Review and Control What Is Sent to the AI

WTFAA sends page content to an AI model as part of a request that you start. Before most requests it shows Review before sending, a summary of what will be sent, with options to leave parts out.

This guide covers what a request contains, how the preview and its options work, the Auto Page Context setting, where requests go, and what local redaction can and cannot do. If you have not set up WTFAA yet, start with Getting started.

What a request can contain

Your message, or the built-in instruction of the quick action you clicked.
Page context: the page URL and title, the page text, and, where available, an outline of the page's headings and color values collected locally from the page.
Form context: the page URL and title and the structure of the forms on the page, including field labels and current field values. See Form Audit and Fill Test Data for the details.
A screenshot, if you captured one and kept it attached. See Screenshot analysis.
Conversation history: the recent messages of the current conversation. When a conversation gets long, earlier messages are summarized by the same AI service, and the summary is sent with later requests.

When the preview appears

Quick actions (the buttons of the Reader, Designer and Developer roles) always open the preview, unless you chose to skip the next one.
Typed messages open the preview when page or form context is attached or a screenshot is waiting to be sent. A message with neither is sent directly.

What the preview shows

Review before sending lists:

Message: the text that will be sent.
Context: the URL and Title; for a page, the Body length (number of characters of page text); for forms, the number of Forms and Fields.
Screenshot: Yes or No.
Model and Mode (Platform or BYOK).
Local redaction: what the local filter found, by type and count, or "No common sensitive patterns detected."

The preview shows these facts, not the full page text.

The preview options

Send page or form context: turn it off to send the request without page or form context. Since version 1.3.4, WTFAA then also does not add the cached page from Auto Page Context.
Send screenshot: available only when a screenshot is attached. Turn it off to send the message without the image.
Use locally filtered context: sends the filtered version of the context. It is available only when the filter found something, and it is turned on by default in that case.
Skip the next quick-action preview: shown only for quick actions. The next quick action is sent without a preview, using the filtered context. It applies once.

Click Send to AI to send, or Cancel to stop.

Auto Page Context

The setting is under Settings → Advanced → Connection & Generation and is on by default.

On: when you switch tabs or a page finishes loading, the side panel reads the active page and keeps its text in local extension storage. That page context is attached to typed messages, and the preview appears before sending.
Off: the side panel no longer reads pages on its own, and page context is not added to typed messages automatically. A typed message with no attached context and no screenshot is sent without a preview. Quick actions still read the page when you click them, and their preview still appears. Context that a quick action collected stays attached to later typed messages and is shown in their preview.

What local redaction does

The filter runs inside the extension before sending. It checks page text and form field values and replaces:

email addresses with [REDACTED_EMAIL]
phone-number-like digit sequences with [REDACTED_PHONE]
bearer tokens, JWTs, strings that look like API keys, and other long tokens (32 or more characters) with placeholders such as [REDACTED_TOKEN]
values of password fields (type password, or a name, id or label containing "password", "passcode" or "pin") with [REDACTED_PASSWORD]

Hidden form fields are removed. Inputs with type="hidden" are not collected from the page in the first place.

The filter does not change the URL, the page title, form labels, placeholders, form descriptions, your own message, screenshots, or conversation history.

Limits of local redaction

Redaction is based on text patterns. No automated filter is perfect:

It can miss sensitive data in formats it does not recognize, such as names, postal addresses or many ID numbers.
It can mask harmless text, such as long product codes or numbers that look like phone numbers.
When page context was sent, the model can ask for another section of the page. If that section is in WTFAA's local page cache, the extension sends it in a follow-up request, shown in the chat as "Fetching Segment …". This follow-up does not open the preview again, and the section text is not filtered.

If a page contains data you must not share, turn off Send page or form context, use a test account, or do not use WTFAA on that page.

Where requests go

Choose the mode under Settings → Advanced → Connection & Generation → AI Service:

Platform (No Key Required): available with a WTFAA Pro subscription. Requests go to the Datafrog AI gateway at ai.datafrog.io. You do not need your own API key.
Use My Own Keys: requests go directly from the extension to the provider of the model you selected: Google, OpenAI, Anthropic, DeepSeek or Groq. Your keys are not sent to Datafrog. They are saved in Chrome's extension sync storage, so they can sync to other browsers signed in to the same Chrome profile when Chrome sync is on. The provider's own terms apply to these requests.

Without Pro, WTFAA uses your own keys.

How Datafrog handles account, payment and usage information: Privacy Policy.
Translating pages: Page translation.
Role overview: Reader, Designer and Developer roles.