Most companies discovering the topic of AI visibility ask themselves one question: "Do I appear?"

That's the wrong question.

The right question is: "In what context do I appear?"

Because a negative citation in an AI response isn't neutral. It doesn't behave like an absence. It behaves like a reverse recommendation — automatically distributed, at scale, to every prospect who asks the question.

What LLMs do that Google didn't

On Google, a result is a link. The user clicks, reads, forms an opinion. The engine takes no position.

LLMs take a position. When ChatGPT responds to "which B2B communications agency do you recommend in Belgium?", it doesn't list ten links. It chooses, prioritises, and formulates a judgement.

This judgement can take several forms depending on what models have absorbed about your brand:

Positive citation: your name appears as a direct recommendation, possibly with one or two favourable contextual elements.

Neutral citation: your name appears in a list, without particular qualification. You exist, but you don't stand out.

Unfavourable comparative citation: your name appears, but immediately followed by an alternative presented as better suited. "X is an option, but for your profile, Y or Z would be more relevant."

Citation with caveat: your name appears with a built-in qualifier. "X is reputed in this field, although some feedback mentions variable response times."

The last two cases are the most dangerous — and the most frequent we observe in French-speaking B2B markets.

Where negative sentiment comes from

LLMs don't build their opinion of your brand from your website. They build it from the totality of sources they've ingested: articles, forums, comparisons, reviews, LinkedIn discussions, Reddit threads, transcribed podcasts.

Sentiment forms from everything the web says about you independently of your own discourse. Your "About" page doesn't enter the equation. What enters the equation is what others say about you — and in what tone.

Three sources particularly feed negative sentiment:

Reviews and sector comparisons. A comparison article that systematically positions you as the second choice, even if two years old, continues to feed the models. LLMs don't always distinguish a dated review from a recent one.

Discussions on specialist forums. Reddit, trade forums, LinkedIn groups — these spaces where professionals exchange freely about their experiences. A well-indexed negative discussion can weigh as much as a dozen favourable articles.

Absence of counter-narrative. When the only available sources are negative or mixed, LLMs have nothing else to balance their synthesis. Editorial silence leaves the field open to unfavourable narratives.

The case that best illustrates the problem

In our June 2026 retesting of Belgian B2B agencies, we observed a particularly instructive case.

An agency appeared in AI responses — which, on the surface, seems positive. But reading the exact context of citations, a pattern emerged: the agency was systematically mentioned as a "possible option", immediately followed by two competitors presented as "particularly suited for industrial SMEs" or "recognised for their responsiveness".

In terms of raw visibility, this agency seemed present. In terms of commercial impact, each citation positioned it as third choice out of three. Sentiment degrades without warning — and without any alert being sent.

This is the sentiment problem in action.

How to detect it without paid tools

The manual method is imperfect but accessible. It requires rigour in reading, not budget.

Step 1 — Build the right prompts

Don't search your name directly. Search what your prospects search. On your sector, your service type, your geographic area. Three to five representative prompts are enough for a first overview.

Step 2 — Read the context, not just the presence

For each response where your name appears, note:

  • Are you cited first, in the middle, last?
  • Is there a qualifier associated with your name?
  • Are you compared to a competitor, and in which direction?
  • Is there a reservation, a caveat, a condition?

Step 3 — Identify the source of the problem

If sentiment is negative or mixed, look for where it comes from. Google your name + your sector terms. Look at what comes up: comparisons, reviews, discussions. What Google indexes well, LLMs have probably read.

Step 4 — Test across multiple engines

Specialised platforms measure four indicators: mention rate, citation rate with link, sentiment, and share of voice relative to competitors. Without tools, you can estimate these four dimensions manually by testing on ChatGPT, Perplexity and Gemini separately — results often vary from one engine to another.

What you can concretely do

Sentiment isn't fixed. It evolves with what the web says about you — and you have more leverage than it seems on what the web says.

Produce a documented counter-narrative

If negative sentiment comes from unfavourable comparisons or old reviews, the response isn't to contest them — it's to produce more recent, richer, better-sourced content that gives LLMs material to rebalance their synthesis. A quantified case study, a substantive article on your method, an interview in a sector publication — each of these elements becomes a potential source.

Activate the right conversation spaces

LLMs synthesise web content to form opinions. If negative reviews dominate their training data, your brand suffers in every response. The reverse is also true: positive and contextualised mentions in well-indexed spaces — professional forums, sector newsletters, LinkedIn discussions — progressively change what models associate with your brand.

Clarify positioning

Sometimes the mixed sentiment comes from blurry positioning. When LLMs don't precisely know what you do, for whom, and with what results, they place you in the generic category — and in the generic category, you compete with everyone. Precise positioning gives models criteria to recommend you specifically rather than generically.

What this changes in the diagnostic approach

When we conduct an AI visibility diagnostic, we don't just measure whether a company appears. We measure in what context it appears, with what associated sentiment, and how that sentiment compares to that of its direct competitors.

This is the dimension missing from most self-diagnostics. Searching your name in ChatGPT and finding it says nothing about the real impact of that presence.

Visibility without sentiment is like a press mention without reading the article.

This article follows our study on Belgian agencies and our June 2026 retesting. Sentiment is one of the six dimensions analysed in our complete diagnostic.

→ Test my AI visibility score for free → See a report example