SEO Case Study: After testing, we found that IT IS POSSIBLE to influence the ChatGPT Search results for your brand. Here's how we did it: We've been doing some monitoring around how ChatGPT Search talks about Go Fish Digital, especially for certain queries that we know are important to the business. One of the queries that performs well for us is "digital PR agency". Historically, we've performed well in Google for this query and have recently seen incoming business referencing ChatGPT as well. This was a query worth monitoring for us. When looking at how ChatGPT Search listed our company compared to competitors, we found that there was a gap in how ChatGPT was producing results for us compared to competitors. For each competitor, ChatGPT Search would list a few "Notable Clients", generally larger companies they've done digital PR work for. However, they weren't listing this information for us. This was a missed opportunity to build trust signals through ChatGPT Search. Fortunately, we have an listicle article on our site that ChatGPT Search was frequently utilizing as a citation. We knew it would scrape content from this article, so we wanted to test if adjusting our content would influence the output. We went to the article and added "Notable Clients" to the article. We utilized a bullet list to better structure this as key value pair that an search engine/LLM would be more likely to view as a data source. After waiting a week, we saw that ChatGPT Search started finally pulling in "Notable Clients" when listing Go Fish Digital in the output! This was a huge win and something that now helps enrich our own listing within ChatGPT Search. So the takeaway here, is that when ChatGPT uses the Search function, right now you CAN have the ability to influence the output. The first thing you need is an article/piece of content that can feed into the citations. If you're able to get that, you can start adjusting the content to better control the output of how ChatGPT Search talks about your brand. We're continuing to run other tests as well. Perhaps we can get ChatGPT to list other things such as Awards or Years Active directly in the output.
Wil Reynolds
2w
OOOOH dude, you and I are seeing similar patterns, I saw our case studies on OTHER sites being referenced, I see the same thing you are with awards, I'm in the middle of building out a page to improve a query we aren't visibile for...I'll share all my learnings too.
Rand Fishkin
2w
NICE. It's rare to have the chance to influence the output so quickly and directly, but this is a great example I'm gonna share with others in the future (and reference y'all of course). 😎
Barry Hurd
2w
If you want to understand how to influence and LLM - I would suggest powering up a decent GPU rack (if you happen to have a boatload of GPUs just sitting around) and deploying one of the larger open source systems and running some models on clustering and token breaks. Once you get some metrics how the data is tokenized for a category in the foundational dataset you can draw some very rough parallels to the OpenAI model, but then you have to consider tokenizing the new search string as well and doing the math for what it takes to affect the probability of the ingested string vs the foundational dataset. Since OpenAI doesn't publish its foundational dataset the nearest comparison is to reverse engineer core pillars of data you can find (things like news sites, Wikipedia, common crawl, etc.) From an intelligence perspective: affecting a LLM is less like SEO and more like code-breaking.
David Mason
2w
Now for a more fun test. Remove the bullets and leave it as a sentence. Let's see if the bullets matter to GPT.
Nate Tower
2w
My main question around experiments like this is whether it will hold up over time and if this training will work on different users since queries on ChatGPT tend to be very different from user to user. In this case, it seems like your training was a success. I just asked ChatGPT for the top three Digital PR agencies and to provide some notable clients. It spit out two versions. One listed Go Fish Digital as number three. The other didn't list Go Fish Digital. It provided these as the notable clients: Notable Clients: Amazon GEICO 23andMe Marriott International Staples Pretty similar to what you provided above. Thank you for sharing this.
Course I could have shown you this a year ago with our search engine 😎 For the first step in making sure that your embedding chunks are showing up in the context window of ChatGPT use our free tool: https://brew.marketbrew.ai/ai-overviews-visualizer.htm
Almost like it's a search engine, wow.
Great case study. ChatGPT is exploring new business models, (i.e. transforming into a "for-profit business". They have recently raised billions toward this development. The most obvious approach is that they will pursue advertising. That's where the big question comes in. Can they provide effective ads (ie targeted ads) without mining user data? How will this impact the optimization of their current search function? My gut tells me their current mode of data analysis will evolve rather quickly and that what we know today will not be the same the day after their "for-profit" model kicks in.
There has also been discussions around influencing content and results via the use of schema - at least as it applies to local businesses (see Miriam Ellis observations on https://www.linkedin.com/posts/miriam-ellis-93855025a_should-you-use-schema-for-your-local-business-activity-7270495178442838016-y96l). There seems to be a maturity level needed for these models - some of this feels very familiar to those of us doing this long enough to remember keyword-stuffing techniques of yore. The question is always short-term gain vs long-term positioning. Where do you think this models need to go in order to gain market position?
VP of Marketing at Go Fish Digital. Speaker at MozCon, SMX, Digital Summit
1wYou can find the full article here along with dedicated screenshots for the different sections! https://gofishdigital.com/blog/seo-case-study-how-we-influenced-the-chatgpt-search-results/