Sentiment analysis

AI sentiment analysis: being named is not being recommended

An assistant can name your brand and, in the same breath, call you the expensive one with the hard setup. That answer counts as a mention on every tool that only counts mentions. We rate how you are described, keep the sentence that produced the rating, and let you argue with it.

Start 7-day free trialAll seven features

Card required, nothing charged for 7 days

Answer · Perplexity · “best project tracker for agencies”

For agency work the strongest options are Northstar, Planwise and Tandem. Northstar handles multi-client billing best, while Planwise has the clearest approval flow for client sign-off…

Sources cited

  1. northstar.com/agenciesnamed
  2. yourbrand.com/blog/agency-billingcited, not named
  3. planwise.io/approvalsnamed
  4. g2.com/categories/project-trackingthird party
Illustrative figures from a demo workspace, not a customer's data.

Three ratings, and the text behind each

Applied to the passage about your brand, not to the answer as a whole.

  1. 1

    Recommended

    You are put forward as an answer to the question, not merely listed among the options.

  2. 2

    Neutral

    You are listed. Named without a judgement attached, which is most mentions in most categories.

  3. 3

    Negative

    You are named with a caveat: price, complexity, a missing feature, a comparison you lose.

Auditable, or it is worthless

Every rating shows its working

A sentiment score with nothing behind it is a number you have to take on faith, and nobody should. Each rating stores the passage it was drawn from, the signals that produced it and a confidence value, so you can read the words yourself and overrule us. This is the same standard we hold every other number to on the methodology page.

The passage is kept.
Read the sentence that was rated, in the full answer it came from.
Confidence is shown.
Ambiguous language is marked as ambiguous rather than forced into a bucket.
Per mention, not per answer.
One answer can praise your category and doubt you. Those are two facts.
States · ChatGPT · 30 answers

  • Named

    You were recommended by name.

    14 / 30

  • Cited, not named

    Your page was read. Someone else got the credit.

    7 / 30

  • Retrieved, not cited

    Fetched, then not used.

    4 / 30

  • Not retrieved

    The assistant never saw you.

    5 / 30
Illustrative figures from a demo workspace, not a customer's data.

Negative sentiment usually has an address

Assistants describe your category from what the web says about it. That means criticism has a source, and a source can be worked on.

When an assistant repeatedly describes you as hard to implement, it did not form that view independently. It is compressing what it read: a comparison page written by a competitor, a forum thread from two years ago, a review site whose entry nobody has updated since your last redesign. The description follows the sources, so the fix is upstream of the assistant rather than inside it.

That is why sentiment sits next to the citations rather than in a chart of its own. Read the rating, then read what the answer cited, and the recurring criticism usually turns out to trace to two or three pages you can do something about. When it starts to shift, monitoring tells you rather than you noticing a quarter later.

  • Sentiment beside sources. The criticism and the pages it came from, in the same view.
  • Per engine. Engines characterise brands very differently. One hostile engine is a specific, fixable problem.
  • No single happiness score. We will not reduce six engines and a hundred passages to one number out of ten.

Common questions about AI sentiment analysis

What people ask before they act on a sentiment rating.

What is AI sentiment analysis?
Rating how an assistant describes your brand in the answers where it names you. Being mentioned is neutral information on its own; being mentioned as the expensive option, or the one with the difficult onboarding, is a different fact entirely and it is the one that costs you deals.
Do you rate the whole answer or just the part about my brand?
Just the part about your brand. An answer can be enthusiastic about your category and lukewarm about you specifically, and rating the whole passage would average those two into something meaningless. We rate the sentence or clause that describes you, and we keep it.
Can I see why something was rated negative?
Yes, and this is the part we would insist on. Every rating stores the signals behind it and the passage it came from, so you can read the actual words and disagree with us. A sentiment score you cannot audit is a number you should not act on.
How accurate is the rating?
Each rating carries a confidence value, and low-confidence ratings are marked rather than hidden. Language about brands is genuinely ambiguous — 'a solid budget option' is praise or damnation depending on who you sell to — so we show you the text and let you make that call rather than pretending the model settled it.
What do I actually do about negative sentiment?
Read the passages. Assistants describe your category from what the web says about it, so a recurring criticism usually traces back to a specific comparison page, a review thread or an old pricing page that still ranks. The fix is upstream of the assistant, which is why we show the citations beside the sentiment.
Is sentiment tracked on every engine?
Yes, on all six, on every plan. Engines differ sharply in how they characterise brands, so the per-engine breakdown is worth reading rather than skipping to the average.

The rest of what we measure

Six more, all on every plan.

See how assistants describe you

Add your brand and the questions your buyers ask, and read the first answers within minutes. Card required, nothing charged for 7 days.

Start 7-day free trialSee pricing