Which AI sales tools disclose which AI models they run on?
Nearly every product in this market is marketed as AI powered. The GTM Tech Index grades all 299 vendors on whether that claim can be checked: what models sit underneath, whether AI generated outreach discloses itself, and whether the scoring logic can be explained. 47 document it at a checkable standard. 252 do not say what the AI is.
Evergrowth A, heddl A, HubSpot Sales Hub A, Leadspicker A, Pipedrive A, Revenue.io A, Säljmotorn A, SAP Sales Cloud A, Scratchpad A, WaLead A hold the only A grades on this axis across the index. At B, with meaningful disclosure and one real gap: Accent Technologies B, Adobe Marketo Engage B, Agentforce Sales B, Aviso B, Bigin by Zoho CRM B, Bombora B, Cadivra B, CallSine B, Clay B, Clevenio B, Clodura.AI B, Cognlay B, Coldreach B, Demandbase B and 23 more. Every name links to the grade record and its sources.
What 299 vendors actually document
Models and providers named, AI generated content disclosure addressed, and scoring or routing logic explained at a level a buyer can interrogate.
Meaningful disclosure of the model stack or the disclosure posture, with one real gap, commonly silence on whether AI authored outreach identifies itself.
The product is described as AI powered with the stack, the disclosure behavior, and the scoring logic all unstated.
No public statement of what models are used, how outputs are produced, or whether recipients are told they are talking to software.
Read the shape before the names. The C band is the market: 195 vendors whose product is described as AI powered while the stack, the disclosure behavior and the scoring logic all go unstated. That is not a documentation oversight at this scale, it is a decision, and it is the exact information asymmetry a buyer inherits at contract signature. The model provider is a subprocessor. Not naming it does not remove it.
The AI heavy lanes, one by one
Lane membership here is primary or secondary, the same rule the per lane answer pages use, because the buying question follows the role a product plays rather than its primary shelf.
AI SDR & Outbound Agents
18 of 71 documentedEvergrowth A, Leadspicker A, WaLead A, Agentforce Sales B, Cadivra B, CallSine B, Clay B, Clodura.AI B, Cognlay B, Coldreach B, Dialpad B, Flocurve B, Instantly B, LeadNitro B, LinkSprig B, Oliv AI B, Qualified B, Reply.io B. The other 53 sit at C or below, with 15 publishing nothing on the axis at all.
Conversation Intelligence
10 of 52 documentedRevenue.io A, Scratchpad A, Aviso B, Demodesk B, Dialpad B, Gong B, Jiminny B, Mindtickle B, Oliv AI B, Seismic Enablement Cloud B. The other 42 sit at C or below, with 3 publishing nothing on the axis at all.
Revenue Intelligence & Forecasting
12 of 51 documentedHubSpot Sales Hub A, Revenue.io A, SAP Sales Cloud A, Scratchpad A, Adobe Marketo Engage B, Agentforce Sales B, Aviso B, Demodesk B, Fullcast B, Gong B, Jiminny B, Oliv AI B. The other 39 sit at C or below, with 2 publishing nothing on the axis at all.
Dialers & Voice
4 of 59 documentedRevenue.io A, Dialpad B, Freshsales B, Salesvue B. The other 55 sit at C or below, with 15 publishing nothing on the axis at all.
The questions buyers actually ask
Which AI sales tools disclose which AI models they run on?
Across the 299 vendors in the GTM Tech Index, 47 document their AI at a checkable standard on the AI Disclosure and Model Transparency axis, and 10 do it in full: Evergrowth, heddl, HubSpot Sales Hub, Leadspicker, Pipedrive, Revenue.io, Säljmotorn, SAP Sales Cloud, Scratchpad, WaLead. An A on this axis means the models and providers are named, disclosure of AI generated content is addressed, and the scoring or routing logic is explained at a level a buyer can interrogate. Another 37 vendors earn a B for meaningful disclosure with one real gap. The finding worth sitting with is the other side: 195 vendors, 65 percent of the index, are described as AI powered with the stack, the disclosure behavior, and the scoring logic all unstated, and 57 publish nothing at all. In the category most sold on AI, AI SDR agents, 18 of 71 vendors document what the AI actually is.
Why does it matter which models a sales tool runs on?
Because the model provider is a subprocessor the buyer inherits. Prospect data, call recordings and CRM content flow to whoever runs the model, under that provider’s terms, in that provider’s regions, and a vendor that will not name the stack has made that dependency invisible rather than absent. It also decides change control: vendors swap underlying models without notice, and output behavior, data handling and cost all move when they do. A vendor that names its models and commits to telling customers when the stack changes has made itself accountable for something specific. One that says proprietary AI has not. And for any tool that writes or speaks to prospects, the stack question connects directly to the disclosure question, since Article 50 of the EU AI Act has made undisclosed AI interaction a compliance matter rather than a taste matter since 2 August 2026.
What does an A on AI Disclosure and Model Transparency mean?
Models and providers named, AI generated content disclosure addressed, and scoring or routing logic explained at a level a buyer can interrogate. One honest caveat, stated here because this index would rather under claim than over imply: the axis bundles three things, the model stack, recipient facing disclosure, and scoring explainability. A grade belongs to the whole axis. A vendor can earn its B mostly on scoring explainability while staying thin on the stack, which is why the vendor profile, with its notes and source basis, is the record to read before relying on any single letter.
How is this different from the outreach compliance evaluation?
Same axis, different side of the desk. The outreach compliance evaluation reads AI disclosure from the recipient’s side: whether a person receiving automated outreach is told they are talking to software, which is where the regulatory exposure lives. This page reads the same axis from the buyer’s side: whether you can find out what you are actually deploying, whose models will see your data, and whether the logic that scores your pipeline can be explained to anyone. A vendor can fail both. The ones that pass both are short lists for a reason.
Grades record what a vendor has published, not what it would say on request, so a low grade means the answer is not public rather than that no answer exists. The axis bundles the model stack, recipient facing disclosure and scoring explainability, and a grade belongs to the whole axis; the per vendor notes on each profile say which parts carried it. Figures on this page regenerate from the live index with npm run model-stats and were last built 2026-09-02. The recipient side of this axis is examined on the outreach compliance evaluation, the model behind data accuracy scores on data provenance, and the outcome claims these models are sold on at AI SDR evidence. Full standards on the methodology page.