Does AI dictation read your screen for context?

Some AI dictation apps read text from the app you are typing into. Reading that text can help the app get names and jargon right. Each vendor defines "context" differently, so you need to read what each app says it reads before you dictate anything sensitive.

Wispr Flow works from text, not screenshots

Wispr Flow's help article on Context Awareness says the feature "works from text and does not take screenshots" ¹. Its privacy page says the setting reads limited text from the app you are dictating into, so that names and jargon come out right ². The help article says the feature is "on by default" ³, so it is active unless you change the setting. If you do not want it, change the setting before you dictate.

The privacy policy says more than the help article does. If you enable the optional feature, Wispr "may collect limited, relevant content from the specific app in use (such as the text on the screen)" ⁴. The policy describes what may be collected, and the help article describes what the feature reads. Both are the company's own statements, and neither gives a count of how often this happens. Read both before you decide.

Superwhisper reads the field you are writing in

Superwhisper describes its feature more broadly. Its documentation says context awareness "lets Superwhisper process your dictation with relevant information from your screen, which produces more accurate results" ⁵.

The application context reads "text from current input fields or text editor" ⁶. That is the text you are already writing in the field where you dictate.

Super Mode is the mode that uses all of this. The documentation says it is the "only built-in mode with all three context types on by default" ⁷. Context is set mode by mode, so a mode you use for notes may read less than the one you use for email.

Screenshots are the open question

Neither app's documentation says it sends screenshots, and Wispr's help article says it does not take them. That is different from proving no screenshot is ever taken. Third-party reports have made claims about screenshots, and we did not verify them. A user who is worried should not rely on any single statement. The practical test is to dictate a non-sensitive passage while watching the app's network activity on your own machine.

The help article also draws a line that matters for sensitive fields. Standard macOS password fields are excluded, but "custom or web-based password fields may be read like normal text fields" ⁸. So the password protection is strong for the usual fields and weaker on some websites.

It depends on the app, the mode and what you write

Four things decide how much an app reads:

• The app's settings. Wispr Flow's setting is on by default ³. Superwhisper's context is set per mode, so the mode you pick decides what is read.
• The text you are writing. Wispr Flow says its reading is there so that names and jargon come out right ².
• The sensitivity of the field. Standard password fields are protected, and custom or web-based ones may not be.
• The accuracy claim. Superwhisper says context gives more accurate results ⁵, and we found no independent test of that claim. Ask whether the gain is worth what the app reads on the text you dictate.

What to do

1. Open the context setting in each app before you dictate anything sensitive. In Wispr Flow it is on by default ³, so decide whether you want it.
2. For Superwhisper, check which context types each of your modes uses, and keep only the ones that mode needs.
3. After a dictation, open the History view and review what was sent. Superwhisper's Super Mode documentation says you can open a dictation in History "to review everything that was sent to the AI" ⁹.
4. For client or medical work, prefer a mode with no context, or a tool that runs on the device. Our question on whether AI dictation can run on your own computer covers the offline options.
5. Do not dictate passwords or account numbers into any app until you have checked how it treats those fields.

Yap, which we make, is an on-device dictation app for Mac and Windows. Check it with the same steps you use for any other app: find its context setting, read its privacy page and test it with a non-sensitive passage first. For how voice data flows in general, read the guide on whether voice to text is safe. The guide to AI dictation explains how context fits into the rest of the pipeline, and the our overview of voice and AI lists the related subjects. For which model reads the text after the context is captured, see which AI models dictation apps use.