Happy Tuesday!
This week, I sat down with Dixon Jones, founder of Waikay (What AI Knows About You) and a 25+ year search industry veteran. Dixon started one of the first search agencies in the UK back in 1999 and has been deep in the weeds of knowledge graphs, entity SEO, and now LLM brand accuracy ever since.
We covered a lot of ground. Here's what stood out:
Watch the full episode here:
It Feels Like the Wild West Again
Dixon made a comparison that immediately clicked for me. The early 2000s search landscape, with AltaVista (Whew! Remember that name?), Yahoo, and the rest, felt like the Wild West. Everyone was making up the rules as they went.
That's exactly where we are again with LLMs. You've got ChatGPT, Claude, Gemini, Copilot, Perplexity, DeepSeek, Qwen. No single dominant player (yet). No settled playbook. It's chaos, and it's also opportunity.
Which is exactly why Jolly has been investing in earned linkbuilding and Reddit Brand Mentions for our clients. The brands building visibility now are the ones LLMs will know later.
Stop Tracking Prompts. Start Tracking Accuracy.
This was the biggest mind shift for me. Most GEO tools right now are tracking prompts. "Best hotel in Jakarta." "Top linkbuilding agency." The idea is to monitor whether your brand shows up when someone types a specific question.
Dixon's argument: that's not how most people actually use AI. They're not typing keyword-style prompts. They're asking their AI assistant to draft a reply, research a vendor, or recommend options. The actual prompt might be "write a reply to this email" and your brand either shows up in the response or it doesn't.
So the question isn't "are we ranking for the right prompts?" The question is: does the AI actually understand your brand correctly?
Waikay's Approach: Knowledge Graphs Meet Fact-Checking
Here's how Dixon's tool works, and I have to admit, it's a different angle than anything else I've seen.
Waikay builds a knowledge graph of your website. That's your source of truth, your definition of who you are, what you do, and what topics you're associated with. Then it asks ChatGPT (and other LLMs): "What do you know about this brand?"
It builds a second knowledge graph from that response and maps the two against each other. The result is a score out of 100 for how accurately each LLM represents your brand. Not just whether it mentions you, but whether it gets you right.
They even break it down by topic. So you might score 85 on "luxury travel" but 40 on "loyalty programs." Now you know exactly where to focus.
Tell AI What You Don't Do
This one hit home. Dixon shared that when they fact-checked their own sister brand, InLinks, they found ChatGPT was calling it a "rank checking company." InLinks doesn't do rank checking. That's Ahrefs and SEMrush territory.
So what did they do? On their About page, they explicitly state what they do not do. It worked! The LLM corrected itself over time.
Think about your own brand. Are there things AI is associating you with that you don't actually do? Every wrong association dilutes the ones that matter. You end up getting inquiries for services you don't offer while missing the ones you're built for.
This is also why the content you publish matters more than ever. It's not just about ranking anymore. It's about teaching AI who you are.
"Share of Model" Scoring Across LLMs
Dixon introduced a metric I hadn't heard before: Share of Model. You give Waikay a prompt like "best luxury hotel brand." It asks every major LLM. Every time your brand gets mentioned, that's one point. Every time a competitor gets mentioned, that's one point for them.
Then you get a league table across all models. It's like share of voice, but for AI. And it gives you a competitive benchmark that goes beyond any single model.
Your Non-English Content Can Now Show Up Everywhere
This one is relevant for anyone working internationally. Dixon explained that LLMs can learn from content in one language and serve it up in another. Your Italian blog post about luxury travel could surface in an English-language ChatGPT response.
That simply wasn't possible with traditional search engines.
Waikay now works in about 40 languages. The underlying concepts get tokenized into entities, so "the Eiffel Tower," "la Tour Eiffel," and "that big metal thing in Paris" all map to the same thing.
For brands with multilingual sites, this is a huge unlock. But there's a catch: make sure your messaging is consistent across languages. If your Chinese site says "luxury hotel" and your English site emphasizes "great destination," you're confusing the models.
When the Algorithm Changes, and It Will
Dixon closed with something I didn't expect but really appreciated. He talked about the emotional toll of algorithm changes. If you've optimized for a specific set of conditions and those conditions shift, by definition you're going to lose ground. It doesn't mean you did anything wrong.
His advice: prepare your clients (and yourself) for it. As Dixon put it: "You can dress perfectly for today's weather, but that doesn't mean you'll be dressed right tomorrow. The weather changes."
It's great advice for anyone in SEO & AEO / GEO.
Dixon's core message maps perfectly to what we've been building at Jolly. Our branded backlinks get your brand mentioned on authoritative sites. That's exactly the kind of signal LLMs use when they ground their responses.
Until next week, Greg
P.S. If you want to make sure AI gets your brand right, the foundation is earned media. Hit reply if you want to talk about it, or get started here.