HomeKoalr

Not just whether AI names you. How it describes you.

Koalr scores every mention of your brand that carries a clear signal, labels it positive, neutral or negative, and averages the scores into 0 to 100 for each AI engine separately. Mentions carrying no signal stay out of the average rather than being scored as a middling result, and when nothing carried a signal the sentiment score reads Unknown rather than a number. Underneath the score sit the named themes: the caveats and limitations the engines raise inside answers that are otherwise complimentary, each one traced back to the answers that raised it.

Your AI Visibility Score in under 5 minutes. No card required.

The AI answer engines Koalr monitors for your brand

Sentiment scored on each of the seven engines separately.

The bands

Three bands, and a fourth state that is not a band.

Every mention is graded once. The grade describes the answer's framing of you, not the answer's subject matter — an answer of plain facts about you reads as positive if it highlights a strength.

Positive

The answer recommends you, praises you, highlights a strength, or puts you forward as a good option. Informational writing counts as positive when it highlights your strengths.

Neutral

A factual listing or a balanced comparison, with no clear framing either way. You are in the answer and the engine has not taken a view.

Negative

The answer warns against you, raises a problem, or steers the reader towards an alternative as the better choice.

Unknown

Not a fourth band. Unknown is what the sentiment score reads when no mention carried a signal, and what a single mention counts as when there is nothing to read — most often a citation-only appearance, where an engine drew on one of your pages without ever writing your name. Those mentions are excluded from the average entirely rather than scored as a middling result, because a number invented from no evidence is worse than an honest gap.

The mechanism

It reads the sentence about you, not the answer.

Scoring a whole answer and calling the result your sentiment is the obvious approach and it is wrong: a glowing paragraph about someone else is still a glowing paragraph.

A window, with the other names taken out

Koalr cuts a window of about fifty words either side of where your brand appears, then removes every other brand named inside it before reading a word. That second step is not tidiness. Brand names are ordinary words — a venue called Dark Arts, a product called Battle, a firm whose name contains "fault" — and leaving them in the window makes their spelling count as your sentiment.

The reading is repeatable

The window is scored with two published methods — a graded sentiment lexicon that handles negation, intensifiers and "but" clauses, and aspect-based analysis, which scores the language around one named entity rather than treating a document as a single lump of tone. On top sits a small lexicon for the terms a general dictionary misses, like "overpriced" and "leading". The window is read by rules rather than by a model, so the same words score the same way twice. What a model decides is the shape of the mention — which kind it was — and that sets the anchor the words are weighed against.

Then the shape of the mention adjusts it

When the words carry a signal they are most of the score and the kind of mention is the rest; when they carry none, the kind of mention is the score. Then position: being named early in an answer nudges the score up a little where being named near the end nudges it down — because an engine putting you first has made a choice about you.

  • A strong endorsement. The engine names you as the answer, not as an option. Anchors highest.

  • A moderate recommendation. You are recommended with a qualifier, or alongside one or two others.

  • A plain listing. You appear in a list with no evaluative language. Anchors above the midpoint rather than at it — see below.

  • A comparison reference. You are named as the thing something else is being measured against. Anchors at the midpoint.

  • A critical mention. The answer raises a problem with you directly. The only anchor below the midpoint.

Worked example

One answer, from the sentence to the score.

Say you sell project management software and one of your tracked prompts is "best project tools for a small agency". An engine answers with five products. Yours is second, described as "a strong choice for small teams, though the reporting is thinner than some".

Fifty words either side of your product name. The other four products are removed from that window, so their names cannot lend or take away tone.

The themes

The objections live inside the good answers.

AI engines almost never write a wholly negative answer about a brand. Filter on the negative band and you find close to nothing, which reads as good news and is not.

What the pass actually does

Once per run, Koalr reads a sample of the answers that mention you — up to thirty, pooled across the latest run and the six before it, and cut down to one per prompt per engine so a single loud prompt cannot dominate — and looks for the limitations, drawbacks, caveats and exclusions the engines actually raised. Those are clustered into at most four named themes: a short title, a sentence or two on what the engines flag, and the list of answers behind it.

There is no fixed list of problem categories to be sorted into. The themes are named in the words of the answers they came from, so a constraint particular to your market gets its own name rather than being rounded to the nearest generic one.

Every theme has to show its evidence

A theme that cannot point at the specific answers that raised it is discarded rather than shown, so nothing arrives on the dashboard you cannot open and read for yourself. Each theme also carries which engines raised it, because an objection coming from one engine is a different job from one coming from all of them.

What is never treated as a limitation

Being the only one of your kind in an area is a strength, and is never reported as though your customers would have to go elsewhere. A deliberate specialism — one cuisine, one sector, one signature approach — is positioning, not a shortcoming, unless an answer explicitly frames it as falling short of what a buyer wanted. A plain fact with no drawback in it is not a theme, and neither is a downside no answer actually raised. All four rules exist because the easiest way to fill a list of objections is to invent them, and a list of invented objections is worse than an empty one.

No themes is a real answer

If the answers about you raise nothing a buyer would hesitate over, the list comes back empty and stays empty. It is not padded to four.

Reading it correctly

What the sentiment score does not mean.

It is not a measure of how often you are mentioned

Sentiment only describes the mentions you have. A brand named in one answer out of forty can outscore a brand named in thirty, because both are averages over the mentions that exist. Read it next to your AI Visibility Score — shown in-app as Prompt Visibility Rate — never instead of it.

One mention still returns a whole number

There is no minimum sample. A score built on a single answer is displayed exactly like one built on four hundred. Early in a brand’s first few runs, treat a high number as thin rather than as proof, and open the mentions behind it.

“Positive” at the bottom of its range means “listed”

Because a plain listing anchors above the midpoint, a brand that is only ever named in lists with no adjectives can still read as positive. That is deliberate — an engine choosing to name you at all is a mild signal in your favour — but it is a weaker result than the word suggests, and the mention type on each answer is what tells the two apart.

It is not a public opinion score

This is how AI engines described you in the answers to your tracked prompts, over the period you are looking at. It is not a survey, not a review average, and not what your customers think. It is what a buyer asking an engine would have been told.

The top of the scale is not where good answers land

The blend leaves headroom: a strongly endorsed mention usually lands in the 80s, and 100 is out of reach by design. Compare a score with your own last run rather than with the top of the range.

Which plans include it

Sentiment is on every plan, including Starter, and on all seven engines. What changes as you go up is how much you are watching: 25 tracked prompts per site on Starter, 40 per site on Growth and 50 per site on Business. More prompts means more answers to grade, so the score settles sooner. Full detail is on the pricing page.

Starter
25

Tracked prompts per site

Growth
40

Tracked prompts per site

Business
50

Tracked prompts per site

Where to go next

The rest of the product

Every capability, defined

What each part of Koalr is, what it measures, what you do with it and which plans include it — on one page.

← Back to the features hubComparing tools

How other tools handle sentiment

Sentiment analysis is one row in the comparison matrix, set against each rival in the comparison set one at a time, using that vendor's own published pricing and documentation.

See the comparisons →
FAQ

Sentiment, answered.

Still wondering? Run your free scan.

Koalr scores every mention that carries a clear signal, labels it positive, neutral or negative, and averages the scores into 0 to 100, per engine. The score is not read from the whole answer: Koalr cuts a window around the sentence your brand appears in, masks out every other brand named nearby so their words cannot move your reading, and scores what is left. Mentions with no signal are left out of the average rather than counted as a middling result, and when nothing carried a signal the sentiment score reads Unknown rather than a number.

Free AI visibility scan

Find out what AI says about you.

Scan your site free and get your AI Visibility Score, plus your first ranked fix list, in minutes. No card, no demo call.

One scan, 25 real prompts, 7 AI engines, under five minutes, zero setup.

Share of Voiceyou vs. rivals
0–100AI Visibility Score
Actionsyour next moves, ranked