Watchdog

Your Sill dashboard tracks how visible your brand is. Watchdog goes deeper, monitoring what AI engines are saying about you and alerting you when something is wrong or has changed.

Sill Watchdog is a risk monitoring layer that analyzes AI responses for caveats, factual claims, competitive threats, and misinformation about your brand.

How Watchdog Works

Watchdog runs automatically after each monitoring run. It reads the same AI responses your dashboard uses and applies risk-focused analysis in three stages:

1. Extract: Identifies caveats and factual claims about your brand from every AI response that mentions you.
2. Analyze: Groups similar findings together, compares them across AI platforms, and detects what's new or changing.
3. Alert: Generates alerts when it finds something that needs your attention.

No extra setup is required. Once you've completed the monitoring wizard, Watchdog begins working immediately.

---

Fact Profile

Sill automatically researches your brand to build a fact profile used to verify claims AI engines make about you.

When you complete the monitoring wizard, Sill automatically researches your brand across the web: your website, review sites, press coverage, and public databases. The result is your Fact Profile: a collection of verified facts about your company and products, including:

• Company details: founding year, headquarters, leadership, funding
• Product information: pricing, features, integrations, known limitations
• Verified claims: facts confirmed or corrected through Watchdog's verification process or your own feedback

Watchdog uses your Fact Profile as ground truth when checking claims that AI engines make about your brand.

Reviewing Your Fact Profile

After the wizard completes, you'll see a banner on your dashboard inviting you to review your Fact Profile. The review page shows what Sill discovered, and you can correct anything that's inaccurate.

Your corrections aren't required. Watchdog works even without review, but correcting any inaccuracies improves the accuracy of future alerts. For example, if Sill found an outdated price on the web but you correct it, Watchdog will know to flag AI responses that still cite the old price.

---

What Watchdog Looks For

Watchdog extracts two types of information from AI responses:

Caveats

Caveats are qualifying statements that limit or hedge an AI engine's recommendation of your brand, such as noting high pricing or a steep learning curve.

Caveats are qualifying statements that limit, hedge, or add conditions to an AI engine's recommendation of your brand. Examples:

• "Brand X is a strong option, but it's expensive compared to alternatives"
• "Great for enterprise teams, though the learning curve is steep"
• "A popular choice, however some users report slow customer support"

Each caveat is categorized by the aspect it qualifies:

When a competitor is positioned as the solution to your weakness (e.g., "Brand X is expensive, but Brand Y offers a free tier"), Watchdog tracks that competitive framing as well.

Factual Claims

Factual claims are specific, verifiable assertions AI engines make about your brand, like pricing, features, or company details, that Watchdog checks against your fact profile.

Factual claims are specific, verifiable assertions AI engines make about your brand. Unlike opinions ("Brand X is great"), claims are concrete and checkable:

• "Brand X costs $50 per month"
• "Brand X integrates with Salesforce"
• "Brand X was founded in 2018"

Watchdog compares each claim against your Fact Profile and assigns a verification status:

---

Alerts

Watchdog generates alerts when it detects new risks, misinformation, or significant changes in how AI engines discuss your brand.

Watchdog generates alerts when it detects issues that need your attention. Alerts appear in a chronological feed on the Watchdog page, with the most recent at the top.

Learning Period

Watchdog needs a baseline before it can meaningfully detect changes. During your first three monitoring runs, Watchdog collects data but does not generate alerts. Alerts begin after the third run completes.

This prevents a flood of alerts on day one when everything would appear "new."

Alert Types

Reading an Alert

Each alert shows:

• Type: The alert category (icon and label)
• Headline: A brief description of what was detected
• Platforms: Which AI engines are involved (e.g., ChatGPT, Perplexity, Gemini)
• Products: Which of your products the alert relates to
• Recurrence: How many times this issue has been observed
• Timestamp: When the alert was first generated

You can expand an alert to see the full detail, including the original AI response excerpts and, for contradictions, a side-by-side comparison with your Fact Profile.

Responding to Alerts

You can take three actions on any alert:

Your feedback makes Watchdog smarter. Dismissing a false alarm teaches it not to re-flag the same issue. Confirming a real problem ensures it gets investigated.

Filtering Alerts

You can filter your alert feed by:

• Alert type: Focus on specific categories (e.g., only contradictions)
• Product: See alerts for a specific product
• Platform: Filter by AI engine
• Status: Show new, acknowledged, or dismissed alerts

---

Verification

Sill periodically verifies unresolved claims about your brand using web research, with weekly verification for premium subscribers and monthly for basic.

Some claims that AI engines make can't be checked against your Fact Profile alone, as they may reference information Sill hasn't seen before. Watchdog periodically runs a verification batch that researches these unresolved claims across the web to determine if they're accurate.

Verification Schedule

What Gets Verified

Not every claim goes through verification. Watchdog prioritizes:

1. Claims you've confirmed as needing investigation (highest priority)
2. Novel claims that appeared for the first time
3. Single-model anomalies that could be hallucinations
4. Claims appearing consistently across many platforms are deprioritized, as broad agreement is a signal of accuracy

Claims you've dismissed as correct are skipped entirely.

Verification Results

After a verification batch runs, you'll see the results reflected in your alerts:

• Refuted claims generate a Factual Contradiction alert with the correct information and sources
• Confirmed claims are added to your Fact Profile, and any related alerts are resolved
• Partially true claims are flagged for your review with additional context

For basic-plan users, monthly verification results are delivered as a digest: a summary of all confirmed issues found that month.

---

Cross-Platform Analysis

Watchdog compares what different AI platforms say about your brand to detect hallucinations and inconsistencies.

Watchdog automatically compares findings across AI platforms (ChatGPT, Perplexity, Gemini, Copilot, etc.) to surface inconsistencies:

• Single-platform claims: If only one AI engine asserts something about your brand while others don't, it's more likely to be a hallucination than a fact.
• Spreading claims: If a claim that was isolated to one platform last week now appears on multiple platforms, it may be gaining traction, accurate or not.
• Platform-specific caveats: If only one AI engine raises a particular concern about your brand, the issue may be specific to that platform's training data.

This cross-platform view helps you distinguish between isolated AI quirks and genuine reputation signals.

---

Basic vs Premium

Watchdog is available on all plans, with additional capabilities for premium subscribers:

---

Quick Reference

---

Need help understanding a specific alert? Reach out to our team at daniel@trysill.com.