How to Use Brand Sentiment to Find Out What AI Actually Says About You

Getting mentioned in an AI answer is not automatically good news. The assistant is not just deciding whether to name you — it is writing a sentence about you, and that sentence is what the buyer reads.

"A solid option, though users report the setup takes some time" and "the standard choice for teams this size" are both mentions. Only one of them wins the deal.

Brand Sentiment is the only screen in ApexaRank.ai that shows you the actual words.

Why the Wording Matters More Than the Mention

In traditional search, a link was neutral. The user clicked and formed their own impression from your page. In AI search, the impression is formed before anyone reaches you — and it is formed by a model summarising other people's descriptions of you.

Before You Run It

Two conditions gate this screen. It sits on a paid plan, and it needs a completed visibility analysis for the site — sentiment is measured from the same answers the report is built on.

If you see "this site has no completed analysis run yet", run the analysis from the Results page first. If you see "not enough credits", the run is priced like any other analysis and the balance is shown before you commit.

Step 1 — Sentiment Distribution

The page opens with Overall Sentiment and a Sentiment Distribution bar split three ways — Positive, Neutral and Negative, each as a percentage.

Read it in this order: negative first, then positive, then ignore neutral.

Step 2 — Brand Attributes

Next comes Brand Attributes (How AI Describes You) — a list of qualities, each scored out of 100 with a bar. Reliability, pricing, ease of use, support, and whatever else the models associate with your category.

This is your positioning, measured from the outside. And the gap between what you claim and what scores highly is usually where the work is.

Step 3 — Sentiment by Platform

Sentiment by Platform gives each assistant its own card: a score, a one-word tone, and the change since the previous run.

The spread here is diagnostic. Each platform reads a different mix of sources, so a difference between them is a difference in what they are reading, not in what you are:

Almost always one bad source that platform trusts — an old review page, a critical comparison article, a forum thread. Read that platform's quoted concerns and you can usually identify it.

No strong sources exist either way. Your own site is doing the describing, and it is not saying much. This is the easiest situation to improve.

Something recent has entered the corpus and is being read widely. Check what has been published about you in the last few months before changing anything on your own site.

Step 4 — The Phrases Themselves

Each platform card splits into two columns: Positive Language and Concerns / Negative, each holding the actual sentences the model produced. Where a run found none, it says so plainly rather than padding.

This is the most useful content on the screen, and the part most people scroll past.

Every line in the Concerns column is an objection being handed to a buyer before they reach your site. And unlike a survey, it comes phrased in the exact language the model will use again next time.

The Positive column is worth as much. Those are the phrases already earning you citations — reuse them in your own copy rather than inventing new marketing language beside them.

Turning a Concern Into a Page

The workflow is short and it is the same every time:

Do not paraphrase it into something more comfortable. "Setup is complex for small teams" is your heading. Rewriting it as "Onboarding excellence" is how the page fails to answer it.

If setup does take two weeks, say two weeks and say what happens in them. A specific honest answer gets quoted. A denial gets ignored, and the original concern keeps circulating.

A dedicated page or a clearly headed section, in plain text. Not a PDF, not a video, not a paragraph buried in a case study.

If the concern traces to one outdated article or review, the durable fix is getting that source corrected or outweighed — publishing alone does not remove what the models are still reading.

Tracking It Over Time

Each platform card carries its change since the previous run, and the page timestamps when sentiment was last analysed.

Sentiment is the slowest-moving measurement in the product, and that is a feature. It changes when the sources models read change — which means a shift here is real, where a two-point move in a visibility score often is not.

Compare runs rather than days, and expect a published answer to take weeks rather than hours to show up in the language.

Common Mistakes

The split is context. The quoted sentences are the instructions.

Neutral is the default state for most brands in most answers. Chasing it into positive with promotional language usually makes content less quotable, not more.

A page that denies a widely repeated criticism reads as defensive to a model too. Acknowledge, qualify, and be specific.

Re-running sentiment a day after publishing tells you nothing. The corpus has not been re-read yet.

Frequently Asked Questions