Tracking mentions isn’t enough: consistency and clarity help AI recognize a business and describe it accurately
If a potential customer asks ChatGPT or Gemini today, “What’s the best brand in this industry?”, will your business be mentioned? More importantly, will it be described accurately? For business owners, this question has a direct impact on conversions and sales.
We explore this topic in the latest episode of SEO Confidential with Dixon Jones, one of the most respected names in international SEO.
CEO of Inlinks and Waikay, Dixon has been working in digital marketing since 1999. He was one of the founders of Majestic, the search engine that maps the web’s links, an achievement that earned the company two Queen’s Awards for Enterprise and led to invitations to Buckingham Palace. We’re delighted to have him as a guest on our series.
In this interview, we break down the key concepts without the technical jargon. You’ll learn:
- why monitoring AI responses helps you assess the results, but isn’t enough to achieve them;
- how to tell whether AI models are actually talking about your business and your competitors;
- what to do when AI gets information about your company wrong;
- why consistency, your “About us” page and internal links matter more than you might think;
- how to showcase your expertise on a web increasingly filled with machine-generated content.
If you want to understand how to navigate AI search without overhauling your business, keep reading. There’s plenty to learn.
“AI prompt monitoring tells you where you stand, not how you got there”
What role does AI prompt tracking play today? In other words, how can monitoring AI responses help you determine whether a brand is being mentioned and recommended? What are its strengths and limitations as part of an online visibility strategy?
Honestly, I think AI prompt tracking is the wrong place to start, but the right place to look at the results. I used to say the same thing about SEO rank tracking, but prompt tracking is like receiving management reports a month AFTER you needed the data. It tells you where you’ve ended up, but it’s less useful for understanding how you got there.
There are other ways to start, such as checking the information AI provides to identify potential hallucinations and making sure it describes your business accurately, in the same way you describe it yourself. That’s work worth doing.
Waikay promises to provide a comprehensive view of a brand’s presence across the main AI systems by analysing hundreds of prompts, citations and comparisons with competitors. In practical terms, what does Brand AI Visibility Tracking actually measure, and what insights can it provide to a business that currently has little understanding of how AI models represent it?
Actually, I think that rather than analysing HUNDREDS of prompts, a better approach is to monitor the ones that matter. That’s the method we recommend.
Meaningful prompts are those that don’t introduce bias through the way the question itself is framed. For example, “What is the best CRM for small businesses?” is better than “Compare brand A, brand B and brand C”. The first question invites the language model to provide an open-ended answer, while the second introduces a context that excludes potentially relevant competitors.
Prompts should therefore reflect actual business objectives, rather than focusing on “prompt volume” and making too many assumptions.
These types of questions encourage responses that mention multiple brands. This allows us to measure share of voice in a tangible way, simply by counting how often your brand is mentioned out of the total number of mentions of ALL brands.
You can still segment these prompts, but introducing bias into the questions is a genuine user error.
Some people fear that Brand AI Visibility Tracking could become another version of traditional rank tracking. Lots of numbers, but not necessarily useful guidance on what to do next. How reliable and actionable are metrics such as Share of Voice, citations and positions in AI-generated results?
Beyond the fact that prompt selection can distort share of voice, another risk is taking competitors for granted or limiting the competitors included in the analysis. Booking.com might argue that it is NOT competing with EasyJet, for example, but if both brands appear in response to a user’s prompt such as “Fly-and-drive holidays in Portugal”, then they ARE competitors.
You therefore need a system that eliminates this distortion as well, automatically labelling the brands mentioned in responses.
There’s another way to make the data more useful, through a metric we call “Topical Presence”. To measure it, you need to look at the context in which competitors and your own brand are presented.
For example, an LLM response might say, “Booking.com is great if you want a wide range of options”, while the same response might describe EasyJet as “a budget-friendly option that offers car hire as an optional add-on”.
Both statements can appear in the same response, but the context in which each brand is recommended matters.
How important is it today to build and strengthen the entities associated with a brand so that Google and AI systems can identify it correctly? Which elements can help strengthen this association in practical terms?
There are many ways to answer this question, but I’ll give you a human answer.
The way you write about topics and ideas on your website defines your identity. When you reduce text to entities, for example in a dictionary, and assign a number to each entry, you turn your content into a digital fingerprint.
Some pages might discuss mustangs as wild horses, associating them with horsepower and nature. Others might discuss the Ford Mustang, the car, linking it to engine power and petrol. The words surrounding the term “Mustang” change its meaning and help determine what it refers to.
In these examples, the difference between “petrol” and “nature” changes a machine’s understanding COMPLETELY.
For a brand, the key is therefore to maintain a consistent brand narrative and stay on topic. Blog posts about the company Christmas party won’t be as useful as an article explaining a specific feature of your tools.
AI hallucinations can also involve very specific information about a brand, from its products and services to the people who work there. To what extent can a better-defined entity and a stronger online presence help reduce these errors, and which interventions are most effective in correcting them?
Consistency and clarity are fundamental, as I mentioned earlier. Your “About us” page can also be extremely useful to a machine trying to work out who you are and what your business does.
Telling the story of how the founder came up with an idea while sitting on a beach in Bali isn’t clarity. Stating the CEO’s name, the company’s headquarters and what the business does is. But you can go further by using open standards such as entitymap.org.
When it comes to hallucinations, though, we’ve found that tracing the original source is often very useful. Something may simply have changed in the real world without you updating your website, or a third-party review site may contain incorrect information.
In either case, IDENTIFY THESE PROBLEMS AND FIX THEM.
There’s little point in increasing your visibility in LLM responses if you haven’t first done everything possible to ensure that, when you are mentioned, you can actually deliver on the promises the LLM makes to users.
How is a brand represented in Google’s Knowledge Graph, and what impact can this representation have on how Google and AI systems understand its identity, activities and relationships? What are the most common mistakes to avoid?
I don’t want to make exaggerated claims about what I can say with certainty regarding Google’s Knowledge Graph. We manage our own knowledge graph and, in fact, generate it on the fly for any website, web page or response to a prompt that we want to analyse. We know that our “map of things in the world” is reasonably accurate: we don’t confuse a flower with a brick, and we assume that Google’s representation of a flower is similar to ours.
That’s how we communicate, both with machines and with people. We use vocabularies that overlap to a considerable extent.
That said, to answer your question, getting your OWN brand (or yourself) into Google’s Knowledge Graph is genuinely useful. If you search for a brand and see a Knowledge Panel with a detailed representation of the brand, including images, along with an AI Overview and perhaps a business listing, look for a “mid” or “kgmid” parameter in the search result URL.
With a little work, you can then investigate what Google’s Knowledge Graph says about you.
More recently, Google has also started developing “profiles”, which are interesting for people as well.
How useful is internal linking today in helping Google understand a website’s structure and the relationships between its content? And what mistakes can make it ineffective?
All roads should lead to Rome. The biggest mistake I see, apart from websites that link EVERY page to EVERY other page through a mega menu, is that people want every page to rank for something specific.
If your business sells ten types of products or has ten important features, what you REALLY want is for Google to recognise you as an authority in those areas. That’s why most of your internal links should point to those pages.
You can then write articles and informational content that cover several main topics within the same piece, providing machines with the context they need.
Link to your pillar pages contextually, within the body of the article, rather than simply placing links in the navigation or at the bottom of the page under a “cited source” label.
When analysing what AI systems “know” about a brand, which signals do you consider most important in determining whether they genuinely associate it with the topics and categories it wants to be known for?
Mathematically evaluate how closely the LLM’s response matches the content on your website. We can ask LLMs to be accurate. That is something we CAN measure.
Comparing a brand with its competitors in AI-generated results can reveal genuine visibility gaps. How can businesses turn these gaps into concrete recommendations for improving their content strategy?
We make comparisons in specific contexts, using prompts that a user might enter and asking LLMs, “What do you know about [Brand] in relation to [feature]?”
I find this second approach much more useful, because what the LLM says about your competitor but NOT about you reveals where the gaps are. It really is that simple.
Sometimes, that gap is NOT something you want to fill. Perhaps you DON’T want to be perceived as the cheapest option. In that case, though, you can be explicit about it. Add a FAQ question such as “Why are we a premium offering?”
Be specific, so that the LLM can help users make an informed choice, rather than simply steering them towards the choice you want them to make.
In your article, you argue that as AI-generated content continues to grow, human provenance could become increasingly important. How can we demonstrate to an AI system that a piece of content comes from a reliable human source and that the information it contains deserves to be preserved?
The central idea behind that article comes from the paper The Curse of Recursion, so I can’t take any credit for it. LLMs are starting to learn from second-hand AI-generated content and, given that they formulate their responses based on probabilities in many respects, they will tend to produce mediocre answers that represent the best possible “fit”. Mathematically, this will exclude the “minorities” (Minority Report), a film that, in my opinion, has taken on a new meaning in an AI-driven world.
Human beings who continue to think for themselves will retain their autonomy and sense of self-worth. Those who rely on prompt-generated answers will gradually, without realising it, begin to feel empty and meaningless.
It’s a serious problem, but it also represents a great opportunity. I’ve just written a book on this very subject, Power in the Modern World, which is available on Amazon.
In your article “The Power of B2B Influencers”, you explain that being mentioned by an authoritative expert can have a different value from simply getting brand exposure. How important is it today for a business to be associated with people recognised as authorities in their field to improve its visibility in AI systems?
If Beyoncé says that “Apple is a better operating system than Microsoft”, would you give her the same authority as Bill Gates making the same statement? Similarly, if Bill Gates talked about the health benefits of twerking, you would probably have some doubts.
We’ve reached a point where we treat everyone as an authority on every subject because, on TikTok or Facebook feeds, the context, background and expertise of the person posting aren’t always obvious. Over time, though, experts will speak knowledgeably about the subjects they know more often than people who simply pretend to be experts.
Choose your field and try to stay within it. That’s my advice. Try not to become an expert on subjects you had never formed an opinion about until they appeared in your feed.
The world is a big place. Find something you enjoy and are good at, then talk about it. For everything else, just keep asking questions.
Don’t pretend to be an expert: talk about what you know!
One thing stood out to me more than anything else: Dixon doesn’t sell shortcuts.
These days, everyone promises tricks to “get featured in ChatGPT”. Dixon, on the other hand, tells you to go back to basics.
It sounds like a small thing, but it means a lot.
Measuring how often AI mentions your brand is useful, but it’s like checking your bank balance. It tells you where you stand, not what went wrong. The real work comes first.
A clear website, an “About us” page with accurate information (who you are, where you’re based, what you do), content that consistently covers the same topics, and internal links pointing to the services you want to be known for.
I start with a twenty-minute test. Write down five questions your customers might ask ChatGPT, Gemini and Perplexity, then check which brands get mentioned. If only your competitors show up, you already have your to-do list.
Then there’s a question worth asking every new customer: “How did you find us?”
Add “An AI recommended you” as one of the possible answers.
For a business owner, that’s the number that really matters, more than any chart, because it tells you whether visibility is generating leads and sales.
Dixon puts a lot of emphasis on hallucinations, and I’d add one more thing. AI models don’t just read your website. They draw information from reviews, industry directories, articles and forums.
If people rarely talk about you outside your own website, or if they talk about you negatively, even a perfect website won’t be enough. Take care of those sources with the same attention you give your “About us” page.
Then there’s the final piece of advice, which is almost liberating for a business owner: choose your field and stick to it.
You don’t have to talk about everything. You have to speak knowledgeably about what you do best.
Well, what more can I say? Just a big thank-you to Dixon for a great conversation.
And thanks to you for reading all the way to the end. See you in the next episode of SEO Confidential. Don’t miss it!
