Hivemind Basic

The surface we map

Hivemind reads the public web for what is documentably true about each of your target accounts this week, then turns only the groundable part into a ready-to-send outreach draft. The object is the per-account public signal (press, announcements, documented events) that a diligent person could find and cite, refreshed on a weekly cadence.

Most AI outreach tools generate a confident message for every prospect, every time. That is exactly the problem: a tool that always has something to say will, when it has nothing real to say, make something up, a flattering assumption or a plausible hook the prospect knows is not true, sent under your name. Hivemind is built on the opposite principle. We would rather send you nothing than send you a claim we cannot trace to a documented public signal.

The evidence standard

Every account moves through the same sequence, and a claim that fails any stage does not advance. The standard is enforced, not just instructed: a finding with no cited signal is rejected before it can become a draft, a cited signal that does not exist in what we actually observed is rejected, and a low-confidence or directory-only source cannot ground a finding.

1. Signal gathering. We search public news and web sources for documented activity tied to the target and their company. Each candidate signal is recorded with its source.
2. Verified findings. Raw signals become findings only when they survive a fidelity check: each finding is tested against the sources it cites and against the brief you gave us. Anything that overreaches its source, or reads back something you told us as if it were discovered, is stripped.
3. Research brief. The surviving findings become a per-account research brief, with depth scored honestly. An account with no groundable findings produces an honest empty brief, not a padded one.
4. The draft. Outreach is generated only from the verified findings. It cannot reach back to raw, unverified signals or to your intake notes. The draft is, by construction, a rewrite of already-verified material into outreach prose.
5. The fidelity gate. Before any draft reaches you, every claim-bearing phrase is checked against the verified findings it is allowed to draw from. The gate catches four things:
   • Claims that overreach their source
   • Claims grounded in no finding at all (fabrication)
   • Claims that trace to your own intake rather than a public signal
   • Claims that no single finding supports but that emerge from combining findings into an implication neither one states, the subtle failure where every phrase is true but the sentence asserts something unproven
   If every claim is anchored and clean, the draft is delivered. If anything fails, the draft is withheld, recorded for our review, never sent to you as if it passed.
6. Delivery. Delivered drafts arrive on your dashboard with their cited sources attached. You review, edit if you want, and send from your own email, under your own identity.

What counts as a signal

We work only from public, documented signals, the kind of thing a diligent person could find and cite:

• Recent press and trade coverage
• Public statements, interviews, and announcements
• Documented company events: appointments, funding, product launches, expansions

We do not scrape private data, LinkedIn profiles, or your CRM. If a claim cannot point to a public, documented source, it does not belong in your outreach, so it does not go in. This is also why some accounts come back with no draft: a prospect with little public footprint gives us little to anchor to, and when that happens we tell you plainly rather than manufacture a hook.

The gate is not a wall

Illustrative example. The following is constructed to show how the gate behaves; it is not drawn from a real account or a real engagement.

A system that withheld everything would be useless, not trustworthy, so it is worth walking through how the gate behaves. Picture a first generation for a target that includes the phrase "the company's focus on innovation and safety," a claim that reads smoothly but traces to no specific finding. The gate withholds it. A second generation makes the same outreach using only claims each anchored to a documented signal, and it is delivered, with its sources cited. Same target, same research: the unanchored version withheld, the anchored version delivered.

You send, we never do

Hivemind arms; you send. We do not send outreach on your behalf, ever. Drafts are delivered to your dashboard for you to review and send from your own identity. If you edit a draft, the edited version is your content: our guarantee covers the draft as we generated it, and the verified original is always shown alongside your edit so you can see exactly what we anchored and what you changed.

Ongoing, not one-shot

Hivemind refreshes intelligence on a weekly cadence. New public signal becomes new findings, new findings become new drafts, and every recurring draft passes the same fidelity gate, so the standard never drops on a re-run. Accounts where your prospect has already engaged (replied, or booked a meeting) are suppressed automatically, so you are not re-pitching someone already in conversation.