đ Works best with Slack. Connect it and Jeremy AI finds the community posts that need a reply. It sends nothing.
Paste your community post export. Get one file back: a post table sorted by age with a status on every post, three reply drafts, two discussion prompts, a shout-out draft, a not-provided list and a check table. Nothing is posted.
You'll see the SAMPLE post export pasted in with the rules, the queue built in about 4 seconds on screen (the real wait was about 74 seconds), the file saved with nothing posted, the post table, the reply drafts, the discussion prompts and shout-out, the not-provided list and the check table.
What this use case does and who it's for
You run a community. Posts pile up. Some get ten replies. Some get none. The ones with none are the ones that matter most, because the member feels ignored. But finding them takes scrolling, and writing a good reply takes a minute you don't have.
Here's the shortcut. Paste your post export. Jeremy AI sorts it by age and sets a status on each post from your rules. Then it drafts the replies. Six parts:
A post table. Four columns: post, replies, hours old and status. One row per post, oldest first.
Three reply drafts. One for each post that needs a reply or is on watch. Each body is 40 words or fewer. It thanks the member and asks one question about the post title.
Two discussion prompts. Each one uses only a topic you listed.
A shout-out draft. For the post you marked as a win. 40 words or fewer.
A not-provided list. Everything the drafts would need that you didn't give, like the moderator name or links.
A check table. Three columns: item, source and status. It confirms the rules were followed.
It's for community managers, course creators and coaches who run a member group. It's drafts only. Jeremy AI doesn't post anything.
What you need before you start
A Jeremy AI account. Nothing to connect to try the sample.
Your post export. The sample is made up: SAMPLE Momentum Community, an online member community for coaches, with five posts (P-01 to P-05). Each row has a title, the number of replies and how many hours old it is.
Your status rules. The sample sets "needs a reply" at 0 replies, "watch" at 1 to 2 replies and 24 or more hours old, and "healthy" at 3 or more replies.
Your topics. The sample lists four: headlines, CRMs, ad spend and booked calls.
A win to celebrate. The sample marks one post as a win.
Honest gaps. The sample says platform name, moderator name, member counts, member names and links are "not provided."
The rules. The sample uses HARD RULES: copy every post id, title and number exactly, never calculate a new number, never give advice or claim a result, never invent a member name, link or feature, never mention giveaways, discounts, prizes or incentives, and never address you by name.
How to set it up
You use the default Jeremy AI worker. There's nothing to build.
Open Jeremy AI. From the dashboard, type in the main box. Check the worker selector says Jeremy AI.
Paste the prompt below. Swap your own posts, status rules and topics in where the sample ones are.
Click Send message.
Wait. The walkthrough speeds this part up. In real time it took a little over a minute.
When it says the file is saved, read the chat summary first.
Open the file card and read the post table.
Scroll to the reply drafts, the prompts, the shout-out, the not-provided list and the check table.
The prompt
This is the exact prompt from the walkthrough. The community, the posts and every number are SAMPLE and made up. Replace them with yours.
Act as my "Community Reply Queue" worker for SAMPLE Momentum Community, an online member community for coaches (all data below is made up). Turn the SAMPLE post export below into a reply queue. Create ONE file named community-reply-queue.md using the file tool (reply with a short summary only) with: 1) a POST TABLE with exactly 4 columns (post, replies, hours old, status), one row per post, sorted from most hours old to fewest, status taken from the status rules below, 2) THREE REPLY DRAFTS, one for each post whose status is "needs a reply" or "watch", each with a body of 40 words or fewer that only thanks the member and asks one question about the stated post title, 3) TWO DISCUSSION PROMPTS, each using only a stated topic, 4) ONE SHOUT-OUT DRAFT of 40 words or fewer for the post marked as a win, 5) a NOT PROVIDED list, 6) a CHECK table with exactly 3 columns (item, source, status). Put the placeholder "[moderator name: not provided]" only as a sign-off on its own last line of each draft, never inside a sentence. Use only the words thanks or congrats as praise. STATUS RULES: needs a reply = 0 replies, watch = 1 to 2 replies and 24 or more hours old, healthy = 3 or more replies. HARD RULES: copy every post id, title and number exactly, never calculate a total, average, percentage or new number; never give advice or claim a result; never invent a member name, link, platform feature or detail not stated; write "not provided"; never mention giveaways, discounts, prizes or incentives; every table has 4 columns or fewer and cells of 12 words or fewer; never address me by name anywhere in the chat reply or the file. Drafts only, nothing is posted.
POST EXPORT (SAMPLE):
- P-01, "Stuck on my offer headline", 0 replies, 30 hours old.
- P-02, "Win: first booked call", 14 replies, 20 hours old.
- P-03, "Which CRM do you use?", 0 replies, 6 hours old.
- P-04, "Ad spend too high this week", 2 replies, 48 hours old.
- P-05, "Weekly check-in thread", 9 replies, 3 hours old.
Topics members raised: headlines, CRMs, ad spend, booked calls.
Not provided: platform name, moderator name, member counts, member names, links.
Prompting tips and how to adapt it
Give the status rules as numbers. The prompt says needs a reply is 0 replies, watch is 1 to 2 replies and 24 or more hours old, and healthy is 3 or more. Jeremy AI applies them. It doesn't guess which posts "feel" ignored.
Keep the replies small. Each draft only thanks the member and asks one question about the post title. That keeps it safe to post. You can add detail yourself.
Put the sign-off on its own line. The prompt says to put "[moderator name: not provided]" only as a sign-off on its own last line. You fill in your name before you post.
Limit the praise words. The prompt allows only "thanks" or "congrats." That keeps the tone steady.
Use IDs and exact titles. The sample uses P-01 to P-05 and says to copy every id and title exactly.
Ban new numbers and advice. The prompt says never calculate a total, average, percentage or new number, and never give advice or claim a result.
Ban incentives. The prompt says never mention giveaways, discounts, prizes or incentives.
Adapt it. Swap in any community: a membership, a course forum or any member group. Change the cutoffs or the topics.
The result to expect
One Markdown file named community-reply-queue.md, titled "Community Reply Queue: SAMPLE Momentum Community." It starts with "Drafts only. Nothing is posted." With the sample data you should see:
A 4-column post table with five rows, oldest first. P-04, "Ad spend too high this week": 2 replies, 48 hours, watch. P-01, "Stuck on my offer headline": 0 replies, 30 hours, needs a reply. P-02, "Win: first booked call": 14 replies, 20 hours, healthy. P-03, "Which CRM do you use?": 0 replies, 6 hours, needs a reply. P-05, "Weekly check-in thread": 9 replies, 3 hours, healthy.
Three reply drafts for P-04, P-01 and P-03. Each thanks the member, asks one question about the post title and ends with "[moderator name: not provided]" on its own line. For example, the P-03 draft thanks them and asks what they need a CRM to do.
Two discussion prompts. One on headlines and one on CRMs. Both use topics you listed.
A shout-out draft for P-02. It says "Congrats" and asks one question about the first booked call.
A not-provided list. Five items: platform name, moderator name, member counts, member names and links.
A 3-column check table. Nine rows, each marked "verified": ids and titles copied exactly, reply and hour numbers copied exactly, statuses applied from your rules, sorted from most to fewest hours, drafts 40 words or fewer, praise limited to thanks and congrats, sign-off on its own last line, no new numbers and the not-provided list included.
The chat reply summarizes the file and says nothing was posted.
These are sample results from a made-up community. Wording changes from run to run. Read every draft before you post anything.
Common mistakes and fixes
A post gets the wrong status. Fix it by hand and say: "Apply the status rules exactly."
A reply gives advice. Delete it. Remind it: "Only thank the member and ask one question about the post title."
A member name or link appears that you never gave. Delete it. Remind it: "Never invent a member name, link or detail."
A giveaway, prize or discount shows up. Delete it. Remind it: "Never mention giveaways, discounts, prizes or incentives."
A new number appears. Delete it. Remind it: "Never calculate a total, average, percentage or new number."
The placeholder sits inside a sentence. Move it to its own last line. Remind it: "Put the moderator placeholder only as a sign-off on its own line."
A question assumes something happened. Check the wording. A question like "What changed with your ad spend this week?" assumes a change. Rewrite it if that doesn't fit the post.
A draft runs long. Remind it: "Each body is 40 words or fewer."
You post a draft with the placeholder still in it. Don't. Fill in your moderator name first.
You post it as is. Don't. Read each draft once, then post it yourself.