Brand mention sweep

Twenty minutes of calls that replace a monitoring subscription for a brand that does not need one yet. Terms in, a dated mention table out, with the threads where a human reply would matter and the talking points for it.

Before you start

• Load the core jell skill (https://jell.run/SKILL.md) if it is not loaded. Confirm access with the free balance tool or jell balance.
• Ask for: the brand and product names (and the misspellings people use), the founder's name and handles, three to five competitors, the category phrase, and the window (default 90 days).
• Inspect social.search, news.search, web.search and reviews.search once and show the prices. Social search is priced per platform: X $0.0007 a post, LinkedIn $0.0035, Reddit $0.0024; the hold covers the limit and the charge covers the posts returned. Quote the sweep: terms x platforms, with Reddit counted at $0.024 a term at a limit of 10.

Steps

1. Terms

Brand, product, founder, each competitor, the category phrase. For each, the exact string and the common variants. Save the list; the sweep is worth repeating monthly with the same terms.

2. Social

jell run -c social.search -i '{"platform":"x","query":"Jell","limit":25}' --max-cost 0.02 --wait 60
jell run -c social.search -i '{"platform":"linkedin","query":"Jell","limit":10}' --max-cost 0.05 --wait 120
jell run -c social.search -i '{"platform":"reddit","query":"Jell","limit":10}' --max-cost 0.15 --wait 120

Run each term on x, reddit and linkedin first; tiktok, youtube and instagram when the brand has a consumer side. Reddit and LinkedIn runs often pass 60 seconds: poll a still-running receipt with runs get -r <run_id> --wait 60 -o file. Quotes in the query are not honoured, so search the bare term and match it in the text yourself. For a generic competitor name (Clay), search the domain or the name with the category word, or the results are pottery. No platform takes a date range: rows carry created_at as ISO 8601 UTC, filter on it after the call. Keep: URL, platform, date, author (a display name) and author_handle when the platform returns one, the sentence with the mention, engagement counts (null on Reddit rows). An empty search comes back as no_match, released, not billed.

3. News and web

jell run -c news.search -i '{"query":"Jell","limit":50}' --max-cost 0.05 --wait 30
jell run -c web.search -i '{"query":"\"Jell\" -site:jell.run","limit":50}' --max-cost 0.05 --wait 30

Web search catches the blog posts, comparison pages, directories and forum threads that social search misses. Search operators are not honoured by the results (-site: and quotes are ignored, and a site: operator is billed at five times the price), so search the bare term and drop the brand's own domains afterwards. Web rows carry no date; about half the snippets start with one (a date such as "Sep 3, 2026" followed by a dash), parse it when present. A news search with nothing behind it comes back as no_match, released, not billed; a failed run is a provider fault, retry it once. Ten X posts cost under a cent when this was written.

4. Reviews

jell run -c reviews.search -i '{"place":"<Google Maps URL or place ID (ChIJ...), or a Yelp business URL>","limit":50}' --max-cost 0.20 --wait 120

For brands with a physical presence or a Google Business Profile. Skip otherwise. place is a Google Maps URL, a place ID or a Yelp business URL, and the platform is inferred from it; a business name fails at the provider. The provider occasionally fails a run outright; a failed run is released, so retry it once before reporting a gap.

5. Classify

First split own from third-party: posts and pages by the brand, its founder or its own domains are own and are counted separately, so the totals say how many other people talked. For each third-party mention: sentiment (positive, neutral, negative, question), type (recommendation, complaint, comparison, question, news), and whether a human reply would matter (a question with no answer, a wrong claim, a comparison where the brand is missing, a complaint). Filter the window to the last 90 days by the mention's own date, not the crawl date.

6. Talking points

For every thread flagged as reply-worthy: what the thread says, what a truthful reply would add, and the disclosure line ("I work at ..."). A human posts, or does not.

Rules

• Never post, reply, vote or engage anywhere. Draft for a human.
• Every mention carries its URL and date. Sentiment is your reading and is labelled as such; quote the sentence so the user can disagree.
• Do not compile personal information beyond the public mention itself. Handles and names stay as they appear in the public post; nothing is enriched from here.
• Every count is from a run and carries the run ID.

Output

mentions-<date>.csv (run ID, URL, platform, date, author, handle, term matched, own or third-party, sentiment, type, reply-worthy, quote) and mentions-report-<date>.md: totals per term and platform with own posts separated, sentiment split, the competitor comparison, the reply-worthy threads with talking points, and the total charged as the sum of every run's billing.charged (not a wallet difference: the balance is shared by every session on the key). Date the file names, so next month's sweep diffs against them.