Matched historical comparison
Matched August and September estimates, changes, and bootstrap intervals.
Public dataset · CRM recommendations
Inspect the evidence behind the CRM benchmark.
Six public CSV files contain the released product metrics, segment results, matched historical changes, wave stability, source counts, and exact prompts from the three-wave September replication.
release_openai_us_crm_replication_2026_09Published September 15, 2026Public download
Research and publication by Benchmark Bureau · Reviewed September 15, 2026 · Report a correction
Download the released files
Download Matched historical comparison CSV
Product metrics
Overall recommendation and first-choice estimates, intervals, support, and rank uncertainty.
Exact prompts
The exact 200-prompt manifest, hashes, and designed buyer dimensions.
Segment metrics
Product estimates across company scale, buying priority, intent, and prompt style.
Source domains
Overall and segmented consulted and inline-cited source-domain counts.
Wave stability
Pairwise recommendation and first-choice rank stability across three waves.
Data dictionary
| Field | Meaning |
|---|---|
release_id | The immutable release represented by the row. |
entity_id | The normalized product identifier. |
recommendation_estimate | Eligible answers actively recommending the product, divided by eligible answers. |
first_choice_estimate | Explicit leader credit, with stated ties split, divided by determinate eligible answers. |
*_ci_low / *_ci_high | The 95% prompt-bootstrap interval for the named estimate. |
segment_type / segment_value | The designed buyer context represented by a segment row. |
inline_citation_count | Visible inline citations attributed to a normalized source domain. |
Each CSV includes additional self-describing headers. Definitions, exclusions, and denominators are controlled by the published methodology.
Verify the downloaded bytes
SHA-256 checksums bind each file to the published release. A changed byte produces a different checksum.
Matched historical comparison
09fa8831b6496bb4afc6ed7f13a78a8f999082f8ccd9a0b898286f806dd98857Product metrics
8a19a642cc3d1c57768d3c7af12a634880c51f945285ad53e457d02ffe959645Exact prompts
de163c3e8ed6b68f911e6f5c83b6f1cb3930adf82c964afeef7cef7c88a00d42Segment metrics
efee93491147868c8327de1f3c5fac165306ff5fd0d370ebd8675d8b36b69673Source domains
2e163862eab02ca9e21e183bb875b57d847df89966d5b680ec4fd62332880f80Wave stability
963a4d8d7efb7bf2c64dbd70026759e7137644815450d188ce18621cd0b1deb4Suggested citation
Benchmark Bureau. “CRM AI Recommendation Dataset: September 2026 Three-Wave Replication.” Release release_openai_us_crm_replication_2026_09, published September 15, 2026. /datasets/crm-ai-recommendations
Research and publication by Benchmark Bureau. Questions go to the research contact; factual issues follow the corrections policy.
Scope and reuse
Factual citations and links are encouraged under the current site terms. Any reuse should name Benchmark Bureau, identify the release, link to this dataset record, and preserve the observation-window and single-provider limitations. No broader data license is asserted here.
These aggregate and normalized research files do not establish product quality, customer outcomes, sales, market share, causal source influence, consumer ChatGPT behavior, or broader AI-provider consensus.