diff --git a/R/slice.R b/R/slice.R
index 36118d5511..2508759977 100644
--- a/R/slice.R
+++ b/R/slice.R
@@ -19,6 +19,12 @@
 #' intrinsic notion of row order. If you want to perform the equivalent
 #' operation, use [filter()] and [row_number()].
 #'
+#' For `slice_sample()`, note that the weights provided in `weight_by` are
+#' passed through to the `prob` argument of [base::sample.int()]. This means
+#' they cannot be used to reconstruct summary statistics from the underlying
+#' population. See [this discussion](https://stats.stackexchange.com/q/639211/)
+#' for more details.
+#'
 #' @family single table verbs
 #' @inheritParams args_by
 #' @inheritParams arrange
@@ -93,9 +99,9 @@
 #' mtcars %>% slice_sample(n = 5)
 #' mtcars %>% slice_sample(n = 5, replace = TRUE)
 #'
-#' # you can optionally weight by a variable - this code weights by the
+#' # You can optionally weight by a variable - this code weights by the
 #' # physical weight of the cars, so heavy cars are more likely to get
-#' # selected
+#' # selected.
 #' mtcars %>% slice_sample(weight_by = wt, n = 5)
 #'
 #' # Group wise operation ----------------------------------------
@@ -293,6 +299,8 @@ slice_max.data.frame <- function(.data, order_by, ..., n, prop, by = NULL, with_
 #' @param weight_by <[`data-masking`][rlang::args_data_masking]> Sampling
 #'   weights. This must evaluate to a vector of non-negative numbers the same
 #'   length as the input. Weights are automatically standardised to sum to 1.
+#'   See the `Details` section for more technical details regarding these
+#'   weights.
 slice_sample <- function(.data, ..., n, prop, by = NULL, weight_by = NULL, replace = FALSE) {
   check_dot_by_typo(...)
   check_slice_unnamed_n_prop(..., n = n, prop = prop)
diff --git a/man/slice.Rd b/man/slice.Rd
index 89a54c74d5..00d673f75f 100644
--- a/man/slice.Rd
+++ b/man/slice.Rd
@@ -89,7 +89,9 @@ reach \code{n}/\code{prop}.}
 
 \item{weight_by}{<\code{\link[rlang:args_data_masking]{data-masking}}> Sampling
 weights. This must evaluate to a vector of non-negative numbers the same
-length as the input. Weights are automatically standardised to sum to 1.}
+length as the input. Weights are automatically standardised to sum to 1.
+See the \code{Details} section for more technical details regarding these
+weights.}
 
 \item{replace}{Should sampling be performed with (\code{TRUE}) or without
 (\code{FALSE}, the default) replacement.}
@@ -123,6 +125,12 @@ each group.
 Slice does not work with relational databases because they have no
 intrinsic notion of row order. If you want to perform the equivalent
 operation, use \code{\link[=filter]{filter()}} and \code{\link[=row_number]{row_number()}}.
+
+For \code{slice_sample()}, note that the weights provided in \code{weight_by} are
+passed through to the \code{prob} argument of \code{\link[base:sample]{base::sample.int()}}. This means
+they cannot be used to reconstruct summary statistics from the underlying
+population. See \href{https://stats.stackexchange.com/q/639211/}{this discussion}
+for more details.
 }
 \section{Methods}{
 
@@ -170,9 +178,9 @@ mtcars \%>\% slice_min(tibble(cyl, mpg), n = 1)
 mtcars \%>\% slice_sample(n = 5)
 mtcars \%>\% slice_sample(n = 5, replace = TRUE)
 
-# you can optionally weight by a variable - this code weights by the
+# You can optionally weight by a variable - this code weights by the
 # physical weight of the cars, so heavy cars are more likely to get
-# selected
+# selected.
 mtcars \%>\% slice_sample(weight_by = wt, n = 5)
 
 # Group wise operation ----------------------------------------