About & editorial policy

AI Tools for PMs is a curated shortlist of AI tools for product managers. The tool landscape is a firehose of hype — you don't need another directory of 10,000 tools, you need a trusted answer to "which tool actually works for this PM task, and how do I use it well?"

How we choose tools

Every entry is hand-written and verified. No auto-scraped listings. Before a tool is published, a human confirms it exists, that our pricing is current, and that the 10-minute recipe actually runs.
Depth over breadth. We keep the catalog small (roughly 50–150 tools) and organized by the job you're trying to do, not by tool type.
Honest verdicts. Every verdict names at least one real weakness. A tool earns its place or it gets cut.
Freshness is a feature. Each entry carries a "last verified" date. We re-check entries on a schedule and cull tools that die or go stale.
Cuts are public. When we remove a tool — it died, it priced itself out of reach, or it simply wasn't good enough to recommend — we say so in that category's intro, with the reason. A shortlist that only ever grows isn't curated.

Written with AI — and what we do about the obvious problem

This site is researched and drafted with AI assistance, specifically with Claude. That creates a bias risk you should know about, because the first draft of this site had it: 45% of the prompts were badged "tuned for Claude," 38% of the tool pairings pointed at Claude, and Claude held an "our pick" badge that an equally-scored ChatGPT did not. None of that was argued for anywhere. It was just what the author reached for.

So we measure it instead of trusting ourselves. Every build runs a neutrality check that caps how much of the site any one vendor can occupy — no vendor may be the named tool on more than 30% of prompts or the target of more than 40% of tool pairings, and playbook concentration is reported on every run. Prompts that are plain text in, text out now say so: they run in ChatGPT, Claude or Gemini, and we don't pretend otherwise. Where we do name one model, the page says which model we actually field-tested with, so you can discount it accordingly.

The residual we haven't fixed yet, stated plainly: Anthropic products still appear in most playbook chains, because those chains record what was genuinely tested and we won't claim a test we didn't run. Rebalancing them means re-running the playbooks against another model. That's the honest fix and it's on the list.

How we stay current

Freshness isn't a promise, it's a procedure. Every entry carries a "last verified" date; a monthly pass re-checks pricing against each tool's live site, playbooks are re-run quarterly so every prompt still works as written, and tools that die or go stale get culled instead of padding the count. The method is written down as a mechanical checklist — verification prompts, kill criteria, field-test rules — so it runs the same way every month regardless of who (or what) runs it.

The weekly loop, and what gets retired

On top of that, one cycle runs every week and publishes as The Cut: sweep what's new, kill most of it against the adoptability bar, field-test what survives, and re-check what we already published. The whole cycle is a written work order, so the standard doesn't drift with whoever runs it.

Anything new enters as experimental: it passed exactly one field test, it says so on the page, and it carries a re-test deadline. On that date it either graduates — we re-ran it, it still works — or it gets retired. There is no third option; a missed deadline breaks our build until someone decides.

Retiring is public and permanent. The page stays up with a post-mortem on top naming what we claimed, what changed, and what we missed when we admitted it — you'll find them all in the retirement record. We don't quietly delete advice that stopped being true, because advice that can't be retired can't be trusted.

The other half of that discipline is the boring one: if nothing clears the bar, we publish an empty issue that says so. No filler, no "5 tools you might like." The running kill rate on the weekly page is there so you can check whether the bar is real.

Our scoring

Verdicts are a whole-number 1–5 editorial call: 1 Skip · 2 Niche · 3 Solid · 4 Great · 5 Essential. It reflects how useful the tool is for the PM job it's listed under — not a generic quality score.

Affiliate disclosure

Our funding model allows affiliate links, and at the moment we use none: zero of the 45 tools in this catalogue carry an affiliate link. That count is read straight from the catalogue every time the site is built, so this page cannot quietly fall out of date. When the number changes, this sentence changes with it.

Affiliate income never influences our verdicts or rankings. Tools are placed and scored on merit alone; a tool with no affiliate program is ranked exactly the same way as one that pays. Where an affiliate link is used, the outbound link is marked rel="sponsored" and disclosed on the page carrying it, in line with the FTC's endorsement guidelines.

Corrections

Pricing changed? A tool shut down? A recipe stopped working? Tell us and we'll fix it — accuracy is the entire point of this site.