Every product with at least 10 labels here, on both axes. The 30% line names a quadrant, not the verdict above: that one needs more than 40%.
“**Don't buy Salesforce** unless you're an enterprise with admin expertise and budget for the true TCO.” DeepSeek V4 Flash, negative prompt
“CRMs to **Approach with High Caution**... **HubSpot**: Tops lists with 300+ negative G2 reviews (capped), Risk Score 100/100” Grok 4.1 Fast, negative prompt
“**Zoho CRM**: 300+ negative G2 reviews... Affordable but frustrating for growing teams.” Grok 4.1 Fast, negative prompt
“Avoid if you have marketing + sales + customer success teams... **Pipedrive** (7/100)” MiniMax M2.5, negative prompt
17 of 72 answers in this category came back with a source list, from 4 of 6 models: citations where the model returns them, or the search results it consulted. 6 of those lists are Google grounding redirects that name no site and are left out of the counts. 138 links across 86 sites, every framing counted. Ranked by the number of answers carrying the site or page. The other 2 models searched through a gateway that returned no source list on this edition.
The ten pages named in the most answers, by full address. A page here is one the models returned with a recommendation, not one the index endorses.
| Prompt | Model | Version string | Time (UTC) | Searched | Sources | Latency |
|---|
Flips between the edition run and its calibration repeat. Six prompts per model is a small sample; the index-wide floor is the number to trust.
Every judgment call made between the raw labels and the numbers above, listed so it is visible and reversible.