Invoke Share answers a competitive question: when an AI agent could use your product or a competitor for the same request, how often does it complete the task through yours?
How it is measured
Only contested intents count: requests for which your integration and at least one competitor are each independently eligible, decided by a neutral category process rather than by your own tool list. The host’s built-in abilities and the “no tool” path stay available in every run, but they are reported separately as the Agent Resolution Split, never mixed into the vendor-against-vendor share. Where a competitor’s results cannot be checked, we report Selection Share instead and say so.
| Integration | Invoke Share | 90% interval |
|---|---|---|
| Integration A | 36% | 31 to 41% |
| Integration B | 33% | 28 to 38% |
| Integration C | 20% | 16 to 24% |
| Integration D | 11% | 8 to 14% |
Reported tie A and B: intervals overlap - reported as tied, so neither is ranked above the other.
What it is not
- Not real-world market share. It is scoped to a category, a host, a runner model, a language, a prompt set and a time window, and states its weights.
- Not a ranking on its own. Every share carries a 90% interval; overlapping intervals are reported as tied.
The full method is on the methodology page. Real category tables appear in the index only once they pass the eligibility and evidence gates.