Connexion

When a buyer asks an AI which software to buy, somebody gets named. We measure who.

A shortlist now comes back from an assistant instead of a page of search results, and the vendors on it have never been told where they sit. So we ask the buying question the way a real prospect asks it, across 44 phrasings, on 2 engines, and count every product named across all 88 answers. The method and the raw answers are published in full — you can check the number rather than take ours.

Engines ChatGPT · Perplexity Phrasings per category 44 Answers read per category 88 Categories live 2

The live leaderboards

Free to read, raw data included. No vendor paid to be listed and no vendor can pay to move.

Applicant tracking systems

ats.connexion.me · 73 vendors named

  1. Workable — 76.1% of answers
  2. Breezy HR — 75.0%
  3. JazzHR — 65.9%
See all 73 →

Help desk & support software

helpdesk.connexion.me · 50 vendors named

  1. Freshdesk — 86.4% of answers
  2. Help Scout — 83.0%
  3. Zendesk — 69.3%
See all 50 →

How the measurement works

Every claim on either board is reproducible from files published alongside it.

The questions come first, the vendors come last

44 distinct phrasings of one purchase decision are written before anything is measured, and no vendor name appears in any of them. Each engine is called through its public API with a near-empty system prompt — “answer the user's question directly and concisely” — and is never asked to produce a list or rank anything. Steering it toward a list would manufacture the thing being measured. The vendor list is then discovered: a product is on the board if and only if an engine named it.

Stability is checked before publication, not after

Each category is measured twice — same questions, same engines, independent calls. The ATS top ten came back 10/10 identical; help desk, 8/10. A board whose head reshuffles between runs is noise and would not be worth publishing, so the check runs first and the page ships only if it passes.

The raw answers are published

All 88 answers per category, verbatim, one JSON object per line, along with the extracted product names per answer and a machine-readable ranking. Linked at the bottom of each board. The measurement is automated end to end, which is why anyone can recompute it.

What this is not

It measures what two answer engines said, on one date, in response to these questions, through their APIs. It is not a quality ranking, not a review, and not advice on what to buy. A product can be excellent and rank low — that gap is the entire point.

What is free, and what is for sale

Both leaderboards are free and stay free — the ranking, the method, every raw answer. Nothing here goes behind a login later.

One thing is for sale: the diagnosis behind a single vendor's row — $199, one-off. When the engines did not name you, who did they name, on which phrasings, and which pages were they citing to do it? What is in the $199 report →

Getting a category measured

Two categories are live and more are being run. If you sell into a crowded B2B category and want to know where the answer engines put you, say which category and we will tell you when it is measured — or tell you it already is.