Do Your Prompts Produce Accurate Lists?

Presented by Landbase

The 2026 GTMBench scores Landbase, Clay, Apollo and ZoomInfo on one question: how well can the platform understand natural-language prompts and provide an accurate list? This report breaks it down by prompt family, by prompt, and by vendor.

26 Prompts Tested.
4 Key Findings Discovered.

  • 3 of 4

    Agents lead three of four prompt families.

    Filters win only where the prompt names attributes the vendor already stores.

  • 18 of 26

    Landbase returns the most precise list.

    76.7% of its rows meet every criterion, against 47.9% for the next vendor.

  • 3 prompts

    Two of the four vendors can’t express three prompts.

    Clay and ZoomInfo declined the same three. Landbase scored 96.1% or better on all of them.

  • 134 seconds

    Landbase is the slowest of the four systems.

    It checks every company against every condition before it returns a row.

Landbase-GTMBench_thumb_v2

A filter can only match what’s stored.

Many fields that determine the ICP criteria aren't stored in clean fields. An agent reads the underlying records and works the criterion out for itself. Natural-language prompting uncovers the real gap.

What’s in the full report?

26

Prompts tested. Each scored per vendor, with the exact wording used.

4

Prompt families, and which kind of system wins each.

100

Rows checked per segment against every stated criterion.

4

Vendors: Landbase, Clay, Apollo and ZoomInfo, run the same way.

See how your vendor scored on the prompts you’d run.

The scores, the misses, and all 26 prompts to test on your own stack.

Download the report