Let two bots keep your catalog fresh: one scouts, one checks
By Šarūnas Vaitkus · about a 6-minute read
If you run a directory, a tool list or a content library, it goes stale fast. Two bots can keep it fresh every month without publishing anything nobody checked. A scout finds and drafts new entries with sources. A reviewer opens every source again, fixes what's wrong and adds only what passes. You send the prompts below, in order, and they ask you before anything sensitive goes in.
A catalog with a clear entry format, such as a database table, a spreadsheet or a content list.
Two bots, or one bot running two separate routines.
About 15 minutes to agree the rules and check the first run.
How it works
You send a prompt, the bot does the work, and you approve what matters.
Highlighted [brackets] are yours to replace. Copy keeps them in the prompt.
1
Step 1. Define an entry and how it's scored
Prompt to send
Look at my catalog in [where it lives]. Don't change anything. Describe one entry: every field, what values each field allows, and which fields visitors actually see. If there's a score, write down the exact formula. Then list which fields need a source to be trusted, for example prices and claims. Save it as a short rules file.
Why you're sending it
Both bots need the same idea of a good entry, or the reviewer spends its time arguing with the scout.
What you'll get
A one-page rules file with fields, allowed values, the scoring formula and what needs a source.
2
Step 2. Let the scout find and draft
Prompt to send
Once a month, search for new [tools, courses or resources] that fit my catalog. Keep a state file of everything already in the catalog and everything you've already looked at, and skip those. For each new find, draft a full entry using the rules file, with a source link for every claim. Save the drafts as one file and hand them to the reviewer. If there's nothing new, send one line saying so.
Why you're sending it
The scout does the slow part, finding and drafting, and the state file stops it re-finding the same things.
What you'll get
A monthly file of draft entries with sources, or one line when nothing's new.
3
Step 3. Mark what it couldn't check
Prompt to send
In every draft, mark any field you couldn't confirm on the vendor's own page as UNVERIFIED. Never guess a price, a feature or a limit. Save a copy of each page you used as evidence.
Why you're sending it
A clear "don't know" is fixable. A confident guess ends up on your site.
What you'll get
Drafts where every gap is visible, plus saved evidence pages.
4
Step 4. Let the reviewer check every source
Prompt to send
For each draft from the scout, open every source again yourself. Correct anything the page doesn't say, remove claims you can't find, and recompute the score with the formula in the rules file. Check the catalog for duplicates by name, address and close matches, including hidden entries. Write down every change you made and why.
Why you're sending it
A second, independent check catches wrong prices and invented claims before your visitors do.
What you'll get
A review file listing each draft, what was fixed and why.
5
Step 5. Sort into CLEAR and FLAG
Prompt to send
Label each reviewed entry CLEAR or FLAG with a one-line reason. CLEAR means every claim is sourced, there's no duplicate and the change is small. FLAG means anything unverified, a brand or legal question, a very new product with no track record, or a change to how scoring works. Add CLEAR entries yourself. Hold FLAG entries for me in one list.
Why you're sending it
Most entries are routine and can go straight in. A few need a human, and they should all arrive in one place.
What you'll get
CLEAR entries added on their own, and one short list of FLAG entries for you.
6
Step 6. Preview the exact change first
Prompt to send
Before you write to the live catalog, save the exact changes you're about to make and a snapshot of the current values so they can be undone. Make all the changes in one step that either fully succeeds or changes nothing. Then check the counts and a few entries on the live site.
Why you're sending it
A saved snapshot means any bad change can be undone in minutes.
What you'll get
A change file, a rollback file and a check that the live catalog looks right.
7
Step 7. Send one quiet summary
Prompt to send
After each run, send me one short summary: how many entries were added, how many were fixed and the FLAG list with reasons. If nothing changed, send nothing.
Why you're sending it
You want to know what happened without reading every draft.
What you'll get
One message per run, only when there's something to say.
8
Step 8. Turn rejections into rules
Prompt to send
Every time you fix or reject a scout draft for the same reason twice, write it up as a rule and add it to the rules file, so the scout stops making that mistake.
Why you're sending it
The reviewer's fixes are free training for the scout.
What you'll get
A rules file that grows from real mistakes, and drafts that get better each month.
All in one (optional)
Prefer one prompt? Send this instead.
Prompt to send
Set up a scout and reviewer pair for my catalog in [where it lives], in this order: a rules file for entries and scoring, a monthly scout run with a state file and sources for every claim, UNVERIFIED marks for anything unchecked, an independent reviewer check with a dedupe, CLEAR or FLAG labels, a saved snapshot before any write, one quiet summary per run, and new rules from repeated fixes. Show me the rules file before the first run. Ask me before adding any FLAG entry, deleting anything, spending money or sending anything outside.
What your bot will always ask you first
Adding any FLAG entry.
Changing how the score is calculated.
Deleting or hiding entries.
Paying for data or a tool.
Contacting a vendor or sending anything externally.
It bundles its questions, so you're not pinged all day.