Forty-six percent of which PM jobs?

Aakash Gupta and Dr Nancy Li's August 27 article reports 35,725 LinkedIn product-management listings and 16,420 results for an AI PM search. Dividing those counts produces the headline figure: roughly 46%. The same article goes further, examining 113 postings for the AI experience employers actually request. Read the analysis.

The division is easy. Establishing that its numerator and denominator describe the same population is the real work. My correction to the tempting critique is that overlap alone does not make a percentage wrong: a subset is supposed to overlap its parent set.

The unanswered question is whether those search results form a clean subset, measured under consistent filters, deduplicated in the same way and classified by the same definition. Without that, I would treat 46% as a search observation, not a measured share of the PM labour market.

The sample answers a narrower question

The authors' closer reading is more actionable than the headline. Within their AI-search sample, generic PM titles and AI-labelled titles both frequently demanded prior AI or ML experience. They report 66% and 70%, respectively. Those figures concern the selected sample; they do not establish that two-thirds of all generic PM jobs require AI experience. Sample and method.

That is still useful. A title can hide an experience requirement, so searching only for an AI prefix may miss relevant work. Reading responsibilities and requirements beats counting a label.

But the reverse matters too. A broad search can return roles because AI appears somewhere in the description, even when it is incidental to the job. Search relevance and occupational classification are different systems. Neither becomes a substitute for the other because the result count looks precise.

Build the denominator before the headline

Here is how I would use this material in a hiring review. This is a proposed exercise, not a study I have performed. Start with one defined market: a geography, seniority band, time window and set of employers that actually compete for the candidate.

Collect unique open roles under a consistent inclusion rule, then classify what each role requires. Distinguish building an AI product, using AI in ordinary PM work and merely working at a company that mentions AI. Keep ambiguous cases visible instead of silently forcing them into the exciting category.

As an illustration, if 30 of 100 unique roles explicitly require experience shipping an AI feature, the relevant figure is 30% of that defined sample. It can inform a decision without pretending to measure every PM opening worldwide. Publish the definition beside the result so someone else can disagree with the boundary rather than guess what it was.

For a candidate, I would turn those requirements into an evidence plan. A portfolio project should demonstrate the recurring decisions employers name: selecting an evaluation, handling failure, choosing a release boundary or explaining a trade-off. Putting AI in the project title does not show any of those capabilities.

For a hiring manager, the equivalent exercise is to rewrite one vacancy as decisions the person must make in the first three months. Then ask which of those genuinely require prior AI experience and which can be learned with domain support. That produces a more defensible requirement than copying a fashionable title.

The article is a useful starting point for those conversations. Its selected postings can reveal expectations worth investigating. The market-share claim needs a population definition before it can carry a broader conclusion.

Use job searches to find requirements; use a defined population to measure their prevalence.