Kira: Got a DM last Tuesday from a guy in the community — runs a two-person dev agency out of Medellín. He said, "Kira, I sent four hundred cold emails last month. Got two replies. Both were people telling me to stop emailing them."
Santi: Four hundred emails. Two angry replies.
Kira: Two angry replies. And here's the part that got me — he said, "I'm using all the tools. Apollo for lists, Instantly for sending, ChatGPT for personalization. I'm doing everything the YouTube guys say to do. Why is nobody responding?"
Santi: Because he's doing everything the YouTube guys say to do.
Kira: Right.
Santi: No, seriously. The playbook everyone's running — scrape a list, generate a personalized first line with AI, blast it through ten inboxes — that playbook is dead. It died in November twenty twenty-five when Gmail started actually enforcing their bulk sender rules. And LinkedIn is scanning for over six thousand browser extensions now. They're actively hunting automation.
Kira: So he's burning domains, risking his LinkedIn account, and annoying four hundred people a month. For zero pipeline.
Santi: For zero pipeline. And the worst part? He doesn't even know his domains are probably already flagged. He's never opened Postmaster Tools.
Kira: He doesn't know what Postmaster Tools is.
Santi: That's the gap. The whole outbound game changed, and nobody updated the playbook for people like us — small teams, nomad founders, no dedicated sales ops person. We need booked calls, but we can't afford to torch our sending domains or get our LinkedIn restricted while we're three time zones from our clients.
Santi: If you're sending cold outreach right now without authenticated domains, without daily caps, without a reply classifier routing interested replies back to you in under fifteen minutes — you're not just wasting time. You're actively poisoning the infrastructure you'll need when you figure out what actually works. Every burned domain is a domain you can't use later. Every spam complaint pushes your sender reputation closer to a cliff you can't climb back from. And Gmail made that cliff permanent.
Kira: So today we're replacing the whole approach. Not tweaking it — replacing it. An AI research agent that finds real angles from public sources, a hybrid LinkedIn-plus-email sequence with hard caps, and a reply classifier that gets you on the phone while the prospect still remembers your name.
Kira: Okay, so let's start with why the old playbook broke. Because it's not just "emails are harder now." The platforms changed the rules, and most people didn't notice.
Santi: Gmail's the big one. As of November twenty twenty-five, they're actively rejecting non-compliant traffic. Not filtering it — rejecting it. Temporary errors, permanent bounces. And here's the number that matters — if you ever send close to five thousand messages in a single day to personal Gmail accounts from one primary domain, you're classified as a bulk sender. Permanently.
Kira: Wait — permanently? Like, one bad day and you're flagged forever?
Santi: Forever. Google's own FAQ says it. Once you hit that threshold even once, you're a bulk sender for life on that domain. And bulk senders have to authenticate with SPF, DKIM, DMARC, align their From address, send over TLS, implement one-click unsubscribe via RFC eighty fifty-eight, and keep their spam rate below zero point three percent. Miss any of those, and your emails start bouncing.
Kira: And zero point three percent sounds like a lot of room until you realize — that's three complaints per thousand emails. If you're sending fifty cold emails a day from one inbox and two people mark you as spam in a week, you're already in the danger zone.
Santi: Exactly. And Microsoft started enforcing the same stuff in May twenty twenty-five for Outlook, Hotmail, Live. So it's not just Gmail anymore. The whole inbox ecosystem is tightening.
Kira: And then there's LinkedIn. Which is the other half of this.
Santi: LinkedIn is on a warpath.
Kira: They're scanning browsers for automation extensions. Tom's Hardware reported they're checking for over six thousand Chrome extensions. And their policy is explicit — no third-party software that automates, scrapes, or modifies LinkedIn activity. Bots, browser extensions, anything. Violations get you restricted or banned.
Santi: Which means every LinkedIn automation tool people are running — the ones that auto-send connection requests, auto-like posts, auto-view profiles — all of that is a policy violation. And LinkedIn is getting better at catching it.
Kira: So the guy from Medellín? He's running Apollo scrapers, blasting through Instantly, probably has a LinkedIn automation tool running in the background. He's violating terms of service on three platforms simultaneously.
Santi: For two angry replies a month.
Kira: Okay, so what's the alternative? Because we still need pipeline. We still need booked calls.
Santi: The alternative is fewer, better touches. And it starts with research — actual research on each prospect before you ever reach out. Not "Hey first name, loved your recent post about topic" — that's AI slop and everyone recognizes it now.
Kira: I get those every day. "Kira, I noticed your agency is doing amazing things with AI content." Delete.
Santi: Right. So instead, you build a lead-research agent. I use Clay for this. You set up a table with input columns — company URL, their about page, blog or news page, maybe a recent press release. All public URLs. No LinkedIn. No Sales Navigator links.
Kira: Why no Sales Navigator?
Santi: Because you want to stay clean on platform terms. Apollo's terms of service explicitly prohibit bots and scrapers outside their built-in features. LinkedIn prohibits scraping. So you feed Clay public web pages only — the stuff anyone can find with a Google search.
Kira: And then what happens?
Santi: Clay runs an AI step — Claude, GPT, whatever model you want — with a system prompt that says: read these public pages, produce five to ten outreach angles, each one referencing a specific thing you found on a specific page. Not flattery. Not "love your work." Each angle has to cite a real quote or section from their site, include the source URL, and explain why this matters right now.
Kira: So you're not generating a personalized first line. You're generating a research brief.
Santi: A research brief with angles. And each angle gets a confidence score. If the sources are thin or outdated — more than eighteen months old — the confidence drops. If the agent finds a "no cold outreach" signal on their site, it returns zero angles and flags the row. You never email that person.
Kira: That's a quality gate I've never seen anyone talk about.
Santi: Because most people skip quality gates entirely. They generate the first line and send. We generate ten angles, reject the weak ones, pick the single best one for the first message, and hold one or two backups for follow-ups.
Kira: And this is the important part — you're picking one angle per message. Not cramming three personalized observations into a paragraph that screams "a robot wrote this."
Santi: One angle. One reason to reply. That's it.
Kira: So you've got your angles. Now what? Because this is where I think most people go wrong — they jump straight to email.
Santi: And that's backwards. For most B2B prospects, LinkedIn first. Manual. All of it.
Kira: And people just groaned.
Santi: They groaned, but listen — twenty actions a day max. Connection requests, DMs, meaningful comments combined. The idea is simple. Before you ever pitch, this person has seen your name three or four times. You've viewed their profile, liked a post, left a real comment — not "great insight," two or three lines that add something. By the time you send a DM, you're not a stranger.
Kira: Okay but twenty actions a day for a solo founder — that's real time.
Santi: Thirty to forty minutes if you batch it. And the DM isn't a pitch. It's a permission ask. "I can email you a two-paragraph teardown of how teams in your space are solving this problem. Okay to send?"
Kira: You're asking permission to email them.
Santi: Which means when they say yes, your email lands in a warm inbox. They're expecting it. They open it. Deliverability stays clean. And if they don't accept the connection request? Skip the DMs entirely. Go email-only. The sequence adapts.
Kira: So LinkedIn is the warm-up layer, and email is the delivery layer. And you never force one channel if it's not working.
Santi: Exactly. On the email side, you ramp each inbox slowly — fifteen a day the first week, twenty-five the second, cap at fifty or sixty ongoing. Send Tuesday through Thursday, randomized within a morning or afternoon window. Offset by seven to eleven minutes so it doesn't look automated.
Kira: And you're not sending from your primary business domain.
Santi: Never. Two to four secondary domains — brand-consistent variants. Each one gets its own SPF, DKIM, DMARC setup. Each one warms up independently over four weeks.
Kira: I want to talk about the pause rules because I actually dug into this. Instantly and Smartlead both document rotation and throttling as core deliverability features — they assume you're running multiple domains. But neither of them will save you if your Postmaster numbers go red. The thresholds are on you.
Santi: Right. And the hard lines are — if your Gmail Postmaster spam rate hits zero point two percent for two consecutive days, pause that domain for seventy-two hours. If it hits zero point three percent on any single day, stop for seven days and consider retiring the domain. Hard bounces above two percent — stop the list, re-validate every address. And Amazon SES will put you under review at zero point one percent complaint rate, which is even tighter than Gmail's ceiling.
Kira: Have you actually had to retire a domain?
Santi: Once. Early last year. I was testing a new ICP segment, got lazy with list hygiene, and my spam rate spiked to zero point four percent in Postmaster. I paused it, cleaned the list, tried again two weeks later — the reputation never recovered. That domain is dead to me now.
Kira: Ouch. And you can't resurrect it? Like, at all?
Santi: I tried. Two weeks of clean sends, low volume, perfect list. Postmaster never moved. Once the reputation tanks, it tanks.
Kira: And that's real money. Domain, mailboxes, four weeks of warm-up time — all gone. Which is why the caps exist. You protect the infrastructure first.
Santi: Pipeline second.
Kira: Okay, so messages are going out. Replies start coming in. And this is where I actually get excited, because most people treat replies like a to-do list they check once a day.
Santi: Which is insane when you look at the data.
Kira: Harvard Business Review published a study — older, two thousand eleven, but it's been validated over and over — companies that responded to leads within one hour were seven times more likely to qualify them than companies that waited longer. Seven times. And the InsideSales research from MIT found that responding within five minutes — five — gives you an order-of-magnitude improvement in contact rates versus waiting thirty minutes.
Santi: You're quoting the HBR study at me.
Kira: I read things! But seriously — imagine you're a solo founder in Chiang Mai. It's eleven PM your time. A prospect in New York replies at ten AM Eastern saying "yeah, let's talk." If you don't respond until your morning — that's twelve hours. By then they've moved on. They've replied to someone else. The window closed.
Santi: So you build a classifier. Every reply gets labeled automatically — Interested, Referral, Not Now, or Spam slash Unsubscribe. The classifier reads the thread, looks for signals. "Let's talk," a Calendly link, a specific time window — that's Interested. "Try my colleague Sarah" — that's Referral. "Circle back next quarter" — Not Now. "Unsubscribe" or anything hostile — Spam.
Kira: And the routing is where it matters. Interested replies trigger an alert — Slack, Telegram, whatever you use — with a task due in fifteen minutes during your working hours. If the reply comes in outside your hours, it queues for nine oh five the next business day. The system drafts a reply with the angle you used and a two-slot calendar link. All you do is review and hit send.
Santi: Referrals create a new contact automatically and send a "so-and-so suggested I reach out" email to the referred person. Not Now replies get logged with the date they mentioned and enrolled in a single check-back step on that date. And Spam or Unsubscribe — immediate suppression. Person and domain. Gone from all future sends.
Kira: That fifteen-minute window is the whole game. Because you're competing against every other vendor in that prospect's inbox. The one who responds fastest with something relevant wins.
Santi: Okay, I want to address something because I know people are thinking it. There's a real argument that AI personalization at scale has diminishing returns. That tight segmentation and good timing can outperform a personalized first line. And honestly — they're not wrong.
Kira: Yeah, I've seen this too. Some operators are reporting that AI-personalized first lines actually underperform compared to well-written segment-specific copy. Because the patterns are detectable now. Prospects can smell the AI.
Santi: And that's a legitimate concern. If your "personalization" is just ChatGPT rewriting the same template with a different company name dropped in, you're not personalizing — you're decorating spam. And Gmail's filters are getting better at detecting those patterns too.
Kira: So why not just skip personalization entirely and send more volume through more inboxes?
Santi: Because volume is what kills you. More volume means more domains, more warm-up time, more surface area for spam complaints, more chances to trip the zero point three percent wire. The math doesn't work for small teams. We don't have twenty domains and a dedicated deliverability person.
Kira: Right. We have two domains and a Postmaster Tools tab we check on Mondays.
Santi: So the answer isn't "personalize everything" or "blast more volume." It's — research deeply, send to fewer people, and make every touch count. The AI agent isn't generating cute first lines. It's doing the research you'd do manually if you had three hours per prospect. It just does it in forty-five seconds.
Kira: And the compliance layer protects you while you figure out what messaging actually converts. Because you can iterate on copy without burning infrastructure.
Santi: That's the whole architecture. Relevance protects your reply rates. Compliance protects your domains. The classifier protects your speed. All three have to work together or the system breaks.
Kira: And the Lisbon Test — can you run this from a café with sketchy wifi?
Santi: The LinkedIn steps are manual, so yes — you just need a browser. Clay runs in the cloud. Your sequencer runs in the cloud. The classifier runs in the cloud. The only thing that needs you is reviewing replies and showing up to calls.
Kira: Which is the part you should be spending time on anyway.
Santi: So back to the guy in Medellín. Four hundred emails, two angry replies. If he ran this system instead — researched twenty prospects deeply, warmed five of them on LinkedIn, emailed the rest with one real angle each, and had a classifier routing replies in fifteen minutes — he'd probably book more calls from twenty sends than he did from four hundred.
Kira: And he'd still have his domains. He'd still have his LinkedIn account. He'd still have the infrastructure to try again next month with better targeting instead of starting over from scratch.
Santi: That's the shift. Outreach isn't a volume game anymore. It's an infrastructure game. Protect the infrastructure, and you can iterate forever. Burn it, and you're buying new domains every quarter wondering why nothing lands.
Kira: We put together the Outbound Safety and Relevance Kit — it's on the Resources page. The Clay agent schema, both eight-touch sequences with the exact pacing and objection handlers, the domain rotation SOP with the pause thresholds, the reply classifier labels and routing logic, and a Postmaster Tools weekly checklist. It's the whole system we just walked through, ready to copy and paste.
Santi: One thing this week. Go to Postmaster Tools, verify your sending domains, and check your spam rate. That's it. If you've never done it, you might be surprised by what you find. And if the number's clean, you've got a foundation to build on.
Kira: And if it's not clean, you just saved yourself from sending another four hundred emails into the void.
Santi: Build from anywhere. Build smart.
Kira: See you Wednesday.