Skip to main content
Available on: all plansFind it in: Analytics > Answer Engine Insights > Sentiment tab
Sentiment is a part of Answer Engine Insights that goes beyond tracking whether your brand appears in AI responses to tracking the tone and content of how AI platforms talk about your brand. The feature surfaces how AI platforms rate and describe your brand: an overall sentiment score, themes that AI engines consistently associate with your brand, and the attributes within each theme that carry positive or negative signals. It also scans the citations behind AI responses so you can see which sources are shaping the narrative.

​
What Sentiment measures

When an AI engine mentions a brand, it does more than cite it. It describes strengths, limitations, and use cases based on the sources it draws from. That framing shapes how potential customers perceive your brand before they ever visit your site. Sentiment helps you understand:
  • Whether the tone is positive or negative, and how that changes over time
  • Which specific themes AI engines surface about your brand, such as pricing, customer support, or product quality, and the attributes within each theme that carry positive or negative signals, such as “opaque pricing” (negative) or “responsive support” (positive)
  • Whether negative themes come from the content of a single outdated cited page or reflect a broader pattern, so you can address the right root cause and take action immediately
  • How your brand’s sentiment compares to competitors, broken down by AI platform and topic

​
How it works

​
Which prompts get analyzed

Profound gathers Sentiment data from prompts assigned the Sentiment type. The Sentiment type can be applied to any prompt in your account, including prompts you configure manually. When a prompt has this type, Profound analyzes the response for any brands mentioned and extracts sentiment from it.
See Prompt Designer to learn how to assign the Sentiment type to prompts.
This means Sentiment data can come from any prompt. For example, a prompt like “What are the best corporate credit cards for small businesses?” might return a response that mentions a competitor but notes that their foreign transaction fees are too high. Profound picks that up and counts it as a negative attribute under a pricing theme.

​
Prompt limits and timing for Sentiment prompts

Sentiment prompts count toward the same total prompt allowance as other tracked prompts in your account. New prompts and prompt configuration changes usually appear in reporting after the next daily processing cycle, and initial data may take 24-48 hours to accumulate.

​
Extracting themes from AI responses

Profound processes AI responses by extracting raw claims, semantically grouping them into clusters determined by the data, and surfacing the themes and attributes that are statistically significant. The overall sentiment score is the average of positive sentiment across each tracked AI model (ChatGPT, Gemini, and others). The claims extracted from AI responses are weighted by mention frequency: a positive claim mentioned 100 times has 10 times the influence on a theme’s sentiment compared to a negative one mentioned 10 times. Because themes are built from aggregated claims, a theme that appears more frequently across responses naturally carries more weight in the overall score.

​
Key features

​
Themes and theme attributes

A theme is a neutral, high-level topic area that AI engines associate with a brand, such as “Customer Support” or “Pricing.” Under each theme, attributes capture the specific claims and patterns that make up that theme, such as “long response times” or “opaque pricing,” each with a positive or negative polarity. This hierarchy is determined by the data: Profound extracts raw claims from AI responses, groups them semantically into clusters, and surfaces only statistically significant groupings. The result is a clean, actionable structure that identifies what is driving a positive or negative signal.

​
Exact text surfacing

When Profound detects a sentiment signal, it surfaces the exact text from the AI engine’s response that the sentiment originates from. This removes ambiguity: instead of inferring why a theme is positive or negative, you can read the precise language the AI used.

​
Citation-level attribution

For each sentiment signal, Profound identifies which specific citations are driving it, whether for an overall score or for a particular theme. The platform shows not just which source was cited, but where on that page the sentiment originates. This is the primary tool for distinguishing a content problem (one article driving a narrative) from a product or service problem (multiple independent sources reflecting the same issue).

​
Negative Sentiment Drivers

The Negative Sentiment Drivers table shows the most cited web pages containing your top negative claims. It’s useful as an actionable priority list of citation sources driving your negative sentiment. Drill down to investigate each claim, or use the Suggested Actions column to have an Agent built automatically to address negative sentiment in each citation source.

​
Competitive comparison

Because Sentiment prompts run across your full competitor set, you can compare your brand’s sentiment themes directly against competitors within the same view. This includes both the overall positive or negative split and the specific attributes AI engines surface for each brand.

​
Time series by theme and competitor

Visualize sentiment trends over time, grouped by theme or by competitor. This lets you spot when a particular theme started trending negatively, correlate shifts with external events or content changes, and track whether corrective actions are working.

​
Language and region support

Sentiment analysis supports prompts and themes across multiple languages and regions: it runs in the user’s native language. Theme names in the Themes table appear in English, as these are labels generated by the Profound platform, but this does not affect the sentiment analysis itself. Prompts grouped under a theme are displayed in the language in which they were run and analyzed.