Sentiment

Insights

Sentiment

• Overview
• Aspects
• Actions

Insights

Sentiment

• the current split and how it is moving
• the themes driving that split
• recommended responses to what the aspects show

Overview#

Sentiment covers how AI answers describe your brand when they mention it, and what is being said that makes them describe it that way.

It matters more here than it would in search because AI answers are written rather than returned. In traditional search you write your own meta description and that is more or less what appears underneath your link. In an AI answer the description is composed by the model out of whatever sources it drew on, in language nobody at your company chose, and you will never see it unless you go looking.

Like Visibility, this screen works in weeks rather than runs. Sentiment moves slowly by nature, and run-level noise would comfortably drown the signal you are trying to read.

The Overview sub-page#

Current Sentiment Share gives you the positive, neutral and negative split across your tracked prompts, with a Change against the previous week, or No change where there was none.

Treat that split as a starting point rather than a conclusion. It is good at telling you there is something worth looking at and almost useless at telling you what. In particular, a large neutral share is the normal shape rather than a problem: most factual mentions of a brand are neutral, because most of the time an engine is describing what you do rather than passing judgement on it. What deserves attention is a negative share that is rising and concentrated on a single theme.

Current Sentiment Share

Across tracked prompts this week

61%

Neutral

• neutral(61%)
• positive(31%)
• negative(8%)

Sentiment trend charts the weekly sentiment score over the last 8 weeks. Eight weeks is a deliberate choice rather than an arbitrary one: sentiment lags reality because the sources behind it change slowly, so a shorter window mostly shows you noise dressed up as movement.

Top drivers this week points at the topics carrying whatever movement there has been, which saves scanning the full topic list to work it out. The topic table itself shows Topics, their Prompts and their Sentiment, with Aspects available for each one.

If a week has too little data behind it you will see No sentiment breakdown this week. That is a message about coverage rather than a score of zero, and the two look alarmingly similar if you read quickly.

Aspects: Where the work is#

An aspect is the specific theme carrying a judgement: pricing, support, quality, ease of use, onboarding, and so on.

This is the layer you can actually act on, and it is the reason the Overview by itself is not enough. Nobody can fix "negative sentiment"; it is not a thing with edges. "Negative on support response times" is a piece of work that can be assigned to someone and finished.

From a topic, View aspects for this topic opens its aspects. View all aspects and Open this aspect on the Aspects page take you through to the full Aspects sub-page, where the detail sits.

Read the answers themselves rather than stopping at the counts. A count points you at a passage and that is all it does. Only the passage tells you what is actually being said, how fairly, and how badly it lands on somebody deciding whether to buy from you.

Best practices#

• Stable or improving across weeks, rather than any particular ratio of positive to negative.
• Negative sentiment concentrated in one named aspect rather than spread thinly across several. Concentrated criticism is tractable; diffuse criticism usually points at something more fundamental than a messaging problem.
• Aspects that match reality. If AI answers are criticising something you fixed a year ago, what you have is a stale-source problem rather than a product problem, and those can often be corrected simply by asking.
• Enough prompts behind the week for the reading to carry any weight.

Next steps#

1. Read Top drivers this week to see which topics moved.
2. Open the aspects for those topics, since the Overview number by itself is not something anybody can act on.
3. Find the theme carrying the negative share and read the language actually being used.
4. Follow it back to the source. Sentiment always comes from somewhere specific: a review, a forum thread, an out-of-date comparison article. Citations is how you find which one.
5. Correct stale information at the source wherever that is possible. An outdated third-party comparison frequently does more damage than anything on your own site, and its owner will often update it if asked politely.
6. Publish something more current answering the same question, so a fresher source exists for an engine to reach for next time.
7. Check Actions for recommended responses tied to what the aspects turned up.
8. Come back after several weeks. Sentiment lags in both directions, so an improvement will not show up any faster than the problem did.

If your brand voice and the claims you want associated with you have not been set up yet, do that in Brand Hub first, so whatever you publish in response is at least consistent with itself.

Limitations#

Sentiment describes AI answers, not your customers. It reflects what has been written about you and indexed, which can sit a long way behind where your product actually is today.

It covers only the engines your plan tracks and the prompts you have configured. It also cannot tell you how many people saw any given answer, only that this is what an engine produces when it is asked.