results do what’s called Search Engine Optimization (SEO). That means putting certain keywords on the page, improving load times, getting more pages to link to the site, and then testing that repeatedly to measure the improvement.

The process of doing this in chatbots is starting to be called AI Optimization (AIO). I’ll explain in the next section our first ideas for how to do AIO, but first I want to give a few examples of what I’m talking about.

I went to Chat GPT and said, “I want to get a new socks that I can wear with shorts” then I said, “Where can I buy crew socks?”

You can see right away that it recommends specific brands and retailers. The implications for those companies are obvious: making sure you get on the list is critically important when search starts in GenAI environments.

But, it gets weirder. I asked ChatGPT the same question with identical wording and got different lists of brands. Here’s a table summarizing the results.

I think this variance is good news for companies because it means that unlike SEO, there’s not a static ranking. There’s less risk that your brand will consistently be left off the list or be at the bottom. Still, you can see some consistency. Amazon is always first, and Nike and Adidas make the list every time. Some smaller brands like Happy Socks only show up once. There’s clearly an algorithm that could be influenced.

The Claude results interestingly are a bit more consistent and inclusive:

You can see more brands on here consistently. The order is different too – department stores are first rather than last, and Amazon is fourth. But, note that Gap, REI, and L.L. Bean only show up in the first one.

Note: Perplexity (an AI search engine) results had fewer brands and were 100% consistent from one run to another.

One final note: This applies equally to B2B as well. I won’t bore you with another set of charts, but you can try for yourself with a prompt like, “I’m the CIO at a sock company. What CRM should we buy?”

How to optimize your brand or product for GenAI systems?

At least today, AIO seems much, much harder than SEO for a couple of reasons. First, the results in search engines like Google (and Perplexity) are consistent from run-to-run, so it’s easier to measure progress. Second, search engines like Google tell you what you need to do to improve rankings: E-E-A-T (Experience-Expertise-Authoritativeness-Trustworthiness).[2]⁴ There is lot of technique in how you implement this properly, but it’s a lot easier than the LLM black boxes.

How do brands get themselves to the top of the list on the LLMs? I asked ChatGPT how to do this, and they offered the following advice:

“Align your brand or product with the GenAI system's objective and logic. How can you make your brand or product relevant and valuable to the GenAI system's objective and logic? How can you make your brand or product compatible and complementary to the GenAI system's data, algorithms, and feedback loops? How can you make your brand or product consistent and reliable to the GenAI system's results and rankings?”

Easier said than done. How do we know what the LLM’s “objective and logic” are? Also, how do we deal with the fact that they seem to be constantly changing. I ran this experiment a couple of months ago, and ChatGPT was a bit more willing to recommend specific products and where to get them. I don’t know whether the changes are the result of the model learning what people want or OpenAI tinkering with it.

Here are a few experiments that companies are trying:

• Adjust affiliate marketing placements and content – The affiliate marketing line can be blurry. People can write copy about a brand they genuinely like and then get paid on the affiliate link, or people can decide that they like a brand that pays a higher affiliate commission. I suspect LLMs struggle to parse those various conflicts of interest and just accept the content as accurate

• Change website text, images, video and information design – this is the same idea as SEO, and since ChatGPT does a direct web search, it’s possible the same tactics will work here. In addition to the standard SEO playbook, site designers are experimenting with technical aspects of the site to make content more AI-discoverable

• Spamming the chatbots – I am not advocating this! In the early days of Google, there were link farms that would setup artificial links to game Google’s algorithm. I suspect we’ll see the same here where companies will pay large numbers of people to search on a GenAI platform and then express a strong preference for the company’s product to train the model. That will probably work for a short time and then the LLM creators will figure out counter measures, and an arms race will begin. At some point, there may be an easier way to at least feed your company information into the models that is facilitated by the LM creators.

All these approaches make take a while to show results, so it is a good idea for brands and agencies to start testing now to learn what works. Also, since the different models give different results, there is the risk that optimizing your site for one might disadvantage it for another although that’s unlikely.

Since the models will change frequently, you will need to monitor your AIO performance constantly to figure out how your brand or product is doing. This will probably be a new line of business for many ad agencies before long.

Conclusion

AIO is going to be an increasingly important marketing problem in the next 12 months as more people use LLMs as search engines. As you saw, we don’t have the answers yet, and I couldn’t find any papers that studied this problem. I will do another post on this if I find more convincing answers for how to properly do AIO optimization.

If anyone has seen any tactics that work on AIO, please leave a comment.

1. If you are too young to know what a rerun is, here’s an explanation. Back before streaming, you just had to watch shows when they aired on specific channels. Shows would produce seasons of 23 episodes, but there are more weeks than that, so many times they would show old episodes from earlier in the season. If you hadn’t seen it the first time, this was your chance to catch up.

2. The original article referred to AEO, an alternate term that was in use at the time. At this point, I think it’s clear we’re calling it AIO

3. This article made me think I need to make a change: https://www.wsj.com/lifestyle/high-socks-or-low-socks-gen-z-debate-cef1ffe8

4. https://developers.google.com/search/docs/fundamentals/creating-helpful-content