As search marketers, we’ve long understood the need to track branded and non-branded keywords separately. In the AI/LLM world, especially as we’re pushed to measure not just links and citations but brand visibility itself, the impact of brand mentions in prompts is even more important (yet, ironically, seems to have gotten less attention).

We conducted an experiment to investigate just how important brand mentions in prompts are in driving brand visibility in LLM output. There’s an even bigger question, though – is brand vs. non-brand even adequate for natural-language prompts?

Every search query or LLM prompt is either brand or non-brand, right? Logically, yes, but consider a prompt like “Who makes the most expensive luxury cars?”:

The result from Google Gemini is above — I’ve asked it to keep the answer short, but the bold highlights are all Gemini’s. What’s the first thing you spot? Brands. Even though no brands are mentioned in the prompt, the question obviously begs a branded answer.

This experiment looked at three sets of prompts (100 each) — brand, “soft-brand”, and non-brand — all of them based on the topic “seo tools” and a handful of pre-selected brands. We intentionally kept the scope narrow, and within a domain we understood well. Of course, results may vary across different topic areas.

Brand prompts

This is the most straightforward group. Brand prompts contained a brand name or branded product directly in the prompt. Some examples include:

• "Can I see historical Domain Authority data in the Moz dashboard?"
• "How many domains does the Moz link index currently track?"
• "Is Moz or Semrush better for a beginner in SEO?"

Note that brand prompts could include brands or branded products and metrics.

Soft-brand prompts

The “non-brand” prompts were split into two groups. The soft-brand group used our query fan-out research to generate prompts in an open-ended way. Examples include:

• “Are premium search suites worth the investment for a small blog?”
• “Can I use a tool to find the most popular questions in my niche?”
• “How do I reconcile keyword scores from multiple search platforms?”

There’s a bias inherent in our topic — questions about seo tools are naturally going to include specific tools and brands in the answers. So, even without including a brand-name or biasing the system toward brands, we’ve already created a soft brand bias.

Non-brand prompts

Given the topic bias, we nudged the system to generate prompts that were more tool-adjacent, resulting in broader, informational questions. For example:

• “How do you measure the organic search visibility of a new website?”
• “Is it better to target one high-volume term or ten low-volume?”
• “What is the best way to handle a sudden drop in rankings?”

We’ll call these our true non-brand prompts. Even from these few examples, it’s probably clear that the line between non-brand and “soft-brand” is a gray one and depends a lot on the topic. Brand mentions are an on/off switch, but brand bias is a volume knob.

How often did each type of prompt result in the presence of a brand in the output? This is an all-or-none measure — if any brand was present, then the output was counted. Note that all responses were collected with Gemini-3-Flash (via the Vertex API).

Every prompt (100%) that included a brand returned one or more brand mentions in the output. For non-brand, this dropped by almost half (53%), with soft-brand prompts somewhere in the middle. Again, our topic for this mini-experiment (seo tools) was brand-biased, so we’d expect relatively high numbers across the board.

While I don’t think these numbers are shocking, they do illustrate how easy it is to bias brand mentions in LLM output. What about the total number of brand mentions?

The results above didn’t surprise me, honestly, but the graph below definitely did. These are the numbers for total brand mentions across each type (100 prompts each):

On average, each brand prompt generated 14.5 brand mentions in the output, which dropped to 1.68 each for the soft-brand group and 0.79 each (<1) for the non-brand group. Seeding the brand in a prompt not only makes a branded response almost guaranteed, but it’s likely to drive multiple brand mentions (and, in many cases, competitor mentions).

As we try to improve our window into “AI” brand visibility, we have to consider not only how dramatically brand mentions can shift output, but the entire spectrum of brand bias. The way we phrase our questions or even the topics that drive those questions can dramatically impact how an LLM surfaces brands.

There will certainly be times when we want to know how LLMs regard our brands and our competitors’ brands, and there’s no such thing as a completely unbiased view of the world, but it’s critical that our prompt tracking and prompt generation reflect a variety of searcher intent, including brand intent. In the world of natural language, we also have to move beyond the brand/non-brand binary and think about how a prompt might naturally drive brand visibility even without mentioning specific brands.

There’s no one-size-fits-all answer here, and I can’t give you a magic number for brand mentions, but as always, the key is awareness and diversification. Just like you might separate branded and non-branded keywords in organic and paid search, it’s important to think about how brand, soft-brand, and non-brand prompts might bias your AI/LLM visibility.