How it works
Most analysis tools read the words. ClearMirror reads the structure underneath.
The gap between words and structure
A document can sound sharp and be structurally broken. A conversation can feel cooperative and be doing something else entirely.
The words land. The structure operates. And the gap between what text says and what text does is where persuasion, manipulation, and self-deception live.
Reading more carefully doesn't close the gap. Neither does running the text through one AI. Both get absorbed into the surface.
Three models, one question
Every analysis runs through three frontier AI models — Claude Opus 4.6, GPT-5.4, and Gemini 3.1 Pro — simultaneously, using the same prompt.
Three models with different training histories looking at the same text produce three independent reads. Where they agree, something real is there. Where they disagree, you see it and judge for yourself.
No single model decides what's true. The comparison is the signal.
Stress Test
For documents. Five analysis methods, each testing something different: what survives when filler is stripped, what the document would need to be true to work, what a skeptical reviewer catches, whether the argument holds under compression. Pick one method or run all five. Each method runs through all three models. You read the outputs side by side.
Multi-Review runs four reviewer perspectives inside one method — skeptic, builder, customer, competitor — giving you more than a single read per model.
Observer
For conversations. Seven detection patterns that watch for specific structural moves: premises hardening without agreement, vocabulary shifting between speakers, who controls the topic, who carries the burden of proof, directives wearing softeners, and more. Run one pattern at a time, or use Full Analysis to run all seven in a single pass. Observer returns structured findings — each with a confidence level and a counterpoint that explains when the pattern might be a false alarm.
You see what each model said
Observer returns structured findings. Each one has a confidence level (high / moderate / low) and a counterpoint — an alternative explanation for what you're seeing. You can see where models agreed and where they diverged.
Stress Test returns the full analytical text from each model, side by side. No scores, no composite judgment. You read what each model said and compare directly.
Both tools save your results to your dashboard so you can come back to them. You can delete saved results anytime.
We're not that
This is not a fact-checker. It doesn't tell you whether claims are true in the world.
This is not a sentiment tool. It doesn't tell you if a conversation felt good.
This is not a summary generator. It won't compress a document for you.
It shows you what's operating under the surface of text. What you do with that is yours.
What happens to your data
When you upload a document or paste a transcript, the content is sent to Claude (Anthropic), GPT (OpenAI), and Gemini (Google) for analysis through their APIs. All three providers' current API policies state that customer inputs are not used to train their models. Those are their policies, not ours — we encourage you to review them directly if it matters to your use case.
We do not store the documents or transcripts you upload. We do store the analysis results to your dashboard so you can return to them. Saved results are protected by database-level security and accessible only to your account. You can delete saved results at any time.
ClearMirror does not train on your data, sell your data, or share it beyond the API calls required to run your analysis.