Avoiding context leak when comparing brands

What context leak looks like

When you write a question that names only one brand, you are not asking a neutral question. You are asking the AI to talk about that brand specifically, and AI models tend to treat the named brand as the centre of the answer. This shows up as context leak: the named brand gets mentioned more often than it would if the question had been asked in a generic way, simply because it was the anchor of the prompt. If you then measure how often each brand appears across your questions, this priming effect can make one brand look far more prominent than it really is in the AI's general knowledge.

This matters most when you are trying to compare your brand against competitors. If your question reads like "What do people think of [Your Brand]?" the AI has nowhere else to go but talk about your brand. If a competitor's question reads the same way, the same inflation happens to them. The comparison then reflects how the questions were phrased, not how the AI actually ranks the brands against each other.

A concrete example

Asking "What do people think of [Your Brand]'s toothpaste?" across a wave of responses can put your brand's Share of Voice close to 90%, simply because it was the only name in the prompt. Asking the generic version instead, "What are the best toothpaste brands?", might only surface your brand in a third of responses alongside several competitors. The first number describes how the AI talks when told to talk about you; only the second describes how the AI ranks you against alternatives on its own.

Three ways to avoid it

There are three practical ways to avoid this.

• Ask generically. Keep the question generic and let the AI decide which brands to mention on its own, for example "What are the best options for [category]?" rather than naming anyone.
• Name everyone, symmetrically. If you do want to ask about specific brands, name all of them in the same question so the priming effect is symmetric across the set, for example "How do [Brand A], [Brand B], and [Brand C] compare?"
• Ask both ways. Add both phrasings, generic and named, as separate questions inside the same campaign. The platform analyzes across questions by default, so you get the benefit of both angles without having to choose only one.

Related reading

These phrasing choices work alongside your campaign intent: even a neutrally-phrased question can still read as one-sided in the final report if the report-writing AI has no framing to work from. And phrasing is only half the story on the way in — an unrelated entity that was never cleaned up can produce the same kind of distortion on the way out, by inflating the denominator a Share of Voice number is built from. Worth checking both before you trust a result.

It's also worth remembering that a single response is just one draw from what the AI could have said — see repetitions and response variability for how to get a more stable read rather than judging context leak off one lucky or unlucky answer.