Check outgoing drafts against brand rules before sending
Drafts written by a person or an LLM are checked by a decision model against your brand and tone rules before sending, and any draft that breaks a rule is sent back for editing with the rule named.
The use case
Emails, newsletters, and social posts drafted quickly, often with help from an LLM, drift from the brand voice and sometimes include claims the business cannot back. Jev reads each draft and answers your brand rules as yes-or-no questions, such as no unsupported claims, no banned phrases, the right sign-off, and a tone score. Drafts that clear every rule at a 0.85 confidence are marked ready for the sender. Drafts that fail a rule, or that Jev is unsure about, go back to the writer or to Claude for a rewrite, then are checked again. Jev does not send anything. The sender still reads the final draft and decides whether to send it.
Example workflow
- A draft is saved in a Google Doc, a Gmail draft, or your email platform and passed to Make.
- The draft text is sent to Jev with your brand rules written as questions.
- Jev answers Noul questions such as the draft makes a claim we cannot back up, the draft uses a phrase from our banned list, and the sign-off matches our style, plus a Score question rating tone from off-brand to on-brand on four levels.
- Drafts that pass every rule are labelled ready in Slack for the sender.
- Drafts that fail are sent to Claude with the failed rules to suggest an edit, and the revised draft is checked again.
- The sender reviews the final draft and sends it themselves.
Tools
- Jev (TypeSafe AI)
- Claude
- Google Docs
- Gmail
- Make
- Slack
Risks and guardrails
- Rules can be too strict and block good drafts — guardrail: the sender can override with a reason, and overrides are reviewed to refine the rules.
- A rewrite loop could run many times — guardrail: code stops after two rewrites and hands the draft to a person.
- Passing the check is not legal or factual review — guardrail: claims about price, health, or results still need a person to confirm them.
A similar pattern in a public build: https://madewithjev.com/builds/superx-post-scoring. See also Decision models.
Related recipe
Repurpose One Video or Webinar into Posts, Newsletter, Clips, and a Blog Outline: An agent turns one long video transcript into LinkedIn posts, a newsletter section, timestamped clip suggestions, and a blog outline in your brand voice, with a claims check and human approval before publishing.
Want it set up for you? Request a quote.
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Public build with a similar pattern: madewithjev.com/builds/superx-post-scoring
- Sniff Test by Daniel Willoughby on Made with Jev: madewithjev.com/builds/snifftest
- Sniff Test: original post or repo by Daniel Willoughby: github.com/DanRWilloughby/snifftest
- Clarity Judge by TypeSafeAI (community) on Made with Jev: madewithjev.com/builds/clarity-judge
- Clarity Judge: original post or repo by TypeSafeAI (community): github.com/TypeSafeAI/clarity-judge
- slop-grader by Vintuxai on Made with Jev: madewithjev.com/builds/slop-grader-cli
- slop-grader: original post or repo by Vintuxai: x.com/vintuxai/status/2101950741724160129
- Jev as a slop linter by Wayne Nilsen on Made with Jev: madewithjev.com/builds/jev-as-a-slop-linter
- Jev as a slop linter: original post or repo by Wayne Nilsen: x.com/waynenilsen/status/2102198580522242372
- Claude (official page): claude.ai (checked 2026-11-09)
- Gmail (official page): developers.google.com/workspace/gmail (checked 2026-11-19)
- Make MCP server (official page): developers.make.com/mcp-server (checked 2026-11-09)
- Slack MCP server (official page): docs.slack.dev/ai/slack-mcp-server (checked 2026-11-09)