Evidence-linked language analysis
Subtell flags observable shifts, omissions, non-responses and pressure language in conversations and statements. Every finding links back to the words. It never decides who is lying, who is at fault, or what you should do.
Start with the sample. No private conversation needed.
Before you paste anything
No verdicts. Findings use calibrated language and name reasonable alternatives.
No training. Submitted content is not used to train Subtell or OpenAI models.
Private knowledge retrieval. Pinecone is queried only for Subtell’s private forensic methodology; your transcript is not uploaded to that knowledge base.
You stay in control. Saved reports can be deleted, and deleting your account removes its associated reports and jobs.
Screenshots, PDFs and DOCX files can be added securely in the dashboard. An account keeps the resulting report private to you.
A real shape of output
This illustrative preview shows how Subtell separates the observed language from possible significance and limitations.
Analysis summary
The exchange moves from a specific question about a fee to a request for a refund. Several questions receive general or indirect responses. These are sensitivity signals, not proof of intent or wrongdoing.
“Can you tell me why?” → “Fees are applied in line with our terms and conditions”
The reply points to the terms as a whole and does not say which term applies or why the fee was charged. This may reflect a scripted response or missing information; it does not establish whether the fee was correct.
“Which part of the terms?” → “We understand this has been frustrating”
The reply acknowledges the customer's feelings but does not answer the question asked. This may be standard courtesy wording. On its own it does not show an intention to avoid the question.
“So will you refund it?” → “Requests like this would normally be reviewed within eight weeks”
The wording describes a usual process rather than committing to an outcome for this case. It may reflect policy limits or caution. More context would be needed to tell these apart.
Illustrative preview. A complete report also records evidential limits, chronology and confidence.
Paste text or upload supported files. You can add neutral context, but Subtell treats the source language as the evidence.
The report groups evidence-linked linguistic features, explains what they may indicate, and states plausible alternatives.
Subtell does not diagnose, blame, rank people, or produce a credibility verdict. You retain the decision.