Cold email reply handling: 13 reply types and what to do with each

Cold email reply handling means sorting every reply into a class and giving each class one fixed action. Answer interested people the same business day, pause for out-of-office notices, never answer machine mail, and suppress unsubscribes at once.

This guide lists the 13 reply classes and the action for each, shows how to tell machine mail from human replies, sets out the unsubscribe deadlines, gives a positive reply rate definition you can defend, and explains what to expect from AI reply classification.

Cold email reply categories: the 13 classes and the action for each

"Can you do Thursday?", "please remove me" and an automatic vacation notice land in the same inbox, and a plain reply count treats all three as replies. They need three different responses, and only one of them is a lead.

Read the table as a default. "Stop follow-ups" means the automated sequence stops for that person. A human can always continue the conversation.

Sort in this order:

1. Machine mail first (bounce, auto reply, out of office). It needs no judgment about intent.
2. Unsubscribes next. They carry a deadline.
3. Human replies last, by what the person wants.

When a reply fits two classes, the stricter consequence wins. "Not interested, please remove me" is an unsubscribe, and "not me, try Priya" is a referral, because a named person is someone new to contact.

Out of office replies and other machine mail

Machine mail is the biggest class in at least one large dataset. Sales.co, a cold email service, classified 61,770 replies from its own campaigns between 2024 and early 2026:

• 45.1% were auto-replies or out-of-office notes.
• 29.9% were negative.
• 14.1% were positive, a group that includes questions and "contact later".
• 2.1% were referrals, 2.7% neutral and 6.1% other.

That is one company's campaigns, and companies that sell outbound publish many of these figures, so do not read the split as a norm. The direction is the point: a reply count that includes machine mail measures mailbox settings as much as buyer interest.

Machine mail follows standards, so check those before reading the words.

Bounces. RFC 3463 defines the status codes in delivery failure reports. A first digit of 4 is a persistent transient failure that may succeed later. A 5 is a permanent failure that is not likely to be resolved by resending the message as it was. A 5.1.1 means the mailbox does not exist, while the RFC says a full mailbox should be reported as transient. Codes with a 7 in the middle (5.7.x) report security or policy refusals, so check your own authentication and content before you blame the address. Count bounces in their own rate, as the deliverability guide explains, never as replies.

Auto replies. RFC 3834 says automatic responses should carry an Auto-Submitted header with the value auto-replied, and should not be sent in answer to any message carrying that header with a value other than "no". Treat the header as a strong signal and never let your own automation answer a message that has it. The RFC says "should", not "must", so a missing header proves nothing. Keep text checks as a backup.

Out of office replies to cold email. Both big mailbox providers limit how often a vacation notice goes out, so a cold email may get none:

• Gmail. Google says the reply goes to a sender the first time they write, and again if they write after four days or the owner edits the reply.
• Outlook. Microsoft says automatic replies go once to each sender, and recommends limiting external replies to contacts, because replying to everyone includes newsletters, ads and junk.

So someone on leave may send nothing, and silence is not evidence the person is at work. A second notice from a paused contact should change nothing. If the notice gives a return date, resume a day or two after it, since the first day back is crowded (our judgment, not a sourced norm). A notice saying the person has left the company is a wrong-person reply.

Unsubscribe replies: the class with a deadline

Three sources set the clock, and they disagree on its length.

• CAN-SPAM (US). The FTC says an opt-out must be honored within 10 business days. A reply email is an acceptable way to opt out, and you cannot require more than a reply or one web page visit. B2B email is covered. Penalties run up to $53,088 per email.
• Gmail. Google recommends honoring unsubscribe requests within 48 hours, and its sender FAQ lists missed 48-hour requests among the issues that cost a bulk sender Google's delivery support. A bulk sender sends close to 5,000 or more messages a day to personal Gmail accounts, counted per primary domain, so most small teams are below it. The FAQ writes this for one-click unsubscribes, so applying it to replies is our judgment.
• GDPR and UK GDPR. Article 21 lets people object to direct marketing at any time, after which their data is no longer processed for it. The UK regulator, the ICO, calls the right absolute, expects action without undue delay and within one month at the latest, and prefers suppression to deletion. Its PECR guidance allows one immediate confirmation but no later contact, even to ask whether they want to opt back in.

Suppress at once, by address, on a list that every sequence checks before every send. Do not delete the contact, or a later import can re-add them. The GDPR cold email guide covers the wider picture.

The hard part is the line between "not interested" and "remove me". Smartlead's category descriptions show it: its Not Interested example includes "Please remove me from your list", while unsubscribe requests sit in a separate Do Not Contact class. The costs are lopsided. A missed opt-out is a complaint and a legal risk, and a wrongly suppressed prospect costs one lead. When wording asks you to stop contact, treat it as an unsubscribe.

This section summarizes primary texts and is not legal advice. Which law applies, CAN-SPAM, GDPR, UK GDPR or another, depends on where you and your recipients are, and the ICO pages we cite say they are under review after the Data (Use and Access) Act.

How to respond to each reply type

Interested, meeting request, question. Answer what was asked, in the order asked, and do not pitch again. End with one next step: two time slots or a booking link. If the question is "how did you get my email?", name the source. GDPR Article 14 lists the source among the information owed to people whose data you did not collect from them.

Objection. Acknowledge the point in half a sentence, answer it once with one fact, offer the cheapest next step, and accept a second no.

Not now. Thank them and repeat their date so they can correct it, then schedule exactly one contact. If they gave no date, pick one, say so, and let them veto it.

Not interested. No reply is acceptable. Do not ask why.

Referral and wrong person. Thank the sender, confirm the name and address as given, and write to the new person separately, saying who suggested them. That contact is data you got from someone else, so under GDPR the Article 14 notice generally applies: within one month, and at the latest at your first message if you use the data to contact them. Exceptions exist. With no name given, one line asking who owns this is enough.

Unsubscribe. One line: you are removed and will not hear from us again. No pitch, survey or resubscribe link.

Other. A person reads it. "Who is this?" gets a plain two-line answer. A hostile or legal reply gets no argument and an immediate stop.

How fast to reply

A 2011 Harvard Business Review article reported a study of 1.25 million online sales leads at 29 B2C and 13 B2B companies in the U.S. Firms that contacted a lead within an hour were nearly seven times as likely to qualify it (a meaningful conversation with a key decision maker) as firms that tried an hour later, and more than 60 times as likely as firms that waited 24 hours or longer.

These were inbound web inquiries, not replies to cold email, and one of the three authors was the CEO of InsideSales.com, a sales software vendor. We found no study of reply speed for cold email replies that we could verify.

Our rule, a judgment: the same business day for interested, meeting and question replies, and within a day for other human replies. Unsubscribes do not wait for a person.

Positive reply rate definition and how to calculate it

A usable positive reply rate definition has three parts, written down before the first campaign:

1. Numerator. Unique people whose human reply is interested or a meeting request. Decide once where questions, not now, objection and referral go. Here, questions are tracked separately. Some vendors count them as positive (Sales.co does), so always say which you use.
2. Exclusions. Auto replies, out-of-office notes and bounce notices are not replies. They leave the numerator and the reply count.
3. Denominator. Unique people contacted, not emails sent. A three-step sequence sends up to three emails to one person, so an emails-sent rate shifts with sequence length even when nothing else does.

Here is a hypothetical example, not FoxKnock data. You contact 400 people with a three-step sequence that sends 1,000 emails, and six addresses hard-bounce. The inbox collects 34 messages: 6 bounce notices, 11 auto replies and out-of-office notes, and 17 from people. Of those 17, 4 are interested, 2 are meeting requests, 2 are questions about the offer, 3 are not now, 4 are not interested, 1 is an unsubscribe and 1 is a referral.

One inbox produces numbers more than ten times apart (8.5% against 0.6%). At 400 people, each positive reply moves the rate by 0.25 points, so compare rates only across hundreds of contacts. For the difference between a reply and a hot lead, see what is a hot lead.

Published benchmarks disagree, partly for the same reason:

• Instantly. Its 2026 benchmark report (2025 data) gives a 3.43% average reply rate: all replies, including follow-up responses, divided by emails sent.
• Belkins. The agency, which runs outbound campaigns for clients, reports 0.45% for 2025 from 34,393 replies on 7,530,489 emails, counting unique replies and excluding auto-replies and bounce notices.
• Sales.co. Its page carries 2.09% per unique contact and a 0.83% human reply rate, and says they are different measurements.

The first two differ by about 7.6 times. One vendor page can carry two definitions as well: Instantly's metrics article describes positive reply rate as the percentage of replies that express interest, and its formula divides positive replies by delivered emails.

All of these come from companies that sell outbound or its tools. Use them as ranges, and compare against your own earlier campaigns under a fixed definition.

AI reply classification: what to expect

Ambiguous replies are hard for anyone to label: "Sure, send it over" can be interest or a brush-off. Three rules keep AI reply classification safe:

1. Set a confidence floor. A label below it goes to a person and triggers nothing.
2. Weigh the cost of a mistake. Automate actions that are cheap to reverse, such as pausing for an out-of-office notice or suppressing an unsubscribe, and keep a person in front of any outgoing message (our judgment).
3. Audit a sample. Check a few labeled replies each week.

Vendor accuracy figures are self-reported. Smartlead's page says its AI sorts replies into six categories and reached 94% accuracy in its internal testing. We found no independent test of any reply classifier.

Taxonomies differ too. Instantly enables three built-in labels by default (Interested, Not interested, Out of Office), with custom labels on its higher plans. Thirteen classes is our choice, made so that each label maps to one action. More classes mean more borderline calls between neighbors, such as interested versus question.

How FoxKnock handles replies

FoxKnock's AI inbox uses the same 13 classes.

• Headers. It reads automatic-reply headers (Auto-Submitted, X-Autoreply, Precedence) as a signal, and header-flagged out-of-office and auto replies can be labeled without calling the AI.
• Confidence. Every label carries a confidence, and anything below 0.6 is marked Needs review, with its own inbox filter.
• Fallback. Labels come from Claude, by Anthropic, so reply text is sent to Anthropic for classification. If the AI is unavailable, a keyword fallback labels instead. Clear opt-out wording such as "remove me" is acted on even when the AI is unsure or unreachable.
• Corrections. You can correct any label. Corrections feed a classifier precision figure on the Analytics page and do not retrain anything. That figure is the share of a label nobody changed, so it cannot see mistakes nobody noticed.
• Cost. At default credit prices, labeling and suggested replies cost no credits, within a daily AI budget.

Four auto-actions follow from the labels:

• Out of office holds the sequence until one business day after the return date, or a week out when there is no usable date (on by default).
• Unsubscribe adds the address to the workspace suppression list (on by default). If you switch it off, a person gets a confirm task.
• Referral creates a contact at the same company and a task to reach out, when the reply names someone. FoxKnock sends nothing to them.
• Meeting request can trigger a booking-link reply. It is off by default, needs confidence of at least 0.9, uses only your own booking URL, applies only to requests under an hour old, and sends at most one per conversation per week. Among reply actions it is the only one that sends without a person approving that message, and it is opt-in.

Replies that call for an answer get a suggested response drafted from your knowledge base, and you send it. Hard bounces are parsed from delivery reports: the address is marked invalid and suppressed, sequences stop, and for addresses FoxKnock found, the email-find credits are refunded once. A meeting is recorded only when a person clicks Mark meeting booked.

Analytics shows reply, positive and meeting rates per person contacted, bounce rate per email, low-sample warnings and cost per positive reply. It counts interested replies and meeting requests as positive, so question replies are not in that number. Bounces, auto replies and out-of-office notices never count as replies.

To try FoxKnock on your own prospects, start free. The Free plan includes 500 one-time credits after you verify your email, with no card, and covers AI onboarding from your website, discovery and standard research. The pricing page lists the AI inbox that sorts replies on the Starter plan.