Online reputation management is all about brand perception: You want people to perceive your brand in a positive, encouraging and compelling way.
While this was mostly focused on search + socials in the past, the rise of AI-based language models has started to add several levels of complexity to the matter.
Here are the key takeaways from this in-depth guide:
- LLMs present additional challenges due to the lack of tracking tools
- Monitoring any mentions of your brand on any platform is crucial
- Engaging with harmful content about your brand can shift language models’ perception
- Now is the best time to bake your brand into the AI ecosystem
Language models get trained by content on the web
Generative AI needs context to produce quality output.
ChatGPT, Perplexity, Copilot or Gemini, to name a few, use various content sources to train their language models.
Input = Output.
This means that whatever goes into the language model as training material, will impact what comes out of it.
Imagine this similar to training a new intern. The more examples of something they have seen, the better they are going to do their job.
Especially when those examples are consistent.
So, if e. g. Gemini is asked whether Nike is a legit brand, it will answer based on its training material about the Nike brand and related information.
The training material will most likely come from official sites like nike.com, but will also factor in elements such as user reviews, trustpilot reviews, videos showcasing products and customer experiences, blog posts, product descriptions, and even podcast recordings where the brand got mentioned.
The growing complexity
The numerous platforms where content can be hosted these days presents a completely new level of complexity to a brand, as training materials will vary greatly:
- Google is spending millions to use Reddit threads as training material these days (apart from obviously using the web).
- We can also expect Google to use all their platforms, from YouTube to Google Drive, to Gmail and more as potential training material to serve users better.
- Microsoft’s Copilot could technically have access to cloud-hosted office documents, outlook.com / hotmail, teams conversations and more.
- Perplexity.io even shows you specifically with references where the answer comes from (honestly, my preferred approach)
What exactly they can use without violating privacy restrictions is another topic I can’t comment on - but we also know that big brands have always happily paid fines while still getting their way to establish themselves in the market, so I do believe they’ll use whatever they can get access to.
Whatever content is out there, will be used as the training material for several language models - and might trigger them to respond about queries relating to your brand in a certain way. (Gert Mellak)
You can’t fix what you don’t know
In order to stay ahead of the game here, as a brand in 2024 you want to make sure you are aware of everything that could concern your brand:
- Where was your brand mentioned?
- Is there a different way to spell your brand name?
- Are there different ways in which people refer to your brand?
- Micro$oft for Microsoft
- “manzana” in Spanish when people refer to Apple
- Will Paulter is often referred to as “the eyebrows guy”
- People might also mention your employees:
- ”The food was good, but the waitress, Cathy, was extremely rude”
- People might only describe your brand …
- “I’ve got shoes from that brand with the stripes and I absolutely love it”
The different ways matter, because they represent varying training material the AI engines can and should connect to the same brand - especially when it’s positive material.
Action step:
- Monitor mentions of your band, products, and relevant staff members in all possible spellings and forms
- Use Google Alerts (free) or more professional tools like brand24 which I personally like (no affiliate).
- Create a set of sample queries people might type into the different platforms, and run them frequently to check their output.
- Leverage custom tutorials like this one from Wil Reynolds showcasing how tracking can start to work with a manual approach.
I found a negative review - what to do now?
In my book “Fix Your Online Reputation” I have described specific action steps around how to react to the different scenarios.
One of the most common ones people encounter is a negative review or opinion. Somebody writing a post, or posting in a forum or a review site like trustpilot, explaining their experience with your product or service.
How should you react?
Should you actually react at all?
Now, first of all, let’s be clear:
It doesn’t matter if they are a real customer or not - it’s going to affect you either way, as the “training material” is there, and ChatGPT won’t know if that’s a real customer.
Whether you should react or not will depend on a variety of factors, such as the overall page’s content and sentiment, frequency, whether the last reviews are positive or not, and more.
In many cases, reacting and offering a solution through customer support is a good first step. However, I prefer to decide this on a case-by-case basis, rather than providing an all-size-fits-all advice.
One factor where reacting to a thread might not be the smartest move, would be if the thread is really old, as adding fresh content to it might just re-ignite its importance again and even bring its relevance up.
Action step:
- If you found a negative review or mention, make an informed decision or consult with a trusted ORM expert before potentially increasing the damage.
Your brand’s content footprint
If you have a brand, you should proactively create content. That content will be controlled training material that the engines can pick up and learn more about your brand.
It is critical that the material they find consistently explains what your brand does, who you help, and how you achieve that.
Pay attention to the wording, and happily reuse phrases to refer to your brand that explain in simple terms what you are all about.
Similar to an artist’s color palette, you want to associate your brand’s entity with a terminology or entity palette. AI language models work on a semantic level, trying to connect relevant concepts with each other to understand how a piece of content works in context.
Pro-actively creating content that connects you with important concepts in your and your ideal clients’ environment will help provide consistent training material and future-proof your brand inside AI’s algorithms.
Next steps
As a brand, you need to establish your status quo, and based on that draw a game plan on what needs to be done consistently to strengthen your brand reputation online.
Hit me up at info@gertmellak.com for more infos!
