diff --git a/package.json b/package.json
index 08372d1141d3..3d611eb798b0 100644
--- a/package.json
+++ b/package.json
@@ -27,6 +27,7 @@
     "delete-orphan-translation-files": "tsx src/workflows/delete-orphan-translation-files.ts",
     "dev": "cross-env npm start",
     "find-orphaned-assets": "node src/assets/scripts/find-orphaned-assets.js",
+    "find-orphaned-features": "tsx src/data-directory/scripts/find-orphaned-features/index.ts",
     "find-past-built-pr": "tsx src/workflows/find-past-built-pr.ts",
     "fixture-dev": "cross-env ROOT=src/fixtures/fixtures npm start",
     "fixture-test": "cross-env ROOT=src/fixtures/fixtures npm test -- src/fixtures/tests",
diff --git a/src/data-directory/scripts/find-orphaned-features/find.ts b/src/data-directory/scripts/find-orphaned-features/find.ts
new file mode 100644
index 000000000000..c516fad8f8eb
--- /dev/null
+++ b/src/data-directory/scripts/find-orphaned-features/find.ts
@@ -0,0 +1,300 @@
+/**
+ * This script will loop over all pages, in all languages, and look at
+ * the following:
+ *
+ *    1. `title` in frontmatter
+ *    2. `intro` in frontmatter
+ *    3. `shortTitle` in frontmatter (if present)
+ *    4. the markdown body itself
+ *    5. The `versions:` frontmatter key (if the page is in English)
+ *
+ * Then it will search out the features mentioned based on `data/features/*.yml`
+ * It will make a Set of these (e.g. `dependabot-grouped-dependencies` and
+ * `ghas-enablement-webhook`) and one by one pluck them away.
+ *
+ * After the pages, it will loop over the reusables in English, and do the
+ * same search there. Once it's done the English, it loops over the
+ * reusables in the translations (if they exist) and does the same search.
+ *
+ * Lastly, it will output the remaining features, as relative file paths.
+ * For example, `data/features/havent-been-used-in-years.yml` so now you
+ * know that file can be deleted.
+ *
+ * NOTE: A lot of translations have corrupted Liquid. So if we can't parse
+ * the Liquid we fall back to string search. A regex will try to find
+ * all `{% ifversion ... %}` (and `elsif`) and search for any features
+ * mentioned inside that as a string.
+ *
+ */
+
+import fs from 'fs'
+import path from 'path'
+
+import chalk from 'chalk'
+import { TokenizationError } from 'liquidjs'
+
+import warmServer from '@/frame/lib/warm-server.js'
+import { getDeepDataByLanguage } from '@/data-directory/lib/get-data.js'
+import { getLiquidTokens } from '@/content-linter/lib/helpers/liquid-utils.js'
+import languages from '@/languages/lib/languages.js'
+import { correctTranslatedContentStrings } from '@/languages/lib/correct-translation-content.js'
+
+type Options = {
+  sourceDirectory: string
+  output?: string
+  verbose?: boolean
+}
+
+type Page = {
+  permalinks: Permalink[]
+  relativePath: string
+  fullPath: string
+  title: string
+  shortTitle?: string
+  intro: string
+  markdown: string
+  languageCode: string
+  versions: Record<string, string>
+}
+type Permalink = {
+  href: string
+  languageCode: string
+}
+
+export async function find(options: Options) {
+  const { sourceDirectory } = options
+  if (process.env.ENABLED_LANGUAGES && process.env.ENABLED_LANGUAGES === 'en') {
+    console.warn(
+      chalk.yellow(
+        `Only English is enabled. Be careful with the output.
+    To include all translations make sure they're available and that
+    ENABLED_LANGUAGES is not set or set to 'all'.`.replaceAll(/\s\s+/g, ' '),
+      ),
+    )
+  }
+  const site = await warmServer([])
+
+  const features = new Set(Object.keys(getDeepDataByLanguage('features', 'en')))
+  if (options.verbose) {
+    console.log(`Found ${features.size} features`)
+  }
+
+  const pageList: Page[] = site.pageList
+  if (options.verbose) {
+    console.log(`Searching ${pageList.length.toLocaleString()} pages`)
+  }
+
+  const t0 = new Date()
+  searchAndRemove(features, pageList, Boolean(options.verbose))
+  const t1 = new Date()
+
+  if (options.verbose) {
+    const color = features.size === 0 ? chalk.green : chalk.yellow
+    console.log(
+      color(
+        `Searched ${pageList.length.toLocaleString()} pages in ${formatDelta(t0, t1)}.
+      And found ${features.size} features remaining (i.e. orphans).`.replace(/\s\s+/, ' '),
+      ),
+    )
+  }
+
+  const remaining = Array.from(features).map((feature) =>
+    path.join(sourceDirectory, `${feature}.yml`),
+  )
+  if (options.output) {
+    if (options.output.endsWith('.json')) {
+      fs.writeFileSync(options.output, JSON.stringify(remaining, null, 2))
+    } else {
+      fs.writeFileSync(options.output, remaining.join('\n'))
+    }
+    if (!options.verbose) {
+      return
+    }
+  }
+  console.log(chalk.bold(`Orphans found (${remaining.length}):`))
+  for (const feature of remaining) {
+    console.log(chalk.green(feature))
+  }
+}
+
+function formatDelta(t0: Date, t1: Date) {
+  const ms = t1.getTime() - t0.getTime()
+  return `${(ms / 1000).toFixed(1)} seconds`
+}
+
+function searchAndRemove(features: Set<string>, pages: Page[], verbose = false) {
+  for (const page of pages) {
+    const content = page.markdown
+    // We actually never bother looking at the `versions:` frontmatter
+    // key in translations, so it doesn't matter if the translated
+    // frontmatter might have `versions: some-old-feature`.
+    if (page.languageCode === 'en') {
+      for (const [key, value] of Object.entries(page.versions)) {
+        if (key === 'feature') {
+          if (features.has(value)) {
+            features.delete(value)
+          }
+        }
+      }
+    }
+
+    const combined = `
+      ${content}
+      ${page.title || ''}
+      ${page.shortTitle || ''}
+      ${page.intro || ''}
+    `
+
+    checkString(combined, features, { page, verbose, languageCode: page.languageCode })
+  }
+
+  // Reusables are a bit special, as they are shared between languages.
+  // There'll always be a slight mismatch between files present on disk
+  // in English vs. translations.
+  // The translations never delete files, so there's often excess reusables
+  // on disk in translations. And the English might be ahead, meaning a file
+  // has been introduced in English but not yet translated.
+  // The code below loops over the English reusables, and takes note of the
+  // their relative paths and content. Then, we re-use the keys of that map
+  // to know which files, in the translations, to check. And when we read
+  // them in, we'll need the English equivalent content to be able to
+  // use the correctTranslatedContentStrings function.
+
+  const englishReusables = new Map<string, string>()
+  for (const filePath of getReusableFiles(path.join(languages.en.dir, 'data', 'reusables'))) {
+    const relativePath = path.relative(languages.en.dir, filePath)
+    const fileContent = fs.readFileSync(filePath, 'utf-8')
+    checkString(fileContent, features, { filePath, verbose, languageCode: 'en' })
+    englishReusables.set(relativePath, fileContent)
+  }
+  for (const language of Object.values(languages)) {
+    if (language.code === 'en') continue // Already did that in the loop above
+
+    for (const [relativePath, englishFileContent] of Array.from(englishReusables.entries())) {
+      const filePath = path.join(language.dir, relativePath)
+      try {
+        const fileContent = fs.readFileSync(filePath, 'utf-8')
+        const correctedFileContent = correctTranslatedContentStrings(
+          fileContent,
+          englishFileContent,
+          {
+            code: language.code,
+            relativePath,
+          },
+        )
+
+        checkString(correctedFileContent, features, {
+          filePath,
+          verbose,
+          languageCode: language.code,
+        })
+      } catch (error) {
+        if (error instanceof Error && 'code' in error && error.code === 'ENOENT') {
+          // That a reusable does *not* exist in a translation is
+          // perfectly expected. It means that English reusable was
+          // most likely added recently and the translation hasn't been
+          // translated yet.
+          continue
+        }
+        throw error
+      }
+    }
+  }
+}
+
+function getReusableFiles(root: string): string[] {
+  const here = []
+  for (const file of fs.readdirSync(root)) {
+    const filePath = `${root}/${file}`
+    if (fs.statSync(filePath).isDirectory()) {
+      here.push(...getReusableFiles(filePath))
+    } else if (file.endsWith('.md') && file !== 'README.md') {
+      here.push(filePath)
+    }
+  }
+  return here
+}
+
+const IGNORE_ARGS = new Set(['or', 'and', 'not', '<', '>', 'ghes', 'fpt', 'ghec', '!=', '='])
+
+function checkString(
+  string: string,
+  features: Set<string>,
+  {
+    page,
+    filePath,
+    languageCode,
+    verbose = false,
+  }: { page?: Page; filePath?: string; languageCode?: string; verbose?: boolean } = {},
+) {
+  try {
+    for (const token of getLiquidTokens(string)) {
+      if (token.name === 'ifversion' || token.name === 'elsif') {
+        for (const arg of token.args.split(/\s+/)) {
+          if (IGNORE_ARGS.has(arg)) continue
+          if (isFloat(arg)) continue
+
+          if (features.has(arg)) {
+            features.delete(arg)
+          }
+        }
+      }
+    }
+  } catch (error) {
+    if (error instanceof TokenizationError) {
+      // If it happens in English, it's a serious error
+      if (languageCode === 'en') throw error
+
+      // The translation might, currently, have corrupted liquid
+      // So treat it as a string
+      if (verbose)
+        console.log(
+          `TokenizationError in ${page ? page.fullPath : filePath}. Treating ${page ? page.fullPath : filePath} as a string and using regex`,
+        )
+
+      for (const feature of Array.from(findByRegex(features, string))) {
+        features.delete(feature)
+      }
+    } else {
+      throw error
+    }
+  }
+}
+
+function findByRegex(features: Set<string>, string: string) {
+  const found = new Set<string>()
+  for (const match of string.match(/\{%\s*(ifversion|elsif)\s*(.*?)\s*%\}/g) || []) {
+    for (const feature of Array.from(features)) {
+      const regex = new RegExp(`\\s${escapeRegex(feature)}(\\s|%)`, 'i')
+      if (regex.test(match)) {
+        found.add(feature)
+      }
+    }
+  }
+  return found
+}
+
+const test = findByRegex(
+  new Set(['placeholder', 'foo-bar']),
+  `
+  placeholder
+
+  {%ifversion placeholder-foo or fpt%}
+  {%   elsif   not-placeholder   %}
+  {%   elsif   foo-bar%}
+  {%endif %}
+
+  {% data reusables.enterprise-migration-tool.placeholder-table %}
+  {% data placeholder %}
+`,
+)
+console.assert(test.has('foo-bar'), test.toString())
+console.assert(!test.has('placeholder'), test.toString())
+
+function escapeRegex(string: string) {
+  return string.replace(/[/\-\\^$*+?.()|[\]{}]/g, '\\$&')
+}
+
+function isFloat(x: any) {
+  return !!(parseFloat(x) + 1)
+}
diff --git a/src/data-directory/scripts/find-orphaned-features/index.ts b/src/data-directory/scripts/find-orphaned-features/index.ts
new file mode 100644
index 000000000000..b4ec025170ca
--- /dev/null
+++ b/src/data-directory/scripts/find-orphaned-features/index.ts
@@ -0,0 +1,18 @@
+import { program } from 'commander'
+import { find } from './find'
+
+program
+  .name('find-orphaned-features')
+  .description(
+    "Compare what's in data/features/*.yml with what's mentioned in Markdown and frontmatter",
+  )
+
+program
+  .command('find')
+  .description('Figure out what features are not being used')
+  .option('-s, --source-directory <directory>', 'Source directory', 'data/features')
+  .option('-o, --output <output-file>', 'Output file')
+  .option('-v, --verbose', 'Verbose')
+  .action(find)
+
+program.parse(process.argv)