AI SDRs are worth it for the research and drafting half of outbound: finding accounts, checking fit, and writing a first email from real signals. They are risky when left to send at volume on their own, because Gmail, Yahoo and Outlook judge you on authentication and spam complaints, and a burned domain takes your normal email down with it.

Picture a ten-person B2B services firm that switches on an AI SDR on a Monday. By Wednesday it has written and sent a few thousand "personalised" emails from the company's main domain. Replies trickle in, but so do spam reports. By the following week, proposals sent to existing clients start landing in junk, and nobody can say which setting caused it.

The tool did what it was set up to do. The problem was the setup: the same domain for cold outreach and client mail, no limit on volume, and nobody reading what went out.

What is an AI SDR?

An AI SDR is software that does parts of a sales development representative's job with AI models: finding target accounts, researching them, writing outreach, sending it and sorting the replies. Products differ mainly in how much of that chain they automate and where a person reviews the work.

A sales development representative (SDR) is the person who books first meetings for the sales team. Most outbound tools now include some AI, so "AI SDR" covers a range:

• Research and enrichment agents. Clay describes its Claygent as "agents in Clay that perform tasks on your behalf": researching accounts, ranking and qualifying leads, and writing personalised copy, with a reasoning trace for each decision.

• All-in-one prospecting platforms. Apollo offers AI research, an AI writing assistant and automated email and phone engagement on top of its contact database.

• Sending platforms with agents. Instantly lists an AI Sales Agent, an AI Reply Agent and inbox warmup alongside its sending infrastructure.

The question is not which category is "real". It is which steps you let run without a person.

What do AI SDRs do well?

AI SDRs do well at work that is repetitive, reads public information and produces something a person can check before it matters. Research, list building, qualification and first drafts all fit that description.

Account research and qualification

An agent can read a company's site, job posts and news, then answer the same questions for every account: do they fit our customer profile, did something change recently, who owns the problem we solve? That is slow, dull work for a person and a good fit for a model, as long as each answer comes with the source it used.

Signal-based first drafts

A first draft built from a real signal ("you're hiring three support agents") beats a template with a first-name token. AI is good at turning a research note into a short, specific opening. A person is still better at deciding whether the signal is worth an email at all.

Reply sorting

Sorting replies into interested, not now, wrong person, out of office and unsubscribe is a classification task models handle well. The value is speed: the interested replies reach a person the same day.

Where do AI SDRs burn your domain?

They burn your domain when volume rises faster than relevance. Mailbox providers watch how recipients react to your mail, and an agent that sends thousands of loosely targeted emails generates spam reports faster than any person could.

Domain reputation is the trust mailbox providers place in mail from your domain, built from authentication results and how recipients treat your messages. When it drops, all mail from that domain suffers, including invoices, proposals and password resets. Four habits cause most of the damage:

1. Sending from the main company domain. Cold outreach and client mail share one reputation.

2. Scaling volume on day one. New mailboxes and domains have no history; a sudden spike looks like spam.

3. Thin personalisation. A company name dropped into a template is still a template, and recipients report it.

4. Unread output. An agent that misreads a site will repeat the same wrong claim to every account on the list.

What do Gmail, Yahoo and Outlook require from senders?

All three require bulk senders to authenticate their mail with SPF, DKIM and DMARC and to keep spam complaints low; Gmail and Yahoo also require one-click unsubscribe for marketing mail. Missing these means junk folders or outright rejection.

• Gmail. Google's email sender guidelines apply to all senders to Gmail accounts from 1 February 2024. Senders of 5,000 or more messages a day must set up SPF, DKIM and DMARC, "keep spam rates reported in Postmaster Tools below 0.30%," and support one-click unsubscribe with "a clearly visible unsubscribe link in the message body."

• Yahoo. Yahoo's sender best practices ask bulk senders to implement SPF and DKIM, publish a DMARC policy of at least p=none, keep the spam rate below 0.3%, support one-click unsubscribe and "honor unsubscribes within 2 days."

• Outlook. Microsoft's requirements for high-volume senders apply to domains sending 5,000 or more messages a day to Outlook.com, Hotmail, Live.com and MSN addresses. Microsoft's support page for error 550 5.7.515 says mail from such domains that fails SPF, DKIM and DMARC checks is rejected.

SPF (Sender Policy Framework) lists the servers allowed to send for your domain. DKIM (DomainKeys Identified Mail) signs each message so it can't be altered in transit. DMARC (Domain-based Message Authentication, Reporting and Conformance) tells receivers what to do when those checks fail. These thresholds are written for mail to the providers' consumer inboxes, but if your prospects use company mailboxes, treat them as the minimum, not the target.

The law still applies to B2B email

In the United States, the FTC's CAN-SPAM compliance guide says "the law makes no exception for business-to-business email." It requires accurate headers and subject lines, a physical postal address and a working opt-out honoured within 10 business days, and it holds you responsible for what a vendor sends on your behalf. Rules in the EU, the UK and elsewhere differ by country. This is general information, not legal advice.

How do you run an AI SDR without burning your domain?

Separate the sending domain from your main one, keep volume low and steady, and put a person between the agent's draft and the send button until the numbers prove it can be trusted. Treat the agent as a fast researcher, not an unsupervised sender.

Checklist before you switch one on

• Send cold outreach from a separate domain and mailboxes, never from the domain your clients and invoices use.

• Set up SPF, DKIM and DMARC on that domain, and check they pass before the first send.

• Warm up new mailboxes gradually and cap daily volume per mailbox.

• Write the target list rules yourself; let the agent research and qualify inside them.

• Have a person approve every draft at first, then sample a share of drafts once quality holds.

• Include a visible unsubscribe link and a postal address, and sync opt-outs to every tool that sends.

• Watch spam complaints in Google Postmaster Tools and pause sending the moment they climb.

• Measure meetings booked and replies from the right people, not emails sent.

Measure the right thing

An AI SDR makes sending cheap, which tempts teams to measure activity. The useful numbers are positive replies per hundred emails and meetings that turn into opportunities. If those don't move, more volume only adds risk.

How we approach it at Vizio AI

We start with the target list and the message, then add automation. With the client, we write down who the customer is, which signals suggest they need help now, and what a good first email says. Only then do we pick tools for research, enrichment and sending, often a mix rather than one platform.

We set up the sending domains and authentication before any campaign runs, and we keep a person reviewing drafts until reply quality is consistent. We also route replies and opt-outs into the client's CRM, so outbound data doesn't live in a sending tool nobody checks. Clean records matter here; If Your CRM Data Is Inaccurate, Your Analytics Will Mislead You explains why.

This is part of the work of our Growth Studio and our marketing services. If you are choosing the model behind an agent like this, Claude, GPT or Gemini? Choosing a Model for Business Workflows covers how to test candidates on your own data.

Sources

• Google, Email sender guidelines (Gmail Help): https://support.google.com/a/answer/81126

• Yahoo Sender Hub, Sender Best Practices: https://senders.yahooinc.com/best-practices/

• Microsoft Defender for Office 365 Blog, Strengthening Email Ecosystem: Outlook's New Requirements for High-Volume Senders: https://techcommunity.microsoft.com/blog/microsoftdefenderforoffice365blog/strengthening-email-ecosystem-outlook%E2%80%99s-new-requirements-for-high%E2%80%90volume-senders/4399730

• Microsoft Support, Fix NDR error "550 5.7.515" in Outlook.com: https://support.microsoft.com/en-us/outlook/fix-ndr-error-550-5-7-515-in-outlook-com

• Federal Trade Commission, CAN-SPAM Act: A Compliance Guide for Business: https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business

• Clay, Claygent: https://www.clay.com/claygent

• Apollo, AI: https://www.apollo.io/ai

• Instantly: https://instantly.ai/