vritax

Do AI assistants actually pick your app?

Vritax measures how ChatGPT, Claude and other assistants surface and select apps, connectors and tools — and shows builders what to improve.

Candidates surfaced per prompt · SAMPLEYour app
SAMPLE · one lit candidate out of 24. Selection share, not a prediction.
How it's measured

Observed behaviour, not self-reported claims.

01
0 claims

Read from real assistant behaviour

Every result comes from watching what an assistant surfaces and selects for a prompt. Nothing is taken from a vendor's own description of its integration.

02
Repeated

Repeated runs, bias controlled

Prompts are repeated across phrasings, orderings and sessions. Position and naming effects are rotated out so a single lucky run never becomes a finding.

03
Directional

Reports state direction, not certainty

Each report says which way a signal moves and how thin the evidence is. Assistants change weekly; a number without a date is worthless.

Research log
Log 001 · 2026-09-10
One description change, and a published ChatGPT app lost the first call in 9 of 10 chatsTwo flight-search tools live in the same ChatGPT conversation, one fixed request, 40 fresh chats. Changing nothing but the test connector's tool description moved the host's first call from 0 of 10 to 9 of 10, back to 0 when I put the old text back, and to 9 again with a description a real publisher could ship.Read the report →

Join the research list.

One report when there is something measured to report. No product news, no cadence promise.