Add accuracy and uncertainty-flagging guidance to always-on content instructions (#61901)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: lecoursen

.github/instructions/content.instructions.md

@@ -16,6 +16,14 @@ Before committing content changes, always:
 3. **Verify [AUTOTITLE] links** point to existing articles
 4. **Run tests** on changed content: `npm run test -- src/content-render/tests/render-changed-and-deleted-files.ts`
 
+## Accuracy and flagging uncertainty
+
+AI assistants can produce plausible but incorrect information. To protect readers and writers:
+
+* Never fabricate product behavior, settings, permissions, API parameters, or limits. If you cannot verify a detail from the source material or the product itself, say so instead of guessing.
+* Flag uncertainty at the level of the specific claim, not the whole article. Identify the individual sentence, step, value, or setting you are unsure about, rather than labeling an entire article as "needs review."
+* After writing or editing, give the writer a checklist of each specific claim to verify, noting where it appears (the exact step or sentence) and why it needs checking. Prioritize claims where being wrong could have serious consequences, such as security, authentication, permissions, billing, or any irreversible action, and recommend confirming those with the relevant product or engineering team.
+
 ## Bullet lists
 
 The bulleted points in a bullet list should always be denoted in Markdown using an asterisk, not a hyphen.