Trust & Methodology
Short answer: yes, and we'd rather show you why than ask you to take our word for it. This page covers the four questions we hear most: is your data safe, is tracking accurate, will you know when something changes, and, since it's the one most often called a "black box," exactly how sentiment gets calculated.
Data security
Full detail lives in our Privacy Policy and Data Processing Agreement.
Citation accuracy
Staying informed
Sentiment methodology
Sentiment is the score most likely to get the "black box algorithm" objection, so it's the one we document most explicitly. Here is the full path from prompt to score.
Step 1
Sentiment is not inferred from a single snapshot. We run the same buyer-relevant prompts against ChatGPT, Google Gemini (with web search), and Claude on a recurring cadence, and record every response verbatim.
Step 2
Each response is scored for how your brand is characterized, favorable, neutral, or unfavorable, across dimensions like reliability, user experience, and support, using the model's own language as the evidence.
Step 3
A sentiment score without the underlying quote is not verifiable. Every score in Halogen Presence™ is shown next to the exact AI-generated language it came from, so you can read the evidence yourself, not just trust a number.
Step 4
If a sentiment issue traces back to something fixable, like inconsistent NAP data or an outdated listing, that goes into your action plan as a specific recommendation. The goal is fixing the underlying cause, not manually re-litigating what an AI model said.
Transparency means naming the parts we don't control, not just the parts we do.
ChatGPT, Gemini, and Claude update their underlying models and retrieval behavior on their own schedules, outside our control. A score can shift because your content changed, or because the model did. We track both and label which is more likely.
A handful of prompt runs is not a statistically stable signal. We show trend lines over time rather than leaning on any single data point, and we're explicit in the product about when a sample is too small to draw a confident conclusion from.
Sentiment analysis measures how an AI model currently describes your brand. If that description is inaccurate, sentiment scoring alone won't tell you that. Our accuracy flagging is a separate, distinct feature for that reason.
Q1
The prompt run cadence and the scoring rubric are the parts that matter for trust, and both are described above. We haven't open-sourced the underlying scoring model itself, the same way most analytics platforms don't publish their exact algorithms, but every individual score in the product is shown with its source quote, so you're never asked to trust a number without evidence.
Q2
We're not going to argue with a chatbot's tone, and neither should you. A sentiment score just reflects the AI's actual language, quoted directly. What matters is why it said what it said. If the sentiment is driven by something fixable, like inconsistent NAP data across directories, Halogen Presence™ surfaces that as a specific recommendation instead of leaving you to dispute a number. If the AI is stating something factually wrong about your brand, that's an accuracy issue, not a sentiment one, and AI Citation Analysis is built to catch and trace those separately.
Q3
No. Halogen Presence™ measures and reports on AI-generated answers; it does not submit content to AI providers to influence model training or responses. The recommendations we generate are actions your own team takes on your own site and off-site presence.
For the broader context on why AI visibility scoring exists at all, see what Answer Engine Optimization (AEO) is and how it differs from traditional SEO measurement.
Run a free AI visibility snapshot and see the source quotes behind your own sentiment score.
See How It Works