Is this just a wrapper around a language model?

A model does the reading, and that is not a secret. What sits around it is the part that matters: quotes verified against the source before anything is shown, transcription damage detected and findings inside it downgraded, names and product terms protected from being flagged as mistakes, metrics computed in code rather than guessed, and a profile that accumulates across meetings. That is the difference between an opinion and a report.

Can I not just write a good prompt?

You can get close on one meeting. What a prompt cannot do is check the model against the transcript afterwards, keep the count across twenty meetings, or stop the model from analysing your client along with you.

Is my transcript safer here than in a chatbot?

Different rather than automatically safer, and worth reading the detail: other participants are pseudonymised before anything leaves the server, the raw text is deleted on a schedule you set, and nothing is used to train a model. Whether a personal chatbot account meets your employer’s policy for a client transcript is a question only your employer can answer.

Why not just paste the transcript into ChatGPT?

What they do

ChatGPT — and Claude, and Gemini — will read a transcript you paste in and tell you about your English. It is free, it is instant, and for catching an obvious agreement error or an awkward phrase it is genuinely good. Anyone who has not tried this should try it before paying anyone, including us.

At a glance

	ChatGPT	betterput
Cost for one look	Free	Free once a month, then $15 USD
Quotes checked against the transcript	No	Every one, or it is dropped
Other participants	Analysed with you	Pseudonymised, never scored
Tracks patterns across meetings	Not on its own	Per 1,000 words, with trends
Answers follow-up questions	Yes	No

Where we differ

It will not tell you when it is guessing

Ask a chatbot what you said wrong and it answers confidently either way. It will paraphrase your sentence while presenting it as a quote, it will tidy up a line and then critique the tidied version, and it will produce a plausible finding about a part of the conversation it has lost track of. None of that announces itself. Here, every finding is matched character for character against the transcript before you see it, and one that fails the match is thrown away rather than rewritten — so a finding you are reading is one your own words support.

It grades everyone in the room

Paste a whole meeting and a chatbot analyses the whole meeting, including the client. Their words are context here and are never scored, and they are replaced by pseudonyms before anything is sent for analysis.

It starts from nothing every time

The value of this is not one report, it is the fourth one — the list of mistakes you keep making, counted per 1,000 words so a long meeting does not look like a bad one, moving down over months. A chat thread cannot do that unless you keep your own spreadsheet, and nobody keeps the spreadsheet.

It does not know what kind of meeting it was

Talking 70% of a client pitch is the point; talking 70% of a 1:1 is the finding. The numbers here are computed in code against the norm for the meeting type you chose, not estimated in prose by a model that was not told which kind of room it was in.

When they are the better choice

If you want one quick opinion on one meeting, a chatbot is free and it is fast, and you should use it. It is also better than this for anything conversational — arguing with the feedback, asking follow-up questions, rehearsing what to say next time. This is not a chat product and does not pretend to be.

Common questions

Is this just a wrapper around a language model?: A model does the reading, and that is not a secret. What sits around it is the part that matters: quotes verified against the source before anything is shown, transcription damage detected and findings inside it downgraded, names and product terms protected from being flagged as mistakes, metrics computed in code rather than guessed, and a profile that accumulates across meetings. That is the difference between an opinion and a report.
Can I not just write a good prompt?: You can get close on one meeting. What a prompt cannot do is check the model against the transcript afterwards, keep the count across twenty meetings, or stop the model from analysing your client along with you.
Is my transcript safer here than in a chatbot?: Different rather than automatically safer, and worth reading the detail: other participants are pseudonymised before anything leaves the server, the raw text is deleted on a schedule you set, and nothing is used to train a model. Whether a personal chatbot account meets your employer’s policy for a client transcript is a question only your employer can answer.

Now find out how you sounded in it

Paste it in and get your talk share, your hedging, and your three clearest English fixes.

Try it on this transcript

No account, and nothing is stored.