The best AI SDR agents, graded on what they send without a human and what they tell the recipient
An AI SDR is bought on the promise that it works without you. That makes two questions decisive and neither of them is a feature list: what does it send with nobody watching, and what does the person on the other end think they are reading. The index grades all 61 vendors in this category on both, from public artifacts.
Assessed 2026-08-27. 248 vendors indexed across the whole index, 61 in this category.
30 of the 61 vendors document what their agent does without a human in the loop. 1 of 61 documents whether the person receiving the message is told it came from one. This category publishes its autonomy and withholds its identity.
That second figure is the thinnest cell in the index, and it sits in the one category built to write as if it were a person. 22 of the 61 publish nothing checkable on it at all. The evidence picture is not much better: this is a category sold on meetings booked, and 15 publish outcome evidence with real numbers while 20 publish none beyond assertion. Not one of the 61 vendors clears all 6 of the axes this category is decided on.
What 61 vendors actually disclose
6 of the seventeen index axes decide a purchase in this category, and every vendor in it carries a grade on all 6. A grade of A or B means a buyer can check the claim without contacting the vendor. A D records that nothing was found in public sources on the verification date, which is far more often an absence of disclosure than a defect in the product.
| Axis | Documented | Share | Nothing to check |
|---|---|---|---|
|
Autonomy and Oversight Model
What the system does without a human. Draft for review, auto send, or fully agentic, and what contains a bad run. |
30 of 61 | 49 percent | 3 |
|
Recipient Disclosure and Authenticity
How the product presents itself to the people it targets: whether automated outreach and AI agents disclose themselves, whether sender personas are real, and whether personalization is grounded in verifiable fact. Measured as known compliance with Article 50 of the EU AI Act, in force since August 2, 2026, which requires AI systems that interact with individuals to disclose that fact. |
1 of 61 | 2 percent | 22 |
|
Operational and Outcome Evidence
Measured outcomes with a stated basis: replies, meetings, pipeline, win rates. Logos are not evidence and prestige is not measurement. |
15 of 61 | 25 percent | 20 |
|
Deliverability and Sending Discipline
The operational craft of sending: warmup, rotation, volume governance, spam rate monitoring, and what happens when reputation degrades. |
21 of 61 | 34 percent | 7 |
|
AI Disclosure and Model Transparency
What models power the product, whether AI generated outreach discloses itself, and whether scoring and routing logic is explainable. |
15 of 61 | 25 percent | 15 |
|
Outreach Compliance Posture
How the product handles regulated outreach: consent, DNC scrubbing, opt out mechanics, caller ID conduct, and the public enforcement record. |
11 of 61 | 18 percent | 14 |
The order of that table is the argument. Autonomy is the feature being sold, so it is documented best. Disclosure, outcomes and consent posture are what a buyer inherits the moment the agent starts sending, and they are documented worst. An agent that runs unsupervised is exactly the case where the first column should be the weakest and the rest should be strongest.
The 30 agents whose autonomy boundaries can be checked
30 of 61These are the vendors graded A or B on Autonomy and Oversight Model, ordered by that grade and then by how many of the other five decisive axes they also document. Read the second column before the first: this shortlist is wide and shallow, and 11 of the 30 document their autonomy model and nothing else on this page. Appearing here means a buyer can find out what the agent does unsupervised. It does not mean the rest of the questions have been answered. Several are indexed primarily in another category and appear here because they also sell into this one.
| Vendor | Autonomy & Oversight Model | Recipient Disclosure & Authenticity | Operational & Outcome Evidence | Deliverability & Sending Discipline | AI Disclosure & Model Transparency | Outreach Compliance Posture |
|---|---|---|---|---|---|---|
|
Agentforce Sales
3 of 6 documented
|
A | C | C | C | B | B |
|
Clodura.AI
3 of 6 documented
|
A | C | C | B | B | C |
|
Reply.io
3 of 6 documented
|
A | D | B | D | B | C |
|
Instantly
5 of 6 documented
|
B | B | C | B | B | A |
|
Amplemarket
4 of 6 documented
|
B | C | B | A | C | A |
|
Cadivra
4 of 6 documented
|
B | C | D | A | B | B |
|
Cognlay
4 of 6 documented
|
B | C | D | B | B | A |
|
Rasayel
4 of 6 documented
|
B | C | B | B | C | B |
|
Evergrowth
3 of 6 documented
|
B | C | B | C | A | C |
|
Reachmigo
3 of 6 documented
|
B | C | D | B | C | B |
|
Tuco AI
3 of 6 documented
|
B | D | A | A | D | C |
|
WaLead
3 of 6 documented
|
B | D | C | B | A | C |
|
Artisan
2 of 6 documented
|
B | C | C | B | C | C |
|
Octave
2 of 6 documented
|
B | C | B | C | C | C |
|
Oliv AI
2 of 6 documented
|
B | C | C | C | B | C |
|
Regie.ai
2 of 6 documented
|
B | C | C | B | C | C |
|
Salesgear
2 of 6 documented
|
B | C | C | B | D | D |
|
Selix AI
2 of 6 documented
|
B | C | C | B | D | D |
|
SendroAI
2 of 6 documented
|
B | C | D | B | D | C |
|
Aimdoc
1 of 6 documented
|
B | C | C | C | C | C |
|
CueGrowth
1 of 6 documented
|
B | C | C | C | C | C |
|
ENAI
1 of 6 documented
|
B | C | D | C | C | C |
|
Fuzzy Sequence
1 of 6 documented
|
B | D | C | C | C | C |
|
lemonreach
1 of 6 documented
|
B | C | D | C | C | C |
|
OnePgr
1 of 6 documented
|
B | C | C | C | C | C |
|
OptaReach
1 of 6 documented
|
B | D | C | C | C | C |
|
RingEmAll
1 of 6 documented
|
B | D | D | C | D | C |
|
Sailes
1 of 6 documented
|
B | D | C | D | C | D |
|
SalesCaptain
1 of 6 documented
|
B | D | D | D | D | D |
|
Samplead
1 of 6 documented
|
B | D | C | C | D | D |
Every grade on this page carries a source basis, and on the deciding axis the mix is worth stating plainly: 59 Vendor Published, 2 Third Party Estimated. Nearly all of it is vendor published, and in this category that carries a specific limitation worth naming. A documented autonomy model describes the default configuration, and an autonomous system is bought precisely so that someone can turn the defaults up. What the index can grade is whether the boundary is written down. What it cannot grade is where your team will set it in month three.
Read the full methodology →6 questions the published materials will not answer
Where the index found nothing to grade, these are the questions that close the gap. Put them in the evaluation, and get the answers in writing rather than on a call.
What exactly sends without human review by default, and what is the largest batch that can go out before a person sees anything?
Autonomy claims deserve the same scrutiny at full volume that they would get one message at a time. The number that matters is the blast radius of a bad run, not the feature name.
What stops a bad batch, who gets alerted, and how quickly can we halt a running sequence mid flight?
Every agent has a failure mode where it keeps going. The control that matters is the one that works after the first hundred messages have already gone.
How does the agent identify itself, and are sender identities real employees, labeled agents, or invented personas?
This is the axis the entire category grades worst on. Article 50 of the EU AI Act has applied since 2 August 2026 and reaches deployers, which means you, not only the vendor.
Whose domain and sending reputation is at risk, and what happens to it when a campaign goes wrong?
Sending reputation is a shared asset an agent can burn faster than a team can rebuild it, and it is usually your domain rather than the vendor’s.
What is your meeting rate, measured across what population and what period?
Pilot anecdotes and a stated population are different kinds of claim. Ask which one the number on the website is.
When a prospect replies asking whether they are talking to a human, what does the agent do?
Ask for the configured behavior rather than the policy, and get it in writing. It is the single question that most reliably separates vendors that have thought about disclosure from vendors that have not.
Common questions
What are the best AI SDR agents for outbound prospecting in 2026?
The index does not publish a single ranked winner, because an AI SDR is bought for a specific motion and the honest differentiator in this category is not capability but disclosure. What it does publish is which agents can be checked before they start sending. Of the 61 AI SDR and outbound agent vendors in the GTM Tech Index, 30 document what the agent does without a human in the loop at a grade of A or B, and those 30 are listed on this page with their grades across the 6 axes this category is decided on. Not one of the 61 clears all 6 at A or B. Buyers should weight the autonomy and disclosure axes above the feature comparison, because those are the two that transfer risk.
Do AI SDR agents disclose that they are AI?
Almost none of them say. 1 of the 61 vendors in this category documents its disclosure posture in any checkable way, not one reaches the top grade on that axis, and 22 publish nothing on it at all. Article 50 of the EU AI Act has applied since 2 August 2026 and requires that a person interacting directly with an AI system is informed of that fact unless it is obvious to a reasonably observant person. It reaches deployers as well as providers, so a buyer running an undisclosed agent carries the exposure alongside the vendor. Establish the behavior in the contract rather than expecting to find it on the website. Nothing here is legal advice.
Which index tracks AI vendors for outbound prospecting?
The GTM Tech Index tracks 61 AI SDR and outbound agent vendors, each graded on the same seventeen capability axes from public artifacts, with a verification date and a source basis on every figure. It is independently maintained, no vendor pays for inclusion or placement, and material changes to capability, pricing and compliance posture are logged as they are verified.
Do AI SDR agents actually book meetings?
The category is sold on that claim and rarely evidences it. 15 of 61 vendors publish outcome evidence with named customers and real numbers, 1 publish it with the measurement basis attached, and 20 publish no outcome evidence beyond assertion on a product sold entirely on its results. A meeting rate without a stated population and period is a pilot anecdote, and in this category most of them are.
The GTM Tech Index is an independent editorial reference. No vendor pays for inclusion, placement, or rating. Grades reflect the public evidence on the verification date and change as vendors publish more. Compliance postures are assessed from published sources and public records; the index is not a law firm and nothing here is legal advice. Vendors who believe a record is wrong are invited to point the index at published documentation through the contact form on any page, and corrections are logged in the change log.