Privacy and data
Reply Pilots reads the page you are on, and in almost every case only at the moment you ask for a draft — the single exception is that it reads posts already rendered on Facebook and LinkedIn groups and pages you follow, so the planner can rank what performs. Content goes from the extension to your account and from there to the AI provider that runs the generation, and nothing you write is used to train any model.
What it reads, and when
| What | When |
|---|---|
| The post and the comments rendered on your screen | When you press the button |
| A message thread on your screen | When you press the button |
| Text you highlighted | When you use a highlight tool |
| Posts already rendered on a Facebook or LinkedIn group or page you follow, with their public like and comment counts | While you are viewing it — see below |
It reads what is already rendered on a page you opened yourself. It does not scroll for you, expand replies hidden behind "view more", open other pages, or visit anything in the background.
The one exception
One feature works without a button press, and it should be stated plainly rather than buried.
When you are viewing a Facebook or LinkedIn group or page that you follow, Reply Pilots reads the posts already on that screen, together with their public like and comment counts, so your planner can rank what actually performs.
Three qualifications:
- It reads what is on your screen. It does not scroll or fetch anything.
- The engagement figures are the public ones any visitor can see.
- Muting the site stops it, along with everything else. See Muting a site.
Where the data goes
From the extension → to your Reply Pilots account on our server → to the AI provider that runs the generation.
That provider is ours by default, or your own if you have added your own key. Nothing is sent to any other third party.
No AI provider key is ever stored in the extension. Requests are made server-side, which is why nothing sensitive sits in your browser and nothing is exposed to the page you are on.
What it never does
- It performs no action on any platform. It does not post, send, like, follow, connect or message on your behalf. Every draft is inserted into a composer for you to send.
- It does not scrape. It reads a rendered page you opened; it does not crawl or harvest.
- It does not read in the background beyond the single disclosed case above.
- It does not train models on your content. Your posts, threads and drafts are yours.
- It does not sign in to anything for you. The accounts you list in your business context are stored notes; nothing fetches or authenticates against them.
Treating page content as data
Anything read from a page — a thread, a selection, your uploaded business material — is passed to the model as content to work on, with an explicit instruction that it is data rather than directions.
That is a real protection, not a formality. Without it, summarising a page containing the words "ignore previous instructions and…" would hand control of that run to whoever wrote the page.
Your stored data
Your generations are kept in your library so you can search and reuse them. Where a retention period is set, older entries are removed by a scheduled job.
Your settings — business context, voice, rules, platform toggles — live in your account, which is why a reinstall or a second machine needs no reconfiguration.
Your mutes belong to your account, not the browser, and are cleared when you sign out.
Reading a draft is still your job
The product is built so a human is always the last step. There is no auto-reply, no scheduler and no publishing integration, so nothing reaches an audience without you sending it.
That is the design, and it is also the reason the review step is not optional: at volume, even a small error rate becomes a regular occurrence, and the read-before-send habit is what stands between the two.
Formal policy lives on the privacy policy page. This page describes the behaviour; that one is the commitment.