We judged 1,635 public posts. 70 were real buying signals

One run, counted from start to end

On 19 October 2026, a software services firm we work with ran one lead search. These are that run's stored counts, shared as totals. The firm stays unnamed.

226 searches found 2,090 posts. 455 were posts found again at the same address, so they were not judged twice. The 1,635 new ones were read and judged. Separately, 179 judged posts copied other posts' text, and were dropped for it.

What judged means here

Every post answered the same fixed questions. Is the author a person? Buying or selling? Our kind of work? Able to pay? How urgent? In the right region?

Unsure answers stay unknown, and unknowns count neither for nor against a company. Each answer keeps the post's own words as evidence. Numbered, locked rules then turn the answers into a grade.

What the 1,565 drops were

One dropped post in three came from a seller offering the same work. The words match your offer exactly, because the author sells what you sell. Reading that as demand is the classic mistake.

The named reasons cover 1,564 of the 1,565 drops. One recorded no reason, and we say so instead of hiding it.

Judged and kept, by source

The loudest source was the worst, so the budget should follow what survives. The plan post adds the search counts and spend.

The broad social feed was not useless. Its one keeper was a direct request worth having. It was just costly attention.

What a real buying signal looked like

The 70 keepers were events, not keywords. 27 were hiring bursts, 23 new funding rounds and 9 new sales or growth leaders. A funding keeper is a raise announced in any source's post, sorted by its kind. That is why the source table has no funding row. The rest: 6 direct requests, 2 requests for proposals, a tender, a product launch and a job ad.

Direct requests are gold and rare. A test that waits only for public requests will starve.

The most common keeper: several dated engineering roles from one company in a short window, none through recruiters. What goes out is the roles, their dates and links, and one honest question. Never a line like 'they are struggling with capacity'.

Right drops, wrong reasons

A day earlier, a judging pass marked 234 of 276 posts as freelancer requests. Most were news or opinions. The lead-search post tells that story and the rule it left.

The one drop with no reason above is the tail of that lesson. Every drop must carry its reason, and one slipped through.

Numbers we do not have yet

How often people agree with the machine: people have recorded only 5 keep-or-drop decisions on this run's grades. That is too few to mean anything.

Replies: the kept leads are new, and replies need real weeks. Both get their own posts when the counts can stand up.

Copy the fit bar, not the numbers

Write the drop reasons before the first search. Make the system record one for every dropped lead.

Review the keepers, not the finds. A person can read 70 graded leads in an afternoon. The cost ladder prices each stage of this run.

Your rates will differ with your market and your bar. The shape should not: most finds dropped for named reasons, and every keeper carrying a dated event.