This guide walks you through installing Second Opinion, running your first analysis, and understanding what you receive back. It assumes no prior familiarity with the product. If you hit something unexpected, jump to the Troubleshooting section at the end, or write to support@pansophyai.com.
Second Opinion is a tool for challenging your assumptions. You give it a document. It reads the language carefully and returns a written opinion describing what the language shows.
Typical use cases:
Second Opinion is not a hiring tool. If you do use it as one input in a process that touches an employment decision, it does not make the decision, does not rank candidates, and does not produce a hireability score. You remain the decision-maker. The tool describes language patterns; you weigh what they mean in your context.
To use Second Opinion you need:
You do not need to create a separate Pansophy AI account. Authentication is handled through Microsoft single sign-on using your existing Microsoft 365 credentials.
There are two install paths. Pick whichever is easier for you.
Visit Microsoft AppSource, search for "Second Opinion" by Pansophy AI Inc., and click Get It Now. AppSource will route you through your Microsoft 365 tenant and add the app to Teams. If your tenant requires admin approval for new apps, your request is sent to your IT administrator and you will see a confirmation message. Once approved, the app appears in your Teams app drawer.
Open Teams. Click Apps in the left sidebar. Search for "Second Opinion." Click Add. Teams will install the app in personal scope (just for you) and open a chat with the bot.
Some organizations restrict Teams app installation to pre-approved apps. If you see a message saying admin approval is needed, your request is automatically routed to the Teams administrator for your tenant. If you want to speed the approval along, point your admin to pansophyai.com/secondopinion/for-it-admins/, which answers the questions IT admins typically need answered before approving a new Teams app.
When you first open Second Opinion, the bot welcomes you and briefly explains what the tool does. There is no account setup, no form to fill out, no profile to build. The bot introduces itself, confirms you are in a US-based Microsoft 365 tenant, and invites you to upload a document whenever you are ready.
Second Opinion does not remember prior sessions. Each session starts fresh. If you analyze an email on Monday and come back on Tuesday to look at a different document, the bot will not reference Monday's session, will not have remembered patterns from it, and will not draw comparisons. This is a deliberate design choice. The tool reads the document in front of it, not your history with it.
You also do not need to tell the bot what kind of document it is. Second Opinion reads the language regardless of whether it is an email, a transcript, or a Slack thread. The intake is simply: upload the document.
A full analysis takes about thirty seconds from upload to opinion delivery. Here are the five steps.
In the bot chat, use the paperclip attachment icon to attach a PDF, DOCX, TXT, or VTT file. The bot will confirm receipt and start processing. You do not paste the text into chat; you upload the file itself.
Behind the scenes, the bot extracts the text from your file, scrubs out personally identifiable information using Microsoft Azure AI Language, passes the cleaned text to the scoring engine, and then asks a language model to write a narrative description of what the scores suggest. You will see status messages so you know the bot has not stalled.
The bot returns a written opinion in the chat. This is prose, not a dashboard. It describes the patterns the tool found in the language and what those patterns suggest about how the person communicated. See the next section for more on how to read it.
After the opinion is delivered, the bot offers four preset rewrite buttons: Shorter, Deeper, Different audience, Different angle. There is also a free-text option called Something else, in case none of the presets match what you want. Rewrites reuse the analysis that was already done on your document, so they are fast (a few seconds) and are not separately billed.
When you are finished, close the Teams chat or move on. The session also closes automatically after thirty minutes of inactivity, or when you upload a new document. At session close, the uploaded file, the extracted text, and the PII-redacted text are all discarded. See the Privacy section below for what is retained and what is not.
The output is a written opinion. A few paragraphs of careful observations about the language patterns in what you uploaded, what those patterns suggest about how the person communicated, and what a thoughtful reviewer might want to look at more closely. We call it an opinion, not a report or a profile, because it is meant to be read and weighed, not filed and cited.
Here is what the opinion is not:
The opinion is framed descriptively throughout. Phrases like "the language shows" or "the writer tended toward" are typical. Phrases like "the writer is" or "this person will" would be predictive and are explicitly avoided. The tool is not always right, and that is not a bug. It is a design constraint. Descriptive observations are what the tool can defensibly produce from word patterns. Predictions about individuals are not, so we do not make them.
Audio files, image files, and video files are not supported directly. If you have an audio recording, transcribe it first (Teams does this automatically for meetings) and upload the resulting transcript as a VTT or TXT file.
For transcripts with clear speaker labels (VTT files from Teams meetings, transcripts with "Speaker 1:" or "Interviewer:" markers, or chat logs with author tags), Second Opinion identifies each speaker and analyzes their language separately. For unlabeled free-flowing conversation, the tool analyzes the document as a whole.
Your first five (5) document analyses are free. After that, usage is billed per word. Rewrites of a given opinion, within the same session, are always free and do not count against your five-document entitlement.
After your five free analyses, usage is billed at $0.0005 per word analyzed, metered through Microsoft commercial marketplace. A few concrete examples:
You are invoiced by Microsoft on your existing Microsoft bill. There is no subscription to sign, no per-seat license, and no tenant minimum. Your organization pays only for the words you actually analyze.
The billing unit is one document analysis, counted after PII scrubbing (so header noise, timestamps, and redacted PII tokens do not pad your word count). Rewrites of an existing opinion within the same session reuse the original analysis and are not separately billed. Only a new document submission starts a new billable session.
Second Opinion runs entirely inside Microsoft Azure in the United States (East US 2 region). No data leaves Microsoft's cloud environment. The pipeline applies PII scrubbing before any analysis is performed.
Pansophy AI applies a split retention model:
A session ends when you explicitly close it, when you submit a new document for analysis, or after thirty (30) minutes of inactivity.
Second Opinion is not used to train Microsoft or OpenAI foundation models. Customer data submitted to Azure OpenAI Service is not used to train, retrain, or fine-tune any model, as contractually enforced by the Azure OpenAI Service terms.
For full detail, see the Privacy Policy and the Responsible AI Policy.
No. The tool analyzes the language in your document against a psycholinguistic lexicon (a dictionary of word categories derived from the Linguistic Inquiry and Word Count research tradition). It does not compare the person you are reading about to other candidates, other customers, or anyone else. There is no leaderboard and no peer ranking.
No. Uploaded documents are not used to train, fine-tune, or retrain any model. This applies to both the behavioral scoring engine and the Azure OpenAI language model that generates the written opinion.
No. The written opinion is the product. Numeric scores exist internally as part of how the tool produces the narrative, but they are not exposed to users. This is a deliberate design choice: numeric scores are easy to misinterpret, easy to misuse in decisions they are not designed to support, and harder to challenge than a written opinion.
Not in the current version. You can copy the opinion text from the Teams chat and paste it into a document of your choice.
English only in the current version. Psycholinguistic analysis accuracy may vary for speakers of English as a second language. This limitation is disclosed during onboarding.
You will get substantively similar observations. Minor variations in the wording of the opinion are normal because the final synthesis uses a language model, which introduces some natural variation. The underlying scoring is deterministic; the prose around it is not.
Yes. Uploading your own writing to see what the tool says about your language is a legitimate and sometimes useful use case. You are the subject of the analysis, so no third-party consent is required.
No. The tool has no access to the speaker's identity, demographics, or any personal characteristic. It analyzes word patterns. It does not know whose words they are.
No. Second Opinion is not a hiring tool. It can be used as one input in a hiring process, but it does not make hiring decisions, does not score candidates on a hireability scale, and does not rank applicants. If you are using it in connection with employment decisions, you are responsible for ensuring compliance with applicable employment and anti-discrimination law in your jurisdiction.
Challenge it. That is what the tool is for. The tagline is "Challenge your assumptions" and that includes the assumption that the opinion is right. Use the rewrite buttons to ask for a different angle, try the document again, or simply disregard the opinion. You are the decision-maker.
Check your internet connection and the Microsoft Teams status page. If Teams is working but the bot is not, wait a minute and try again; the bot may be cold-starting. If it still does not respond, send a simple message like "hi" to the bot. If that gets no reply within a minute, write to support@pansophyai.com.
Check that the file is under 10 MB and is one of the supported types (PDF, DOCX, TXT, VTT). If the file type is correct but upload fails, try saving the file with a simpler filename (no special characters, no spaces) and uploading again.
Most analysis errors trace to one of three causes: the document contains very little text (PDFs that are mostly images, for example), the document is in a language other than English, or the document is a file format that technically matches the extension but was not created in the expected way (a PDF that is actually a scanned image with no OCR layer, for instance). Try the document again, or try a different format of the same content.
You get five free document analyses, lifetime. Rewrites of an existing opinion do not count. If the counter looks wrong, write to support@pansophyai.com with your Microsoft 365 account email and we will verify.
If you installed from AppSource and your tenant requires admin approval, the app will not appear until your administrator approves the request. Check with your IT administrator. If you are the administrator, check your pending app approvals in the Teams admin center.
That is not necessarily a bug. The whole point of the tool is to surface patterns you might have missed or weighed differently. Before assuming the opinion is wrong, consider whether it is noticing something genuine that cuts against your first impression. If you still disagree, use the rewrite buttons to request a different angle, or simply weigh it against your own reading and move on. The tool is one input.
Two email addresses cover everything:
support@pansophyai.com
Technical issues, billing questions, account problems, general how-do-I questions
privacy@pansophyai.com
Data subject requests, privacy inquiries, legal matters
Response times: general support within two business days, security issues within one business day, data subject requests within seven business days. If you need faster attention, write to support@pansophyai.com and put URGENT in the subject line.