On this page
A founder reads an AI answer that names their brand third, right after the phrase “a solid option if you don’t mind paying more,” and spends the next hour worried about it. Meanwhile the same brand is absent, completely, from eleven other answers nobody on the team has read this month. The cautious mention got noticed because someone saw it. The eleven absences didn’t get noticed because there was nothing to see. That asymmetry, not which problem is objectively bigger, is usually what decides where a team’s attention goes, and it’s worth checking whether it’s pointing at the right thing.
The two failure modes aren’t the same shape
Being described with a caveat and not being mentioned at all look like points on the same scale, bad to worse, but they behave differently enough that treating them as one dial misreads both. A caveat means the assistant considered you, had something specific to say, and chose to hedge. An absence means the assistant either didn’t retrieve you as a candidate or didn’t have enough to say anything at all. One is a quality problem with your existing presence. The other is a presence problem that sentiment tracking can’t even see, because there’s no answer to score. Monitoring tone and monitoring presence are already split into separate tracked lines for this reason, but a lot of teams still read them as two views of one underlying health number instead of two different problems with two different fixes.
Absence is the common case, not the exception
Here’s the part that reframes the question: outright negative framing is rare. BrightEdge’s analysis of Google AI Overviews and ChatGPT across Apparel, Electronics and Education found negative brand sentiment in only about 2.3% of Google’s mentions and 1.6% of ChatGPT’s, and even the “44% more likely to criticize” headline is 44% of a very small base rate (BrightEdge). Compare that to how often a brand doesn’t show up at all. A separate study of 22,295 B2B answers across ChatGPT, Perplexity and Google AI Mode found zero brand mentions in 66.5% of ChatGPT answers, 58.5% of Perplexity answers and 65.2% of Google AI Mode answers for the tracked organizations in that set (Analyze AI). Negative framing shows up on roughly 1 or 2 answers out of 100. Silent absence shows up on roughly 60 out of 100. If you only track sentiment, you’re watching the rarer event and missing the one that’s actually eating most of your answer set.
Why the rare problem feels bigger than the common one
This isn’t just a measurement gap, it’s a reading gap, and there’s real research behind why. A 2022 study in Frontiers in Psychology on “omission neglect” found that people routinely fail to notice and discount what’s missing from a description, filling the gap with a generous assumption instead of a penalty, unless something explicitly flags the absence for them (Mantel, Montag-Smit, Kardes and Barchetti, Frontiers in Psychology). That study tested product descriptions and restaurant listings, not AI answers specifically, so the extension here is ours, not the authors’: a caveat you can point to and quote feels concrete and threatening in a way a blank space in an answer never will, because there’s nothing to read, no sentence to screenshot, no line to flag to your team. The visible problem gets the Slack message. The invisible one, which is larger by volume, keeps compounding with nobody watching it.
What each one actually costs, and where
The two failures also bite at different points in the funnel, which matters for deciding which to fix first. Absence costs you at discovery: you’re not in the set of brands a buyer even considers, and a worse-reviewed competitor can occupy that spot instead simply because you weren’t retrieved as a candidate. A cautious mention costs you later, at the comparison stage, after you’ve already cleared the harder bar of being named. It’s a real cost, but it’s a cost paid from a position absence never lets you reach. There’s also a reliability problem baked into sentiment specifically: on the overlapping prompts where BrightEdge found both Google and ChatGPT surfacing negative sentiment, the two engines flagged different brands 73% of the time (BrightEdge). A sentiment problem on one engine often isn’t a sentiment problem on the next one you check, which is one more reason it reads as noisier and more fixable than a flat, consistent absence.
How to tell which one you actually have
Pull your tracked prompts and split them into two counts before you look at tone at all: how many came back with zero mention of your brand, and of the ones that did mention you, how many carried a caveat or negative lean. Measuring share of voice is the mechanical version of the first count. If the zero-mention count is the bulk of your weak answers, you have a presence problem wearing a sentiment costume, and fixing tone on the handful of answers where you do appear won’t move the number that matters. If you show up reliably and the caveats are what’s dragging the read down, you have an actual sentiment problem worth the Step-3-style root-cause work the sentiment guide walks through. These two diagnoses call for different fixes, and a blended “visibility score” that averages presence and tone into one number is exactly the kind of thing that hides which one you have, the same way a healthy blended score can hide a dead-zone query.
Fix the common problem before the rare one
If both numbers are available and you have to pick where to spend the next sprint, spend it on absence first. It’s the larger share of your weak answers in most categories, it’s the one that silently compounds because nobody notices a blank space, and it’s the precondition for the other problem even existing, since an engine can’t hedge about a brand it never names. Sentiment work is real and worth doing once presence is solid, particularly on the specific comparison prompts where a caveat is actually costing a sale, but chasing tone while two-thirds of your answers have no mention of you at all is optimizing the smaller, louder number instead of the bigger, quieter one.
Track both, as two numbers
Run a free Visibility Check and look at the raw answer set before you look at any score: count the blanks first, then read the tone on what’s left. From there, Monitor tracking keeps presence and sentiment as two separate lines instead of one blended figure, and tracking who else gets named in your place will tell you whether your absences are going to a specific competitor consistently, which is the fastest way to confirm it’s a real gap and not noise.