Talk about visibility in AI answers and people almost always think in terms of competition: who gets the recommendation, us or them? That framing misses a third category, and in the case we analyzed it was the second-largest one of all.
In the more recent survey, 59 of 162 answers named not a single brand. That's 36.4 percent. In the survey two weeks earlier, with a different model, it was even higher at 47.1 percent.
That repetition is the actual point. Two different models, two different dates, measured independently of each other, and both show the same pattern. This isn't a quirk of one particular model, it's a structural feature of this kind of advisory situation.
Look at the brandless answers and they almost never show up on product comparisons. They happen on basic advisory questions. In the case we studied, for example:
The AI answers all of it competently. Just without a brand, because its sources don't name one either at that point. Advice portals, trade articles and forums explain pruning timing without recommending a manufacturer.
Because this space belongs to no one. On comparison questions, you're fighting established review sites and competitors with years of built-up authority. On advisory questions, you're fighting nobody.
There's a second, often underrated effect. These questions tend to sit near the start of a conversation. Show up there as a source, and you stay present as the conversation gets more concrete. The data backs this up directly: answers with the brand's own domain among the sources named the brand in 87.7 percent of cases, versus just 1.6 percent without it.
Customer service, sales and specialty retail know what gets asked. Collect 20 to 30 actual phrasings, verbatim. You don't need search volume here, because you're not optimizing for ranking, you're optimizing for citability.
Ask each question in a fresh chat and note the cited sources. You'll immediately see whether trade portals, forums or competitors show up, and how high the bar is.
No brand building, no company-history preamble. Someone asking about branch diameter needs the number. Models cite passages that conclusively answer a question, not passages that invite you to keep reading.
A subheading that carries the user's question word for word makes the paragraph underneath recognizable as a self-contained answer unit. That helps models extract individual passages.
Concrete values, standards, timeframes, your own measurements. Sentences like "highest quality" are worthless to a model, because they don't claim anything citable.
A sensible reference at the end of the answer is enough. Turn the text into a brochure and it loses the citability you wrote it for in the first place.
Many companies already have this content. It just sits in the wrong place: in a product PDF, in a dealer portal behind a login, in a video with no transcript, or in a blog post that only answers the question after eight paragraphs of company history.
Before producing new content, it's worth asking what you already have that's simply not accessible yet. That's usually the faster lever.
We can't say that from a single case. We measured 36.4 and 47.1 percent across two surveys of the same category. It's plausible that the share is higher where a lot of foundational knowledge gets asked about, and lower for pure product comparisons. That's not something we've verified.
That would be a fair objection in classic search. The logic here is different: you're not optimizing for clicks, you're optimizing for the AI treating your content as a source. If your domain shows up in an early answer, it stays more present as the conversation continues.
We don't have solid data of our own on that. It depends on how fast the page gets crawled and whether the model uses live web search or only its training data. Systems with live search can move fast; models relying purely on training data wait until the next training cycle.
You'd end up with content that gets cited without your name attached. A factual reference at the end is worth including for that reason. It's a narrow line: too promotional costs you citability, no mention at all costs you the effect you were going for.
Share of answers with zero brand mentions across all need states, personas and phases. Survey from July 26, 2026 (gpt-5.6-terra, n=162): 36.4 percent, 59 answers. Survey from July 12, 2026 (gpt-4o-mini, n=119): 47.1 percent. Both surveys conducted independently of each other.
A note on this article: research, analysis and writing were produced with the support of AI systems (the Ex Tenebris agent team) and reviewed editorially by Felix Zeh. Every figure cited comes from a real LUX/GEO analysis; the analyzed brand is anonymized to protect the client relationship.
The methodology behind this article is the same one we use for every assessment we run, tailored to your category, your competitors and your need states.