Demand assisted buying-signals

Who is showing buying intent this week

No receipt. This lab has not run this loop. This lab sells nothing, so it has no ICP to match signals against and no sales team to hand them to. Running it here would mean inventing a buyer. Everything in the catalog has run here and published what it found. This has not, so judge it on the prompts rather than on our word.

The file

take it and run it.

The full loop as one markdown file: what it needs, the goal prompt, the verifier prompt, and the stopping condition. MIT licensed, no account.

curl -fsSL https://vibemarketers.ai/templates/buying-signals.md -o .claude/loops/buying-signals.md

Most buying intent is public before anyone fills in a form. People ask for tool recommendations in communities, complain about a competitor in reviews, post a job that only makes sense if they are about to buy, or announce the change that creates the need. This loop reads those sources on a schedule and writes down what it finds, with the link.

Use it when you want sales and content to work from the same evidence of what the market is asking for this week, instead of from a vendor’s intent score nobody can check.

You need an ICP file, a list of public sources, and a recency window.

You get a signal log. Each row is one signal: what was said or done, who did it where that is public, the URL, the date, the signal type, and a strength rating. The log is split in two: demand themes, which feed content, and account signals, which feed sales.

The gate is a person. It produces a log for someone to act on. It never contacts anyone.

Signal types, strongest first

Rate each signal by what it proves, not by how exciting it sounds:

  • Asking to buy. A public post asking for a tool, a vendor, or a recommendation in your category.
  • Leaving a competitor. A review, thread, or post describing why they are switching away from something you replace.
  • Hiring for the problem. A job post that names your category, a competitor, or the work your product does.
  • A trigger event. Funding, a launch, a new leader in the buying role, an expansion that creates the need.
  • Talking about the problem. A question or complaint about the pain, with no mention of buying anything.

The last type is the weakest account signal and the best content signal. It is demand you can write for, not a lead you can call.

Goal prompt

Read <icp-file> and <sources-file>. For each source, collect items published
within <window> that match the ICP's problem, category, or competitors.

For each match, record: a one-line summary, the verbatim phrase that makes it
a signal, the URL, the publish date, the signal type (asking to buy, leaving a
competitor, hiring for the problem, trigger event, talking about the problem),
and a strength rating with one sentence of reasoning.

Split the log into two tables. Demand themes: group problem talk and questions
by theme, with every URL listed under its theme. Account signals: one row per
named company or public profile, only where the name appears in the source.

Write to <output-file> with the collection date. Only read pages that are
public without logging in. Never guess an email address, a phone number, or a
person's name that the source does not show. Never contact anyone.

Verifier prompt

Open the output file in a fresh session and load every URL in it.

FAIL if any signal has no URL, or its URL does not load.
FAIL if the verbatim phrase does not appear on the page at that URL.
FAIL if any signal's publish date falls outside <window>.
FAIL if any row contains an email address, phone number, or person's name that
does not appear on its source page.
FAIL if any account signal names a company its source does not name.
FAIL if any source required a login to read.

Otherwise PASS. On FAIL, print the row, the URL, and the reason. Drop rows that
fail the URL or date checks and report how many were dropped.

Stopping condition

Every source was scanned or has a recorded error, every remaining signal has a URL that loads, a verbatim phrase found on that page, and a date inside the window, and the verifier printed PASS.

What it may and may not do

May: read public pages, write the signal log, group themes, rate strength.

May not: log in to anything, scrape behind a paywall or a login, guess contact details, enrich a person from a data broker, message or email anyone, or add an account to a CRM.

What this will not tell you

A signal is evidence that someone said or did something on a date. It is not proof they will buy, and it is not proof they will buy from you. Treat account signals as a reason to look, not a reason to send.

Public sources also skew toward people who post. The quiet buyer who reads everything and asks nothing will never show up here, which is why the demand themes matter more than the account list over time.

This becomes a published loop the day it runs here and passes its checker. Until then it stays on the templates shelf.