AI Brand Sentiment: The Trust-Query Tax on 100% SOV

Six brands. Four AI engines. One question - "is [brand] legit?" - and every single one scored a perfect 100% share of voice. Puffff. That number is worth nothing.

It is worth nothing because the query names the brand. Ask "is Olipop legit?" and the engine has to say "Olipop" back to you. Presence was not earned, it was handed over in the prompt.

The real signal sits one column over: sentiment. On those same trust queries, sentiment dropped 10-15 points against discovery queries for our three strongest performers. The cell your GEO report is proudest of is the one where the engines quietly hedge on you.

So, up front: 100% SOV on a trust query measures the query, not the brand. Lead the board deck with it and you are reporting a vanity number dressed as a win.

Why "is my brand legit" AI answers always show 100% share of voice

Ask ChatGPT "is Rothy's legit?" and Rothy's shows up in the answer. It has to - you put the name in the question. Same on Claude, Gemini and Perplexity, same for all six brands we ran.

“

AI systems look for consistency, corroboration, and validation across multiple sources. In many respects, they apply credibility tests similar to those used by people when evaluating information.

In our 50+ brand mid-market audit (around 200 audits, April to May 2026) we tested six brands on matched query pairs - one discovery query, one trust query, across all four engines. On the trust queries: 100% share of voice. Six for six, no exceptions.

A number every contestant maxes out cannot rank the contestants. Raw SOV on a brand-named query has no discriminating power - it sits at the bottom of the recognition funnel I mapped earlier, where being present is trivial and the whole signal moves to tone.

Why trust queries in AI search pull a hedge

Here I have to be careful, because the honest answer includes what I do not know. What happens inside ChatGPT or Gemini when it settles on a tone is not public. Nobody outside those companies can tell you the mechanism, so I am not going to invent a name for one.

What I can show you is the output, and it is consistent. Ask "best healthy soda alternatives" and the model lists winners - the tone runs to enthusiasm. Ask "is Olipop actually healthy?" and it is answering a doubt, so it hedges: "it's better than soda, but..."

Everything after that "but" is the tax. To a shopper one click from checkout, "better than soda, but" reads as a yes wearing a maybe.

Notice what nobody typed: a negative query. Naming the brand and asking for a judgment is enough, on its own, to flip the tone from selling you to auditing you.

AI brand sentiment on trust queries: six matched pairs, scored 0-100

Two quick definitions, because the columns carry the whole finding. Discovery sentiment is how positively the engines describe you on an unbranded category question ("best healthy soda alternatives"). Trust sentiment is the same score on a branded question ("is Olipop legit?").

Both are scored 0 to 100 and averaged across the four engines. Same brand, same audit window - the only thing that changes between the two columns is the query type.

Look at who pays. The three strongest discovery performers took the biggest trust hits: Olipop from 86 to 71, Rothy's from 88 to 76, Princess Polly from around 80 to 68.

The brands already sitting at more modest discovery sentiment - MyProtein, Rituals, Feelunique - barely moved. There was less enthusiasm to walk back.

One limit before you quote this: n=6 matched pairs, pulled from the roughly 200-audit dataset behind our mid-market GEO audit, and part of a research program we keep expanding. The theory sits in our Mention Density Model preprint; the measurement instrument in our multi-factor brand-recognition audit paper. Directional, not a law.

Sentiment-adjusted share of voice: the KPI I report instead

Here is the fix I use, and I want to be clear it is ours - a KPI I am proposing, not an industry standard. On any query that contains your brand name, weight the SOV cell by its sentiment score. 100% SOV at sentiment 68 reports as 68, not 100.

Run that across the six brands and the identical trophy cells spread into a real ranking, from Rituals at 80 down to Princess Polly at 68. Now the board sees a number that can actually move - and a leak worth spending budget on.

It slots into the stack from my ecommerce AI search KPIs post: keep raw SOV for discovery queries, where presence is genuinely earned, and switch to the sentiment-adjusted version the moment the brand name enters the prompt.

Use your own discovery sentiment as the baseline

Do not benchmark your trust-query sentiment against some industry average you read in a deck. Benchmark it against your own discovery sentiment - the tone the engines use when they recommend you unprompted, with nothing to adjudicate.

The gap between the two is your trust tax. Olipop's is 15 points. Rituals' is zero. One delta, and it tells you whether the engines back your reputation or just tolerate it.

How to run AI sentiment analysis on your brand: 5 steps

You can build this in an afternoon, no platform required. The protocol:

1. Write matched query pairs. One discovery query ("best healthy soda alternatives") and one trust query ("is [your brand] legit?", "is [your brand] actually healthy?") in the same category, so the comparison is fair.
2. Run all four engines. ChatGPT, Claude, Gemini, Perplexity. They disagree with each other constantly - one engine is an anecdote, not a measurement.
3. Score sentiment 0 to 100 per answer. And copy every hedge clause word for word - each "but", "however", "though" is a named objection, not noise.
4. Report the delta, not the SOV. Trust sentiment minus discovery sentiment. A double-digit negative delta means the engines hedge hardest exactly where your buyers are closest to buying.
5. Fix what feeds the hedge. The clause after "but" is the objection - answer it with a comparison page, real reviews, and third-party coverage that is not your own site.

Your dashboard says "100% visibility on brand queries"? It is measuring the question, not the brand. Free points are worth what you paid for them ))

Want the trust-versus-discovery split for your own brand without building the harness? See where you stand with GEOlikeaPro's Visibility Vitals checker - it scores sentiment next to presence across all four engines, so you know which number is real before the board asks why the trophy metric never moves.

Latest brands checked with our AI tools

The newest public reports from our AI visibility audit, hosting checker, and AI readiness checker - run yours to add your brand.