Skip to content
Saleswoman

AI SDR that checks its own draft before you send it

It researches the account, writes the email, and marks every factual claim as cited or unverified. It scores spam risk before the send, and it stops at the daily cap you set.

Your data stays yours. Prospect data is never used to train a model. Nothing is charged today.

Or paste your own prospect
Try it on

Spam risk Grade words today

Spam signals found: . Footer check: .

The pre-send checklist comes with it. Nothing is charged today.

The draft

To (domain redacted)

Claim ledger

of cited

This draft asserts nothing checkable, so there is nothing to cite.

Run it on your own prospect

Four inputs, all prefilled. No signup.

Angle

Daily cap, per mailbox

In the product, Saleswoman also reads the account's public pages and your CRM. In this demo it only has what you pasted, so anything it cannot source here is flagged rather than assumed.

The domain is the asset, and the bot is pointed at it

Outbound still works. What changed is the cost of getting it wrong, and the person who has to approve an AI sender knows exactly what is at stake. This is hard to sign off on for three reasons, and none of them is a matter of taste.

  1. The thresholds are published, and they are not generous

    Google and Yahoo's bulk sender requirements have been in force since 1 February 2024. Senders of 5,000 messages a day or more must keep the spam complaint rate reported in Postmaster Tools below 0.30%, and Google tells senders to stay under 0.10% and never reach 0.30%. SPF, DKIM, DMARC and one-click unsubscribe are requirements, not suggestions. Cross the line and the domain takes the hit, not the sequence.

    CITED
    • Google, Email sender guidelines. In force 1 February 2024.
    • Yahoo, Sender best practices. Same date, same thresholds.
    • The 0.30% and 0.10% figures are the rates as reported in Google Postmaster Tools, not a rate we measured.

  2. Penalties are counted per message

    Under CAN-SPAM, each separate email in violation is treated as a separate violation, and the FTC adjusts the maximum civil penalty for inflation every year. A bot that sends four thousand non-compliant messages is not one problem to fix. It is four thousand.

    CITED
    • CAN-SPAM Act, 15 U.S.C. 7704.
    • FTC, annual civil penalty inflation adjustment rule. The current maximum is deliberately not quoted here, because it changes every year and a stale figure is a wrong one.

  3. A hallucinated detail fails in public, in writing

    A model that invents a funding round, a headcount or a tool the prospect does not use will not fail quietly in a log. It fails in front of the buyer, from your domain, with your name on it. That is the sentence a VP Sales imagines when they are asked to hand a bot the keys.

    UNVERIFIED
    • No source. This is the failure people describe to us, not a rate anyone publishes, so it goes into the ledger as unverified rather than dressed up as a statistic.
    • That is the same call the product makes on your draft, and it is why this column exists on our own page.

Two of the three check out. One does not, and it stays marked. That is the whole product, run on this page.

It has to show its work before it is allowed to send

This is what supervised actually means: three mechanics, all of them visible in the demo above. Nothing here is a setting buried in an admin screen.

Cited or flagged

Every factual sentence is tied to a source, or it is marked unverified.

The claim ledger lists each sentence that asserts something checkable. A cited claim carries the line it came from. An unverified one carries a suggested rewrite that removes the guess, usually by asking instead of stating.

Scored before it sends

The checks run on the draft, not on a post-mortem report a week later.

Spam risk with the specific triggers named, reading level, word count, unsubscribe present, address placeholder present, subject line not deceptive. Each one passes or fails on its own line, so the score is arithmetic you can follow.

Capped by you

A hard daily ceiling per mailbox. The agent stops at it.

You type the number. There is no override inside the agent, no burst mode, no quota it is trying to hit. The verdict strip shows the count against the ceiling on every draft it produces.

Capability by capability, with the limits stated on the same row, on sales automation software. The check itself is broken down on spam score checker.

Research, draft, check, hold

Four steps, in this order, every time. The fourth is the one the category skips. A draft that fails a check does not go out and does not sit in a report nobody opens. It waits for a person.

Research

It reads the account's public pages and the records you connected, and it writes down what it found as source lines. Anything it did not find stays empty rather than being filled in.

What it found

  • CRM note, 14 Mar: renewal on the routing tool is in July.
  • Careers page: two dispatch roles open in Columbus.
  • Headcount growth: nothing found in the sources connected.

Draft

It writes the email in your angle and your tone, at a reading level a busy operator gets through in one pass, with the unsubscribe and address lines already in place.

Subject and open

  • empty miles on the ohio lanes
  • Your careers page has two dispatch roles open in Columbus.
  • Most fleets your size are losing a double digit share of backhaul.

Check

Every factual sentence goes into the ledger, cited or unverified. Spam risk, reading grade, length and the three compliance lines are scored on the draft itself.

The read-out

  • Claims: 3 total, 2 cited, 1 unverified.
  • Spam risk 12/100, no trigger phrases, one link.
  • Unsubscribe present, address placeholder present.

Hold

A clean draft goes into the queue against your cap. Anything unverified stops and waits for you to supply the source or take the rewrite. That is the whole approval loop.

What happens next

  • 1 draft queued, 1 of 40 today.
  • 1 draft held: one claim has no source behind it.
  • Nothing sends while a claim is unresolved.

The long version, one screen per step, is on ai sales agent. Or skip the description: run it on a prospect and read the ledger yourself.

You are signing off on the sending reputation, so you get the controls

For the person who owns the domain. If you are the VP Sales or the RevOps lead being asked to approve this, these are the three answers you need, and none of them is a promise about our uptime.

What you control

  • The daily cap per mailbox, typed by you, enforced by the queue.
  • Which claims the agent may reuse, kept in an approved claim library.
  • Whether every draft needs approval, or only the ones the ledger flagged.
  • Which mailboxes and which domains are connected at all.

What you can prove afterwards

  • Every send with the draft, the ledger and the verdict stored against it.
  • The source line behind each cited claim, as it stood at send time.
  • Suppression and unsubscribe handling, per contact.
  • An audit export you can hand to legal without rebuilding the story.

What it structurally cannot do

  • Send past your cap. There is no override inside the agent.
  • Send a draft with an unresolved unverified claim in it.
  • Send without an unsubscribe line and an address block.
  • Use your prospect data to train a model. It is not in the pipeline.

The full control list, the data handling and the subprocessor categories are on email compliance software, and the same argument is made per role on sales engagement platform.

Adding a rep does not add a bill

Flat monthly, not per seat: priced against what the work is worth rather than how many people watch it happen. Annual billing gives you two months free. Nothing is charged on this page.

Solo

$149 a month

or $1,490 a year

A founder doing their own outbound, with one domain to protect.

  • 1 seat
  • 1 mailbox
  • up to 40 sends a day

Team

Most popular

$399 a month

or $3,990 a year

A sales team that needs CRM sync, per-rep approvals and a shared claim library.

  • 3 seats
  • 5 mailboxes
  • up to 80 sends a day

Scale

$999 a month

or $9,990 a year

Multi-mailbox outbound where every send has to be defensible afterwards.

  • 10 seats
  • 20 mailboxes
  • up to 150 sends a day

Enterprise

Custom

Custom terms and invoicing

SSO, approval policies, a signed DPA, invoicing and custom terms.

  • Unlimited seats
  • Custom mailboxes
  • Cap set per domain
Talk to sales

No card required to start. Nothing is charged today. The full table, the billing questions and the reason there is no free tier are on AI SDR software pricing.

What is included on each plan
What is included Solo Team Scale Enterprise
Seats 1 3 10 Unlimited
Connected mailboxes 1 5 20 Custom
Daily send cap, per mailbox 40 80 150 Set per domain
Claim ledger and spam scoring Included Included Included Included
Review queue and approvals Approve all Per rep Roles and policies Custom policies
Claim library, approved claims it may reuse 10 100 Unlimited Unlimited
CRM sync and deal notes HubSpot, Pipedrive plus Salesforce plus custom fields plus sandbox
Audit export 30 days 12 months Scheduled, unlimited To your bucket
Suppression lists and one-click unsubscribe Included Included Included Included
Support Email Email, 1 business day Priority, shared channel Named contact, SLA
SSO / SAML, SCIM provisioning Not included Not included Not included Included
DPA, security review, invoicing, custom terms Not included Not included On request Included

Sequencers execute, agents send, this one holds

How the category divides up, stated by capability rather than by brand. Every one of these tools has a real job, and for some teams the right answer is one of the other three.

Capability comparison by category
Capability Sequencing tools Data plus sequences Autonomous AI SDR Saleswoman
Writes the first draft for you No, you write it Assisted Yes Yes
Cites a source for each factual claim No No No Yes
Scores the draft before the send No No Post-send reporting Yes
Hard daily cap the agent cannot pass Throttles, editable Throttles, editable Usually a target, not a ceiling Yes, set by you
Holds a draft that fails a check Not applicable Not applicable No Yes
Audit trail of draft, claims and verdict Activity log Activity log Activity log Yes, per send
Pricing shape Per seat, annual Per seat Annual contract Flat monthly

Category descriptions, not vendor claims. Where a tool in one of these categories does a thing differently, its own documentation is the authority. The longer version, including when to buy one of the others, is on the AI SDR tools comparison.

Answered plainly, limits included

The six questions that decide it. The one that loses the sale is first, and it is open.

Will this get my domain blacklisted?
That is the right question to lead with, and no tool can promise it will not happen. What Saleswoman does is remove the ways it usually happens. You set a hard daily ceiling per mailbox and the send stops at it, not near it. The sequence pauses itself if your complaint rate reaches 0.10%, which is the level Google tells senders to stay under, well below the 0.30% threshold in the same guidance. Suppression is checked at draft time rather than at send time, so a suppressed contact never reaches a queue. And nothing goes out without an unsubscribe line. If you want the extra layer, send from a separate domain: there is an answer on that further down.
What stops it from making things up about my prospect?
Every sentence in a draft is classified as either asserting a checkable fact or not. A sentence that asserts one has to point at a source you supplied: a CRM field, a note, a page you connected. If it cannot, the sentence is marked unverified in the claim ledger, the draft is held, and you either supply the source, take the suggested rewrite or cut the sentence. You can watch it happen in the demo on the homepage: leave the "one fact you can prove" field empty and read what the ledger says about the draft that comes back.
Does the recipient know it is AI?
Yes, if you leave the disclosure line on, and it is on by default. The draft carries a line saying the email was written with AI assistance, and you can edit the wording. Turning it off entirely is possible, and we think it is a bad idea in most markets, particularly in the EU where transparency obligations are tightening. The compliance check tells you which of the three lines are present in every draft: disclosure, unsubscribe and a postal address.
How many emails a day will it actually send?
Exactly as many as you allow. The cap is a number you type, per mailbox, and the plans are built around it: up to 40 a day on Solo, 80 on Team, 150 on Scale. Those are ceilings, not targets, and there is no override inside the agent. No burst mode, no catching up after a quiet day, no quota it is trying to reach.
Does it work for GDPR, UWG and CASL markets?
It is built so you can operate inside them, but it is not legal advice and it cannot consent on your behalf. Practically: consent status is a field, suppression is enforced before drafting, every send stores the lawful basis you recorded, and data can be held in an EU region on Enterprise. In Germany, cold B2B email without prior consent is restricted under UWG ยง7 and no tool changes that. In Canada, CASL requires consent or a documented existing business relationship. Saleswoman records which one you claimed for each contact, so the record exists when someone asks for it.
How is this different from an autonomous AI SDR?
Two ways, and both are structural rather than a matter of quality. First, the output is inspectable: every draft arrives with a ledger saying what it asserted, where each claim came from, and why it did or did not clear. Tools sold as an autonomous employee return an email and a dashboard of activity. Second, this one is built to stop. Caps, suppression, a pause on complaint rate and a hold on unsourced claims are core behaviour, not settings that can be turned off to hit a number. On price, the top self-serve tier here is $999 a month flat rather than per seat, while the sequencing tools anchor per seat and the autonomous agents sit on annual contracts. There is a capability comparison, by category and without naming vendors, on the comparison page.

The other twelve, grouped by what they are about, are on the ai sales assistant FAQ.

Run it on a prospect, then decide.

The demo is the product surface. Draft an email, read the claim ledger, and see the verdict before anything is connected to your domain.

Run it on a prospect

Your data stays yours. Prospect data is never used to train a model. Nothing is charged today.