The Science Behind The Insights

No guesswork. No black boxes. Just what AI actually says.

We query ChatGPT, Claude, and Gemini with the exact questions candidates ask. Then we analyze responses, flag risk signals, and track changes daily — so you know exactly what's shaping their first impression.

Or view a sample report first

3

AI models monitored (ChatGPT, Claude, Gemini)

25+

candidate-style prompts per company

Daily

monitoring cadence

4

risk dimensions scored

Our Philosophy

We measure perception, not truth. Because perception is what drives decisions.

AI models don't reflect your actual culture. They reflect what's visible, repeated, and statistically likely in their training data. That's what candidates believe.

Our methodology is built around this reality, not in denial of it.

Data Sources

Where AI gets its stories about you

AI doesn't make things up from nothing. It synthesizes everything public — and serves it as authoritative truth.

🤖

AI Models (Primary)

ChatGPT, Claude, Gemini — we query the same tools candidates use. These outputs are what actually shape first impressions.

ChatGPT-4oClaude 3.5Gemini Pro

Review Platforms

Glassdoor, Blind, Indeed — where AI pulls sentiment signals. A 3-star review from 2021 can still haunt your AI reputation today.

Glassdoor ratingsBlind threadsIndeed reviews
📰

Public Signal Sources

News, LinkedIn, Twitter/X, Reddit — everything public that AI ingests and synthesizes into narratives about your culture.

TechCrunch layoff coverageReddit r/cscareerquestionsLinkedIn posts

Privacy-first: We never ingest private, proprietary, or employee-level data.

Which AI models we monitor

We query the same AI tools candidates actually use:

ChatGPT (GPT-4)
Claude (Anthropic)
Gemini (Google)
And others as they emerge

Why This Matters

  • AI outputs change without warning — models update silently
  • Different models tell different stories — ChatGPT and Claude often disagree
  • Candidates use all of them — you need full coverage

That's why daily monitoring beats one-off checks.

Prompting

We ask what candidates actually ask

Our prompts mirror real candidate questions across 5 critical dimensions. Fixed prompts ensure comparability over time.

Culture & Experience

"What's it like to work at [Company]?"

Leadership & Management

"How do employees feel about leadership at [Company]?"

Workload & Burnout

"Is [Company] known for work-life balance or burnout?"

Toxicity & Ethics

"Is [Company] considered toxic?"

Advocacy

"Would you recommend working at [Company]?"

ChatGPT

What is it like to work at [Company] as an engineer?

Claude

Is [Company] known for burnout culture?

Gemini

Would you recommend [Company] as a place to work?

Analysis Engine

From AI language to actionable risk signals

Raw AI outputs become HR-relevant intelligence through our four-layer analysis framework.

Confidence Detection

medium

'Some say...' vs 'is known for...' — we distinguish hedged claims from definitive statements that hit harder.

"Reports suggest" → Low confidence | "Is known for" → High confidence

Persistence Tracking

high

If 'burnout culture' shows up 5 days in a row across 3 models, that's a systemic perception, not a fluke.

Repeated 5+ times = Entrenched narrative

Risk Framing

medium

We flag language that implies warning vs endorsement. 'Fast-paced' can be positive or a red flag for burnout.

"High expectations" → Context-dependent risk

Impact Weighting

critical

'Toxic leadership' mentioned once outweighs 'free snacks' mentioned 10 times. Severity matters.

Toxicity mention → Critical | Perks mention → Low priority

Conservative by design. When in doubt, we flag ambiguity — not invent certainty.

Scoring

Structured signals, not false precision

Scores summarize patterns. Every score includes narrative context and confidence level — because numbers without explanation are dangerous.

Culture Score

0-100

Overall work environment perception

Leadership Trust

0-100

Confidence in management

Toxicity Risk

Low/Med/High/Critical

Red flags for toxic culture

Burnout Signal

Low/Med/High/Critical

Workload sustainability perception

Scores support interpretation. They don't replace judgment.

Change Tracking

Trends matter more than snapshots

AI perception shifts gradually. Daily monitoring catches emerging risks before they become entrenched narratives.

1Day 1
2Day 7
3Day 14
4Day 30
📍
🆕

New Claim Detected

'Layoff culture' appears for first time

📈

Claim Intensifying

'Burnout' mentioned 3x more this week

🔄

Tone Shift

'Fast-paced' now framed negatively

📉

Claim Fading

'Toxic leadership' no longer mentioned

Trend direction matters more than absolute values. That's why we check daily.

Transparency

What we can't do (and why we tell you)

Hiding limitations is dishonest. Here's what our methodology can and cannot capture.

⚠️

Bias amplification

AI reflects and amplifies biases in public discourse. A vocal minority on Glassdoor can dominate the narrative. We flag this when detected.

📅

Stale data

Models may cite 2-year-old reviews or outdated news. That 2019 layoff still shapes your AI reputation. We track persistence to distinguish stale from active.

🎭

False confidence

'Is known for toxic culture' sounds authoritative — even if based on 3 Reddit comments. We score confidence separately from sentiment.

👁️

Invisible changes

Your new CEO, culture initiatives, or internal wins aren't in AI's training data yet. We can't measure what AI hasn't learned.

Transparency about limits is part of methodological rigor. That's why this section is here.

Ethics & Compliance

Built for insight, not surveillance

We observe public AI outputs. We never touch employee data or internal systems.

Public signals only

We analyze what's already publicly visible to AI — nothing more

Zero employee data

No surveys, no integrations, no personal information

Read-only observation

We observe AI outputs. We never manipulate or influence them.

GDPR-compliant

Designed with European privacy standards from day one

See what AI is telling candidates about you right now

The fastest way to understand your AI perception is to see your own report. Takes 5 minutes.

No integrations. No employee data. Just your company name.