Today, Google’s search results still look (mostly) like lists of snippets and links, e.g.
But, plenty of search marketers and industry observers suspect we’re in the closing days of this format, and the future holds something closer to how AI-tools like ChatGPT and Bard answer user-inputted prompts, e.g.
If that’s the case, what can/should marketers do to give themselves a fighting chance at continuing to earn, if not traffic, at least potential references and nudges from Google (or whatever replaces it)?
This week’s 5-minute whiteboard takes a look at how those answers are calculated, and what it means for how marketers can potentially impact the results.
Howdy SparkToro fans and welcome to another edition of 5-minute whiteboard.
This week, we’re chatting about how to future proof your marketing for a potential world in which AI and large language models are the core of how search and discoverability works on the web.
So, let’s imagine that you think in 2026 Google’s interface functions much more like this rather than saying, best dishwasher and getting a list of results, you instead type in what’s the best dishwasher brand for fast drying. And Google gives you an answer that is similar to their generative search experience that they that they’ve demoed today, and that is essentially a paragraph or a couple of paragraphs, maybe a list of text rather than a bunch of links that you can go visit on the web and maybe some images and, information.
So Google might say something like the best brands for drying dishes fast are the (not sure how to pronounce it; I think it’s) Bosch.
Bosch 900 and the KitchenAid SuperDry. Fantastic. Great. How did Google come up with those? Well, what we know for certain is how large language models work.
They essentially predict tokens based on their input. So what has happened is this information comes from the token prediction system of a large language model. When you go to ChatGPT or you go to Bard or you go to Google’s generative, search experience, what you’re getting is a crawl of massive amounts of data. Similar to what Google crawls on the web today.
And then that crawl has been tokenized into words and phrases and combinations of words and phrases and numbers and all that kind of stuff. Then it’s trained by machines, right, in a in a classic machine learning style and also with lots of human input. You can see lots of stories about how, individual human beings all over the planet, especially in developing economies are very poorly paid to help train, classify, categorize, and improve large language models for use in services like OpenAI’s ChatGPT. These are then aggregated and the system is designed to predict the common responses.
Right? So it’s trying show you the kinds of things that you would get if a human being who had written on the web had written that content. So in this case, right, If the large language model, Google’s, you know, index of content of the web shows that when people on the web have written about and talked about fast drying dishwashers, and those sources often mention Bosch and KitchenAid, then those are the ones most likely to show up in these generated answers.
Via How to Get Better Outputs from Your Large Language Model on NVidia
This is the core of how AI answers work. It it’s not surprising to a lot of folks; plenty of people professionally in the search world and the AI world have talked about this before. What hasn’t been talked about too much, unfortunately, is what do we do if we’re a marketer?
Well, I think what’s great about this explanation is it actually makes the marketing very obvious. What you need to do as a marketer is make sure that if your brand is tied to words and phrases that people might enter into search experiences now or in the future, you are mentioned. Your brand is mentioned hopefully in positive ways in all of the places where a large language model might derive their generated answers from.
That’s the core of marketing in an AI centered world. Now what I wanna be clear on is we don’t know what that AI centric world will look like in the future. And I believe it’s honestly a little premature to say, “Oh, I’m gonna try and make sure that my brand is present in the places that I predict in the future, Google’s gonna use, you know, in the in the large language model training set.”
You don’t know for sure.
But here’s the thing. What you know for sure is that in the present day today, there are lots of sources where people go to get information about your brand:
What are the best brands? Where should I get software? Where should I get which video game should I buy? Which dishwasher should I buy? What brand of shirts are really good for, you know, men with tiny shoulders like mine?
The reported training set corpus for an early version of Google’s Bard LLM
And you can be present in those places right now. You can through content marketing and through PR and through outreach and sources-of-influence style marketing you can be present in those places. If you think Reddit is a likely source where lots of people are discussing your topic and Google’s almost certainly training on it and Reddit’s gonna sell them that data, fantastic.
How do you make sure that editors or people who want to talk about your brand in connection to these words and phrases that might be used the future? That’s something you can do today that will also for future proof your marketing.
Alright. Look forward to seeing you next week on another edition of five minute whiteboard. Take care!