Brand Sentiment Monitoring
Monitor how AI platforms talk about your brand across five dimensions. Get alerts when sentiment shifts before it impacts reputation.
The five aspects
Every brand mention is scored independently on each aspect, giving you a multi-dimensional view of how AI platforms perceive your brand.
Quality
How AI platforms characterise the quality of your products, services, or expertise. High scores mean LLMs describe your offerings as excellent, reliable, or best-in-class.
"Nous Group is widely regarded as one of Australia's top public sector consultancies."
Price
How AI platforms frame your pricing relative to competitors. High scores mean LLMs describe your pricing as fair, competitive, or good value. Low scores may indicate mentions of premium pricing without corresponding value justification.
"Their fees are competitive for the quality of strategic advice you receive."
Trust
Whether AI platforms position your brand as trustworthy, credible, and reputable. This aspect captures mentions of awards, certifications, longevity, client testimonials, and governance.
"With over 20 years in the market, they have built a strong reputation for integrity."
Support
How AI platforms describe your customer service, client support, and engagement experience. High scores reflect mentions of responsiveness, accessibility, and client care.
"Clients praise their hands-on approach and dedicated project managers."
Ethics
Whether AI platforms associate your brand with ethical practices, sustainability, social responsibility, or transparent governance. This aspect is increasingly important as LLMs incorporate ESG signals.
"The firm has a strong commitment to reconciliation and First Nations engagement."
Score Scale
-1.0 (strongly negative) → 0.0 (neutral) → +1.0 (strongly positive)
Each aspect score includes a confidence value (0–1) and the specific evidence text.
How the pipeline works
Response Analysis
When our prompt monitoring pipeline receives an LLM response mentioning your brand, the full response text is stored with its context window — the surrounding paragraph that frames the mention. This ensures we have the complete context for accurate sentiment classification.
Aspect Classification
Each mention and its context window are sent to a Claude model with a structured prompt that scores each of the five aspects independently on a -1.0 to +1.0 scale. Each aspect also receives a confidence score (0–1) and the specific evidence text that informed the score.
Qualified Mention Detection
The classifier also flags whether the mention is "qualified" — meaning the LLM expressed a substantive opinion about your brand rather than merely naming it in a list. Qualified mentions carry more weight in theme analysis and alerting. Hedging phrases ("might", "could", "some say") are detected and stored for auditing.
Theme Clustering
Weekly, the platform analyses the last 30 days of classified responses to identify 3–8 recurring themes (e.g. "Competitive Positioning", "Client Support Experience"). Each response is assigned to its best-matching theme with a similarity score. Themes track their aggregate score and trend (improving, stable, or worsening).
Alert Evaluation
Daily, the platform checks all active alert rules against recent data. Rules can trigger when an aspect score drops below a threshold you set, or when it drops by more than a percentage over a lookback window (e.g. quality drops >15% over 14 days). Triggered alerts are stored and sent via configured notification channels.
Frequently Asked Questions
How is aspect sentiment different from overall sentiment?
Overall sentiment (positive, neutral, negative, mixed) gives you a single label. Aspect sentiment breaks this down into five dimensions. A response might be positive about quality but negative about price — the overall label would be "mixed", but the aspect scores reveal exactly where the sentiment diverges.
Can I create custom alert rules?
Yes. You can set thresholds on any individual aspect, optionally filtered by platform. Rules support three comparison types: "below" (absolute threshold), "above" (ceiling trigger), and "drops_by" (relative decline over a lookback period).
How are themes different from categories?
Categories (awareness, consideration, decision, support) are manually assigned to prompts when they are created. Themes are discovered automatically by clustering similar sentiment patterns across responses. A theme like "Competitive Positioning" might span prompts from both the consideration and decision categories.
What model is used for classification?
Aspect classification uses a fast, cost-efficient Claude model so every mention can be scored at scale. Theme clustering uses a more capable Claude model for its stronger reasoning about thematic groupings across large response sets.
Can I link a sentiment theme to a content brief?
Yes. Each content piece can be linked to its source sentiment theme. This creates a closed loop: detect a negative theme, generate a content brief to address it, track whether publishing that content improves the theme score over time.