diff --git a/.github/workflows/update-stats.yaml b/.github/workflows/update-stats.yaml
new file mode 100644
index 0000000..6206cf1
--- /dev/null
+++ b/.github/workflows/update-stats.yaml
@@ -0,0 +1,45 @@
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: '0 6 * * 1'  # Every Monday at 6am UTC
+
+name: Update stats
+
+jobs:
+  update-stats:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    env:
+      GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
+      CROSSREF_EMAIL: ${{ secrets.CROSSREF_EMAIL }}
+    steps:
+      - uses: actions/checkout@v7
+
+      - uses: r-lib/actions/setup-r@v2
+        with:
+          install-r: true
+          use-public-rspm: true
+
+      - name: Install system dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install libcurl4-openssl-dev
+
+      - uses: r-lib/actions/setup-renv@v2
+
+      - name: Collect package and paper metrics
+        run: source("snippets/update_stats.R")
+        shell: Rscript {0}
+
+      - name: Commit the new rows
+        run: |
+          git config user.name "epiforecasts-bot"
+          git config user.email "epiforecasts-bot@users.noreply.github.com"
+          git add _data/package-stats.csv _data/paper-citations.csv
+          if git diff --staged --quiet; then
+            echo "No change this week"
+          else
+            git commit -m "Weekly package and paper metrics"
+            git push
+          fi
diff --git a/_automation/get_stats.R b/_automation/get_stats.R
new file mode 100644
index 0000000..848a9e4
--- /dev/null
+++ b/_automation/get_stats.R
@@ -0,0 +1,189 @@
+## Collect weekly metrics for the group's packages and papers.
+##
+## Two sources of truth, both already maintained elsewhere: the r-universe
+## registry plus _data/software-extras.yml decide which packages count, and
+## _data/papers.bib decides which papers count. Nothing here needs its own
+## hand-kept list.
+
+openalex_mailto <- function() {
+  email <- Sys.getenv("CROSSREF_EMAIL")
+  if (identical(email, "")) NULL else email
+}
+
+## Citation counts are tracked for papers only. Packages rarely declare a
+## citable reference: of sixteen, three do, and one of those is a CRAN package
+## DOI that is almost never cited, so a package citation column would be blank
+## or misleading nearly everywhere.
+##
+## OpenAlex is generous but rate-limits anonymous callers, so identify us where
+## an address is available.
+openalex_citations <- function(doi) {
+  url <- paste0("https://api.openalex.org/works/doi:", utils::URLencode(doi))
+  mailto <- openalex_mailto()
+  if (!is.null(mailto)) url <- paste0(url, "?mailto=", mailto)
+  res <- tryCatch(jsonlite::fromJSON(url), error = function(err) NULL)
+  if (is.null(res) || is.null(res$cited_by_count)) {
+    return(list(
+      citations = NA_integer_,
+      title = NA_character_,
+      year = NA_integer_
+    ))
+  }
+  list(
+    citations = as.integer(res$cited_by_count),
+    title = res$display_name %||% NA_character_,
+    year = as.integer(res$publication_year %||% NA)
+  )
+}
+
+## ---- packages --------------------------------------------------------------
+
+## Every repo the software page shows, from the same two sources it uses.
+package_repos <- function() {
+  universe <- jsonlite::read_json(
+    "https://github.com/epiforecasts/universe/raw/main/packages.json"
+  )
+  flagged <- purrr::keep(universe, ~ isTRUE(.x$display_website))
+  from_universe <- purrr::map_chr(flagged, function(e) {
+    sub("^https://github.com/", "", e$url)
+  })
+
+  extras <- yaml::read_yaml("_data/software-extras.yml")
+  from_extras <- purrr::map_chr(extras, "repo")
+
+  unique(c(from_universe, from_extras))
+}
+
+github_stats <- function(repo) {
+  info <- tryCatch(
+    gh::gh("/repos/{repo}", repo = repo),
+    error = function(err) NULL
+  )
+  if (is.null(info)) {
+    return(list(stars = NA_integer_, forks = NA_integer_, issues = NA_integer_))
+  }
+  list(
+    stars = as.integer(info$stargazers_count %||% NA),
+    forks = as.integer(info$forks_count %||% NA),
+    issues = as.integer(info$open_issues_count %||% NA)
+  )
+}
+
+## r-universe reports a download count per package, sourced from cranlogs. For
+## a package that is not on CRAN that count is always zero, which would read as
+## "nobody downloaded it" rather than "this figure does not apply", so keep the
+## CRAN flag alongside it and blank the count when it does not apply.
+universe_downloads <- function() {
+  pkgs <- jsonlite::fromJSON(
+    "https://epiforecasts.r-universe.dev/api/packages",
+    simplifyVector = FALSE
+  )
+  stats <- purrr::map(pkgs, function(p) {
+    on_cran <- isTRUE(p$`_cranurl`)
+    list(
+      package = p$Package,
+      on_cran = on_cran,
+      downloads = if (on_cran) p$`_downloads`$count %||% NA else NA
+    )
+  })
+  ## registry names and repo names differ in case (RBi vs rbi)
+  purrr::set_names(stats, tolower(purrr::map_chr(stats, "package")))
+}
+
+## Packages listed from outside the epiforecasts registry have no entry there,
+## so ask cranlogs directly. A package absent from CRAN returns a zero, which
+## is why the answer is only trusted when crandb knows the package.
+cranlogs_downloads <- function(package) {
+  known <- tryCatch(
+    !is.null(
+      jsonlite::fromJSON(paste0("https://crandb.r-pkg.org/", package))$Package
+    ),
+    error = function(err) FALSE
+  )
+  if (!isTRUE(known)) return(list(on_cran = FALSE, downloads = NA_integer_))
+
+  res <- tryCatch(
+    jsonlite::fromJSON(
+      paste0("https://cranlogs.r-pkg.org/downloads/total/last-month/", package)
+    ),
+    error = function(err) NULL
+  )
+  count <- if (is.null(res) || length(res$downloads) == 0) {
+    NA_integer_
+  } else {
+    as.integer(res$downloads[[1]])
+  }
+  list(on_cran = TRUE, downloads = count)
+}
+
+collect_package_stats <- function(on_date = Sys.Date()) {
+  repos <- package_repos()
+  downloads <- universe_downloads()
+
+  rows <- purrr::map(repos, function(repo) {
+    package <- basename(repo)
+    gh_stats <- github_stats(repo)
+    dl <- downloads[[tolower(package)]] %||% cranlogs_downloads(package)
+
+    data.frame(
+      date = as.character(on_date),
+      package = package,
+      repo = repo,
+      stars = gh_stats$stars,
+      forks = gh_stats$forks,
+      open_issues = gh_stats$issues,
+      on_cran = dl$on_cran,
+      downloads_last_month = dl$downloads,
+      stringsAsFactors = FALSE
+    )
+  })
+
+  dplyr::bind_rows(rows)
+}
+
+## ---- papers ----------------------------------------------------------------
+
+dois_from_papers <- function(bib = "_data/papers.bib") {
+  lines <- readLines(bib, warn = FALSE)
+  pattern <- "DOI = \\{[^}]+\\}"
+  hits <- regmatches(lines, regexpr(pattern, lines, ignore.case = TRUE))
+  unique(gsub("DOI = \\{|\\}", "", hits, ignore.case = TRUE))
+}
+
+collect_paper_citations <- function(on_date = Sys.Date()) {
+  dois <- dois_from_papers()
+  rows <- purrr::map(dois, function(doi) {
+    info <- openalex_citations(doi)
+    data.frame(
+      date = as.character(on_date),
+      doi = doi,
+      title = info$title,
+      year = info$year,
+      citations = info$citations,
+      stringsAsFactors = FALSE
+    )
+  })
+  dplyr::bind_rows(rows)
+}
+
+## ---- writing ---------------------------------------------------------------
+
+## Append this week's rows, keeping one row per subject per date so a rerun on
+## the same day corrects rather than duplicates.
+append_stats <- function(new_rows, path, key) {
+  if (nrow(new_rows) == 0) return(invisible(NULL))
+  combined <- if (file.exists(path)) {
+    old <- utils::read.csv(path, stringsAsFactors = FALSE)
+    dplyr::bind_rows(old, new_rows)
+  } else {
+    new_rows
+  }
+  combined <- combined |>
+    dplyr::distinct(
+      dplyr::across(dplyr::all_of(c("date", key))),
+      .keep_all = TRUE
+    ) |>
+    dplyr::arrange(date, .data[[key]])
+  utils::write.csv(combined, path, row.names = FALSE)
+  invisible(combined)
+}
diff --git a/_data/package-stats.csv b/_data/package-stats.csv
new file mode 100644
index 0000000..9790601
--- /dev/null
+++ b/_data/package-stats.csv
@@ -0,0 +1,17 @@
+"date","package","repo","stars","forks","open_issues","on_cran","downloads_last_month"
+"2026-08-25","EpiNow2","epiforecasts/EpiNow2",140,40,66,TRUE,598
+"2026-08-25","RBi","sbfnk/RBi",25,8,0,TRUE,197
+"2026-08-25","baselinenowcast","epinowcast/baselinenowcast",10,2,63,TRUE,192
+"2026-08-25","cfrnow","epiforecasts/cfrnow",2,0,0,FALSE,NA
+"2026-08-25","contactsurveys","epiforecasts/contactsurveys",2,3,0,TRUE,263
+"2026-08-25","distspec","epiforecasts/distspec",1,0,7,TRUE,164
+"2026-08-25","epichains","epiverse-trace/epichains",10,4,33,TRUE,229
+"2026-08-25","epimixr","sbfnk/epimixr",4,0,5,FALSE,NA
+"2026-08-25","epinowcast","epinowcast/epinowcast",67,22,120,FALSE,NA
+"2026-08-25","forecastbaselines","epiforecasts/forecastbaselines",0,0,2,FALSE,NA
+"2026-08-25","lopensemble","epiforecasts/lopensemble",7,3,8,FALSE,NA
+"2026-08-25","qrensemble","epiforecasts/qrensemble",3,0,5,FALSE,NA
+"2026-08-25","rbi.helpers","sbfnk/rbi.helpers",4,2,0,TRUE,161
+"2026-08-25","ringbp","epiforecasts/ringbp",18,12,16,TRUE,161
+"2026-08-25","scoringutils","epiforecasts/scoringutils",64,24,119,TRUE,934
+"2026-08-25","socialmixr","epiforecasts/socialmixr",44,15,1,TRUE,903
diff --git a/_data/paper-citations.csv b/_data/paper-citations.csv
new file mode 100644
index 0000000..c548a6b
--- /dev/null
+++ b/_data/paper-citations.csv
@@ -0,0 +1,40 @@
+"date","doi","title","year","citations"
+"2026-08-25","10.1016/j.epidem.2024.100765","Characterising information gains and losses when collecting multiple epidemic model outputs",2024,11
+"2026-08-25","10.1016/j.idm.2025.04.005","Visual preferences for communicating modelling: a global analysis of COVID-19 policy and decision makers",2025,4
+"2026-08-25","10.1016/s2214-109x(20)30074-7","Feasibility of controlling COVID-19 outbreaks by isolation of cases and contacts",2020,2830
+"2026-08-25","10.1038/s41467-021-22213-0","Implications of the school-household network structure on SARS-CoV-2 transmission under school reopening strategies in England",2021,43
+"2026-08-25","10.1038/s41467-021-25207-0","A pre-registered short-term forecasting study of COVID-19 in Germany and Poland during the second wave",2021,86
+"2026-08-25","10.1073/pnas.2203019119","Simulating respiratory disease transmission within and between classrooms to assess pandemic management strategies at schools",2022,11
+"2026-08-25","10.1098/rsfs.2025.0007","Improving modelling for epidemic response: a progress update from a community of UK infectious disease modellers",2025,0
+"2026-08-25","10.1098/rsif.2020.0084","Probabilistic reconstruction of measles transmission clusters from routinely collected surveillance data",2020,12
+"2026-08-25","10.1098/rstb.2020.0283","Exploring surveillance data biases when estimating the reproduction number: with insights into subpopulation transmission of COVID-19 in England",2021,52
+"2026-08-25","10.1101/2022.01.08.22268920","Estimation of the test to test distribution as a proxy for generation interval distribution for the Omicron variant in England",2022,68
+"2026-08-25","10.1101/2022.10.12.22280917","Evaluating an epidemiologically motivated surrogate model of a multi-model ensemble",2022,2
+"2026-08-25","10.1101/2025.03.04.25323088","The utility of infectious disease modelling in informing policy for outbreak response: a scoping review",2025,0
+"2026-08-25","10.1101/2025.04.03.25325159","Fine-Grid Spatial Interaction Matrices for Surveillance Models, with Application to Influenza in Germany",2025,0
+"2026-08-25","10.1101/2025.04.10.25325611","The influence of model structure and geographic specificity on predictive accuracy among European COVID-19 forecasts",2025,1
+"2026-08-25","10.1101/2025.08.01.25332807","Mind the Baseline: The Hidden Impact of Reference Model Selection on Forecast Assessment",2025,2
+"2026-08-25","10.1101/2025.08.09.25331484","The influence of ensemble size and composition on the performance of combined real-time COVID-19 forecasts",2025,0
+"2026-08-25","10.1111/rssa.12974","Sebastian Funk, Sam Abbott and Johannes Bracher’s Discussion Contribution to the Papers in Session 2 of The Royal Statistical Society’s Special Topic Meeting on Covid-19 Transmission: 11 June 2021",2022,2
+"2026-08-25","10.1126/science.abf9648","The impact of population-wide rapid antigen testing on SARS-CoV-2 prevalence in Slovakia",2021,195
+"2026-08-25","10.1126/science.add4507","Heavy-tailed sexual contact networks and monkeypox epidemiology in the global outbreak, 2022",2022,220
+"2026-08-25","10.1186/s12916-019-1288-7","Real-time analysis of the diphtheria outbreak in forcibly displaced Myanmar nationals in Bangladesh",2019,56
+"2026-08-25","10.1186/s12916-022-02271-x","Comparative assessment of methods for short-term forecasts of COVID-19 hospital admissions in England at the local level",2022,34
+"2026-08-25","10.12688/wellcomeopenres.15718.1","The transmissibility of novel Coronavirus in the early stages of the 2019-20 outbreak in Wuhan: Exploring initial point-source exposure sizes and durations using scenario analysis",2020,83
+"2026-08-25","10.12688/wellcomeopenres.15842.3","Estimating the overdispersion in COVID-19 transmission using outbreak sizes outside China",2020,676
+"2026-08-25","10.12688/wellcomeopenres.16006.2","Estimating the time-varying reproduction number of SARS-CoV-2 using national and subnational case counts",2020,195
+"2026-08-25","10.12688/wellcomeopenres.16344.3","Implication of backward contact tracing in the presence of overdispersed transmission in COVID-19 outbreaks",2021,91
+"2026-08-25","10.12688/wellcomeopenres.19380.2","Human judgement forecasting of COVID-19 in the UK",2024,3
+"2026-08-25","10.12688/wellcomeopenres.19601.1","Improving modelling for epidemic responses: reflections from members of the UK infectious disease modelling community on their experiences during the COVID-19 pandemic",2024,15
+"2026-08-25","10.12688/wellcomeopenres.25027.2","Baseline nowcasting methods for handling delays in epidemiological data",2026,1
+"2026-08-25","10.1371/currents.outbreaks.406ae55e83ec0b5193e30856b9235ed2","Temporal Changes in Ebola Transmission in Sierra Leone and Implications for Control Requirements: a Real-time Modelling Study",2015,131
+"2026-08-25","10.1371/journal.pcbi.1006785","Assessing the performance of real-time epidemic forecasts: A case study of Ebola in the Western Area region of Sierra Leone, 2014-15",2019,114
+"2026-08-25","10.1371/journal.pcbi.1010405","Comparing human and model-based forecasts of COVID-19 in Germany and Poland",2022,31
+"2026-08-25","10.1371/journal.pcbi.1011393","Scoring epidemiological forecasts on transformed scales",2023,50
+"2026-08-25","10.1371/journal.pcbi.1011453","Evaluating the use of social contact data to produce age-specific short-term forecasts of SARS-CoV-2 incidence in England",2023,14
+"2026-08-25","10.1371/journal.pgph.0004675","How does policy modelling work in practice? A global analysis on the use of epidemiological modelling in health crises",2025,9
+"2026-08-25","10.1371/journal.pgph.0005120","The utility of infectious disease modelling in informing decisions for outbreak response: A scoping review",2025,2
+"2026-08-25","10.21105/joss.03290","covidregionaldata: Subnational data for COVID-19 epidemiology",2021,13
+"2026-08-25","10.7554/elife.70767","Inference of the SARS-CoV-2 generation time using UK household data",2022,73
+"2026-08-25","10.7554/elife.81916","Predictive performance of multi-model ensemble forecasts of COVID-19 across European nations",2023,109
+"2026-08-25","10.7554/elife.98005.3","Forecasting the spatial spread of an Ebola epidemic in real time: Comparing predictions of mathematical models and experts",2025,1
diff --git a/_software-item.Rmd b/_software-item.Rmd
index ad78816..ba77b74 100644
--- a/_software-item.Rmd
+++ b/_software-item.Rmd
@@ -23,6 +23,9 @@ if ("{{language}}" != "" && "{{language}}" != "NA") {
 if ("{{updated}}" != "" && "{{updated}}" != "NA") {
   meta <- c(meta, paste0("Updated ", format(as.Date("{{updated}}"), "%b %Y")))
 }
+if ("{{downloads_text}}" != "") {
+  meta <- c(meta, "{{downloads_text}}")
+}
 if (length(meta) > 0) {
   cat('<div class="project-meta">', paste(meta, collapse = " · "), '</div>\n')
 }
diff --git a/snippets/update_stats.R b/snippets/update_stats.R
new file mode 100644
index 0000000..0152b63
--- /dev/null
+++ b/snippets/update_stats.R
@@ -0,0 +1,21 @@
+## Weekly metrics for the group's packages and papers.
+##
+## Appends one row per package and per paper to two CSVs under _data/. The
+## history is the point: these numbers cannot be reconstructed later, so the
+## job records them as it goes.
+
+source("_automation/get_stats.R")
+
+`%||%` <- function(x, y) if (is.null(x)) y else x
+
+today <- Sys.Date()
+
+message("Collecting package metrics")
+packages <- collect_package_stats(today)
+message(sprintf("  %d packages", nrow(packages)))
+append_stats(packages, "_data/package-stats.csv", key = "package")
+
+message("Collecting paper citations")
+papers <- collect_paper_citations(today)
+message(sprintf("  %d papers", nrow(papers)))
+append_stats(papers, "_data/paper-citations.csv", key = "doi")
diff --git a/software.qmd b/software.qmd
index f7ef50b..f991125 100644
--- a/software.qmd
+++ b/software.qmd
@@ -14,6 +14,20 @@ library(magrittr)
 stale_months <- 24
 ```
 
+```{r load-stats}
+# Weekly download figures, collected by .github/workflows/update-stats.yaml.
+# Absent until that job has run at least once, so treat it as optional.
+downloads_by_package <- list()
+stats_file <- "_data/package-stats.csv"
+if (file.exists(stats_file)) {
+  stats <- read.csv(stats_file, stringsAsFactors = FALSE)
+  stats <- stats[stats$date == max(stats$date), ]
+  stats <- stats[!is.na(stats$downloads_last_month), ]
+  # registry and repo names differ in case (rbi vs RBi)
+  downloads_by_package <- split(stats$downloads_last_month, tolower(stats$package))
+}
+```
+
 ```{r load-team}
 # Load team members and create lookup by GitHub username
 team <- fs::dir_ls("_data/team", regexp = "\\w+\\-\\w+\\.yml") |>
@@ -216,9 +230,22 @@ all_packages %>%
       )
     }
 
+    downloads <- downloads_by_package[[tolower(e$Package)]]
+    downloads_text <- if (is.null(downloads)) {
+      ""
+    } else {
+      # name the source, so a package without a figure reads as "not on CRAN"
+      # rather than "not used"
+      paste(
+        formatC(downloads, big.mark = ",", format = "d"),
+        "CRAN downloads/month"
+      )
+    }
+
     knitr::knit_expand(
       "_software-item.Rmd",
       Package = e$Package,
+      downloads_text = downloads_text,
       logo_html = logo_html,
       part_of_html = part_of_html,
       description = e$description,