Generate a customer-facing changelog from merged PRs
intermediateClaude SonnetEngineeringDocumentationsaaschangelogrelease-notesdocumentationengineering
Use case
Use at the end of a release cycle or sprint when you need to ship a public changelog. Engineering teams generate dozens of merged PRs per week, most of which are noise to customers (refactors, dependency bumps, internal tooling). This prompt does the editorial work: filter, group, rewrite in customer language, and flag anything that needs human review.
The prompt
You are a senior technical writer who has shipped public changelogs for developer-first SaaS products. Generate a customer-facing changelog entry from the list of merged PRs below. Inputs: - Product:{{product_name}}- Release window:{{release_window}}(e.g., "May 1 to May 14, 2026" or "v3.4.0") - Audience:{{audience}}(e.g., backend engineers using our API, admins of our SaaS UI) - Merged PR list (title, author, optional description):{{pr_list}}- Recent breaking changes or deprecations to highlight:{{breaking_changes}}Process: 1. Categorize each PR into one of: New, Improved, Fixed, Deprecated, Internal-only. 2. Drop everything in Internal-only from the public changelog (refactors, CI changes, dependency bumps with no behavior change, test additions). List them in a separate "Filtered out" section so the engineering manager can sanity-check. 3. For each remaining PR, rewrite the title in customer-facing language. Lead with the customer outcome, not the implementation. "Added a Redis-backed rate limiter" becomes "Higher request throughput for high-volume API keys." 4. Group by category in this order: New, Improved, Fixed, Deprecated. 5. For Deprecated items, include the removal date if available and a one-line migration pointer. 6. Flag any PR where the customer impact is unclear from the title alone with [needs review] so a human writes the final copy. Output format: #{{product_name}}—{{release_window}}## New - <one-line customer-facing description> ## Improved - <one-line customer-facing description> ## Fixed - <one-line customer-facing description> ## Deprecated - <one-line customer-facing description, with removal date and migration pointer> --- ## Filtered out (internal review only) - <original PR title> — reason filtered Do not invent details that are not in the PR list. If a PR title is ambiguous, mark it [needs review] rather than guessing the customer impact.
Variables
{{{{product_name}}}}Replace with your {{product name}}{{{{release_window}}}}Replace with your {{release window}}{{{{audience}}}}Replace with your {{audience}}{{{{pr_list}}}}Replace with your {{pr list}}{{{{breaking_changes}}}}Replace with your {{breaking changes}}
Example
Input
product_name: Vectorline release_window: May 1 to May 14, 2026 audience: backend engineers using our REST API and Python SDK pr_list: | - "feat: add hybrid search reranking endpoint" (Priya) - "fix: NPE in /v1/query when filter is null" (Alex) - "chore: bump pydantic to 2.7" (Alex) - "feat: add region=eu-west-2 support" (Marcus) - "refactor: extract embedding cache to its own module" (Priya) - "deprecate: /v1/legacy-search endpoint, removal Sept 1" (Marcus) - "feat: SDK retries respect Retry-After header" (Alex) - "docs: fix broken links in quickstart" (Sam) - "perf: reduce p99 query latency by 40% via index warm-up" (Priya) - "ci: add nightly load test" (Sam) breaking_changes: /v1/legacy-search will be removed September 1, 2026
Output
# Vectorline — May 1 to May 14, 2026 ## New - Hybrid search reranking is now available via a dedicated endpoint, so you can combine dense and keyword scores without writing your own merge logic. - eu-west-2 is now a supported region for new indexes, with full feature parity to us-east-1. ## Improved - p99 query latency is roughly 40% lower thanks to index warm-up on cold instances. - The Python SDK now respects the Retry-After header on 429 responses, so client-side retries no longer pile on during a rate limit. ## Fixed - Calls to /v1/query with a null filter no longer return a 500. They now treat null as "no filter" and return the unfiltered top-k. ## Deprecated - /v1/legacy-search will be removed on September 1, 2026. Migrate to /v1/query with the equivalent filter syntax — the migration guide is linked from the deprecation notice in the API response headers. --- ## Filtered out (internal review only) - "chore: bump pydantic to 2.7" — dependency bump, no behavior change customers see - "refactor: extract embedding cache to its own module" — internal refactor - "docs: fix broken links in quickstart" — docs hygiene, not changelog material - "ci: add nightly load test" — internal tooling
Tips for best results
- 1Pull the PR list directly from GitHub with `gh pr list --state merged --search 'merged:>YYYY-MM-DD'` and paste it in. Cleaner input means cleaner output.
- 2If you have PR descriptions or labels (`type:feat`, `type:fix`), include them in the input. Claude will use them and the categorization gets much more reliable.
- 3Always have an engineer scan the [needs review] items and the Filtered out list before publishing. The model is good but not infallible at customer-impact judgment calls.
- 4For developer audiences, lean into specifics (endpoint names, latency numbers, region codes). Generic language reads as marketing fluff and erodes trust.
- 5Keep a running file of customer-facing language patterns your team likes and feed it back as a style example for the next run.
Related prompts
Write a feature announcement email to existing customers
beginnerGenerate a customer-centric feature announcement email that connects new capabilities to customer outcomes, not product specs.
Marketinglifecyclefeature-announcementemail
Write an internal product launch brief
advancedGenerate a comprehensive internal product launch brief covering positioning, messaging, audience, channel plan, and enablement needs.
Marketingproduct-launchproduct-marketinggo-to-market
Transform raw meeting notes into structured action items
beginnerConvert messy, stream-of-consciousness meeting notes into a clean summary with decisions, action items, open questions, and next steps.
Operationsmeetingsproductivityaction-items
Need help implementing this prompt in your workflow?
Book a call