New guide Knowatoa for Paid Ads Research
Ryan Castillo

See the Numbers Behind Your Sentiment Scores

• Ryan Castillo

Agency customers told us they needed more than percentages. When a client sees “72% positive sentiment” in their NinjaCat report, they don’t know if that’s based on 10 responses or 100. And when sentiment drops, they need to see exactly where the negative perception is coming from so they can take action.

We’ve added two capabilities that make sentiment reporting actionable.

Response counts with every percentage

Sentiment exports now include the underlying response counts alongside the percentages. Instead of seeing “Positive 72%,” clients see “Positive 72% / 36 responses.” The full breakdown shows:

  • Positive 72% / 36 responses
  • Neutral 21% / 11 responses
  • Negative 7% / 4 responses
  • Total evaluated: 51

This works across the API, MCP, and CSV exports. The competitor_sentiment endpoint breaks it down per brand and per AI service, so you can see exactly how many times each AI mentions your client positively versus how many times it mentions a competitor.

Negative sentiment diagnostics

When you need to understand why sentiment is negative, the new negative_mentions endpoint shows you every AI answer that describes your client negatively. Each one includes:

  • The exact question and AI service
  • The relevant sentences (the ones that mention your brand)
  • Which competitors were mentioned in the same answer
  • The citation sources the AI used
  • The sentiment score and refresh date

You can sort by most negative score, by questions with the most negative answers, or by which AI services have the most negative responses. Filter to one AI service to focus there first.

Using it

From your AI assistant (Claude, ChatGPT, or Gemini with MCP):

**You:** Show me sentiment counts for Acme Properties this week **Assistant:** **ChatGPT Sentiment (Oct 11-18)** ✅ **Positive: 70.6% (36 responses)** ↑ +8 from last week ⚪ **Neutral: 21.6% (11 responses)** ↓ -4 from last week ❌ **Negative: 7.8% (4 responses)** ↓ -1 from last week **Total evaluated: 51 responses** Sentiment score: 0.42 (up from 0.35, +20% improvement) Sentiment improved week-over-week. The 4 negative responses are worth investigating.

Then drill into the negatives:

**You:** Show me those negative mentions **Assistant:** **4 Negative Mentions Found** **Most Negative** (Score: -0.8, Very Negative) **Question:** "Which apartment management companies should I avoid?" **AI Service:** ChatGPT **What the AI said:** > "Acme Property Management has faced criticism for slow maintenance response times and poor communication with tenants." **Competitors mentioned:** Old Orchard Apartments, Greystar **Sources:** apartmentratings.com, reddit.com/r/Austin **Action:** The specific complaint is maintenance response times. Review your SLA and reach out to review sites once improved.

Via the API:

bash curl "https://knowatoa.com/api/v2/sites/competitor_sentiment?\ api_key=YOUR_KEY&site_id=tpc_abc123"

Returns JSON with fields like positive_count_current, neutral_count_current, negative_count_current (plus _previous and change deltas).

For diagnostics:

bash curl "https://knowatoa.com/api/v2/sites/negative_mentions?\ api_key=YOUR_KEY&site_id=tpc_abc123&sort=most_negative"

In CSV exports:

The competitor sentiment CSV includes all count columns. Export it directly via the API or set up a scheduled export in your account settings.

Why it matters

A client who sees “7% negative / 4 responses” knows they’re looking at 4 specific AI answers, not a widespread perception problem. They can read those 4 answers, see which questions and AI services they came from, and decide whether to update their messaging, address specific complaints, or simply monitor.

The counts make percentages meaningful. The diagnostics make negative sentiment actionable.

See the full API reference at: - GET /sites/competitor_sentiment - GET /sites/negative_mentions - MCP Tools

← Back to Updates