What Facebook shows logged-out visitors: 2026 data

Ten posts, and that is the end of it

A public group hands somebody without an account ten posts. A public page does the same. Scrolling down loads no more of them — the page simply ends, and whatever it did not render does not exist for anyone without an account. Three weeks earlier, in mid-August 2026, the same probe returned about four. The number belongs to Facebook rather than to us, which is why we never promise it in the product.

The consequence is practical and it matters to anyone comparing tools. A vendor promising "every post in the group" is either signing in to somebody's account or saying something untrue. The time window we filter results by works only on what the page managed to show; it does not reach into an archive, because for a logged-out visitor there is no archive.

The default view hides fresh posts

A group opens sorted by popularity, not by date. On a quiet group the two orders coincide and the difference is invisible, which is exactly how somebody concludes that the sort parameter changes nothing. On an active group, checked on 2 September 2026, the default view had an eight-day-old post at the top and was missing a post from that same day, which the chronological order showed without trouble.

So we open a group's address with the chronological parameter rather than the default. A page has no equivalent parameter and we take it as it comes. The post-age filter only ever works on what the page rendered, so without this change it would have nothing left to keep.

From a datacentre address there is nothing to see

A probe on 14 August 2026, on live proxies and with no account anywhere: from an address belonging to a datacentre, Facebook closes every path with a login wall. That includes the lightweight mbasic version and an ordinary public page, both of which the guides online describe as open to everybody. From a residential address the same page opens without an account and shows its content.

This is the one fact on this list that decides whether a product like ours can exist at all. There is no way around it in code: no header change, no browser mode and no delay between visits alters what Facebook does with a datacentre address.

The order of obstacles is fixed

The cookie consent banner comes first, the login prompt second, the content last. While the banner is standing it intercepts every click, and the page stays half-loaded. Skipping that step understates two things at once: the number of visible posts and the transfer, which are the two numbers every cost calculation rests on.

Hence the methodological point, which is worth more than the figure itself: a measurement of Facebook without a described order of steps is a measurement of something else. Our first runs, in which the banner was never dismissed, returned about four posts — and for a while that looked like a platform limit rather than our own mistake.

What is never visible

The list of people who reacted to a post is unavailable. The counter is visible, the names are not, and no plan changes that, because the boundary is Facebook's. A tool promising that list is signed in to somebody's account, and that account carries the risk.

Comment and share counters are sometimes there, but they carry no labels. They sit right behind the reaction bar as two buttons with numbers, and only their order tells them apart. With two we read both; with one we stay with comments and a zero for shares. We understate deliberately, because an invented number on that screen is worse than a small one.

The date of a video post and of a photo post sits behind a different link from the date of a plain post. A parser that knows only the post address returns such an entry with no date and under an arbitrary identifier. On one large page that was half the entries, so it is not an edge case.

What it weighs and how long an address lasts

17 August 2026, sixty visits, 222.7 MB in total: a group 4.22 MB, a page 3.65 MB. Comments render in the same load, so a wider scan costs not one byte more — which is the reason there is no surcharge here for scope. The earlier figure, 5.8 MB, predates resource blocking and is no longer usable for costing anything.

A sticky session holds one exit address for exactly thirty minutes; we checked that on the same day with thirty-two probes, one a minute. At roughly twenty-one seconds per visit, one lease will serve about eighty-five loads. That is a harder ceiling than any ban threshold, and it is what decides how many addresses parallel work needs.

What this measurement does not say

Sixty visits from one address did not trigger a block. That is a lower bound, not a threshold: we do not know where the threshold sits and we do not pretend to. Anyone sizing a pool of addresses against ban risk rather than against parallelism is sizing it against a number nobody measured.

Facebook changes without notice, and this page is the evidence: the post count grew from about four to ten in three weeks. That is why every measurement here carries a date and why we re-run them instead of copying them forward. A measurement without a date is an anecdote, even if it was true on the day it was taken.