Reports
Report an inappropriate AI suggestion.
If englisher suggested something wrong, offensive or harmful, write to us. We read every report.
We usually reply within 5 working days.
When to report
This page is for cases where the model’s output was not OK, not just stylistically off:
- an offensive, vulgar or demeaning response;
- harmful or dangerous content, or content encouraging something bad;
- a correction that reverses the meaning of what you wrote;
- output containing things that were not in your text and look made up;
- anything you think should not come out of the application.
What to include in the report
The more of these you describe, the better the chance that we can reproduce the problem:
- What you asked for — a correction or a change of tone.
- What came back — the content of the response or a description of it.
- Approximate date and time — to the nearest hour is enough.
- Application version — you will find it in the englisher settings window.
- Mode — correction on your computer or in englisher Cloud.
Do not paste confidential data
The report goes by ordinary email. It is not covered by the guarantees we enforce on the AI gateway — mail passes through mail servers and stays in your sent folder and our inbox.
Do not send passwords, medical data, client data, content covered by professional secrecy or anything you do not want to leave in an email. If the problematic sentence contains such data, describe it in your own words or replace names and numbers with example ones.
Why we ask for a description instead of checking logs
Because there is nothing to check. englisher Cloud does not store the content of corrected texts or the model’s responses. Only counters remain in the database: the number of input and output tokens, cost, model name, request status, its ID and the SHA-256 hash of the system instruction. A sentence cannot be recovered from these.
This works the other way too: the application opens this page without automatically attaching the selected text or the model’s output. You decide what reaches us. Details are in the privacy policy.
Where suggestions come from
Style corrections and tone variants come from a generative model. A model can be wrong — it may change the meaning of a sentence, add something that was not there, or suggest wording that does not fit the situation. Read the output before you send the text on.
We use reports to improve the instructions for the model and to choose the model itself. We do not train anything on them.