Buying signals that convert, and the ones that waste your month

The difference that matters

There are two kinds of phrase, and they look identical on a whiteboard.

A topic phrase catches people interested in a subject. "AI for sales" catches consultants, students, competitors, and people who like posting about AI.

A problem phrase catches people currently suffering something. "our consultants hate prospecting" catches somebody who just lost an argument about it internally.

The second converts. The first fills a list.

Why keywords fail

"hiring" is a keyword. It catches job adverts, congratulations posts, recruiters advertising, and people announcing they got a job. Essentially none of them are buying.

"we cannot keep up with enquiries" is a phrase. It is longer, it matches less, and almost everyone it matches has the problem right now.

As a rule of thumb, three words is where a fragment starts carrying intent. Anything shorter matches the whole internet and bills you for the privilege.

Write them in their words, not yours

The most common failure is writing signals in vendor language.

You sell "pipeline optimisation". Nobody has ever typed that into a post. They typed "our BD is inconsistent" or "pipeline dries up whenever we get busy delivering".

If you would not say it out loud on a call, it is not a signal. It is positioning, and positioning does not match text.

Measure them or they are guesses

Here is the part almost nobody does. Every phrase you add is a standing cost: it opens searches, those searches run, and running costs money whether or not anyone good turns up.

So count, per phrase:

• how many searches it opened
• how many people those found
• how many of those cleared your bar

That last number is the only one that settles an argument. A phrase that has opened three searches and produced nobody is costing you money on every cycle, and no amount of re-reading the phrase will tell you that. It reads fine. It just does not work.

A workable starting set

Five phrases, not twenty. Taken from things customers actually said to you, ideally from call recordings or support tickets rather than memory.

Run them for two weeks. Kill anything that produced nobody. Write two more from what the survivors turned up, because the people you caught will have phrased the problem in ways you did not think of.

That loop, run monthly, beats any list you could write in one sitting.