The ICP Gate: Filter Guests Before You Invite

Most shows pick guests on how well they will interview, and that one habit is why a full calendar produces no revenue. We run invite led outbound for 50 plus B2B companies at a 4.6% reply rate, against the 3.43% average Instantly measured across billions of cold email interactions, and almost all of that edge is decided before a word of copy gets written. Below, the gate that decides who earns an invite, the 5 checks it runs, the filters worth deleting today, and how to tell when you have set it too tight.

What Is an ICP Gate for Podcast Guests?

The gate is not a scorecard and it is not a vibe. It is a sentence you could hand to a stranger, and the stranger would return the same verdict you would on the same 100 names.

That is the whole test. If 2 people reading your criteria disagree on a row, you do not have a gate yet, you have an opinion with bullet points under it.

ICP Gate

A binary fit rule run on every name before an invite is sent. Output is pass or fail plus one line of reasoning, stored on the row so the decision is auditable later.

Pull Parameter

A countable filter set at list build time in Apollo or a similar tool. Headcount, revenue band, geography, seniority, keyword. These belong in the query, never in the gate.

The distinction between those 2 is where most guest lists quietly break, and it is worth more than any wording change you will make to the invite itself. How you build the guest list sets what reaches the gate. The gate decides what survives it.

Why Does a Written Gate Beat Judgment on Every Name?

Judgment does not scale and it does not hold still. Reviewed on a Monday, a mid market operations lead is a strong fit. Reviewed on a Thursday, with 400 rows left to go, the same person is a maybe, and maybes get invited because deleting a row feels like losing progress.

Volume settles the argument. At 10,000 invites a month nobody is reading 10,000 profiles, so either the standard is written down and runs the same way every time, or the standard is whoever was reviewing rows at 4pm.

There is a second reason, and it is about who is actually on the other end. Gartner puts the typical B2B buying group at 6 to 10 decision makers, with 77% of buyers describing their last purchase as complex or difficult. A guest list built on impressive titles invites 6 people who each own a slice of the decision. A guest list built on a gate invites the 1 who owns the problem.

That person is usually less senior than the title you were reaching for and far more useful, which is the same finding behind defining an ICP for cold email and what an ideal customer profile actually is.

What Belongs in the Gate and What Belongs in the Query?

The query holds everything countable. The gate holds everything that needs a judgment call written down once. Mixing the 2 is the most common way a list gets thin without anyone noticing.

Here is what that looks like in practice. Your Apollo query already restricted the pull to companies of a certain size, in certain regions, at certain seniority levels. Every row that reaches the gate has cleared those. Write those same limits into the gate prose and the grader starts rejecting rows the query already approved, usually because a headcount estimate on the enrichment text disagrees with the one in the database.

You paid for that filter once at pull time. Restating it charges you a second time, in rows you never see, and the shrinkage shows up as a list that ran dry a month early.

So the gate never mentions company size or revenue. It asks a different kind of question entirely: does this person sit at a desk and own the problem we solve. The desk part sounds trivial and it is not. A buyer who works a computer as part of the job reads email, replies to email, and can join a recorded conversation from a quiet room. A buyer who spends the day on a job site does none of those things, regardless of how well the company matches on paper.

The rest of the row is context for that judgment, which is why enrichment runs before the gate rather than after it, and why researching a guest happens after the verdict, not before it.

What Is the Desk Access Test, and Why Did It Replace the Industry Kill List?

We used to keep a list of industries that were not allowed to receive an invite. It came from pattern matching on bad recordings and it grew every time a slot went sideways. In August 2026 we deleted the whole list and replaced it with one question: does this buyer work a computer as part of the job.

That is the desk access test. It is not a judgment about the work and it is not about collar color. It is a reachability question, and an invite motion lives or dies on reachability.

Run it forward and the logic shows itself. A buyer at a desk reads email during the workday, replies from a keyboard, and can join a recorded conversation from a quiet room with a working camera. A buyer on a roof, in a truck, or on a plant floor does none of those 3 reliably, no matter how cleanly the company matched on headcount, revenue, or vertical.

The industry kill list was a cruder version of the same test, and it failed in both directions at once. It deleted real buyers because a peer in their category once wasted a slot, and it never caught the desk problem inside an approved category, which is where the problem actually lives. A marketing agency clears any industry filter you write. The field services division inside one does not clear the desk test, and the industry filter will never tell you that.

So the gate prose now carries an explicit instruction in the other direction: never grade a company down for its industry, its vertical, or the kind of work it sells. Branding, creative, design, public relations, content, video, production. Every one of those sells to other businesses at high ticket, with buyers who answer email, which is the only thing the gate is allowed to care about.

There is a second layer to this when what you sell is itself a client acquisition service. Then the gate has to read one step past the guest, because the engagement only works if the guest's own buyers can be invited onto a show too. A consultancy that sells exclusively into public school districts is a strong company that we cannot deliver for, and that is a gate decision rather than a sales decision. Make it on the list, in writing, before anybody spends a conversation finding out.

What Are the 5 Checks Every Name Has to Pass?

In the order they should run, cheapest first.

1. Do they own the problem you solve. Not adjacent to it, not interested in it. It is their number, their team, or their quarter that suffers when the problem goes unsolved.
2. Can they decide on spend. Either they sign, or they are the person whose recommendation ends the conversation. Anyone who has to build an internal case from scratch is a longer road than a guest seat can carry.
3. Do they sit at a desk. Reachable by email during the workday, able to join a recording without leaving a site. This single check replaces every industry blacklist we used to keep.
4. Is the company visibly operating. A live site, recent activity, a real team page. Dead companies are the largest source of silent list decay, and they never announce themselves.
5. Would you take the sales conversation if it booked tomorrow. The honest one. If the answer is no, the invite is a favor you are doing your calendar, and the recording will confirm it 45 minutes at a time.

Check 5 is the one people skip, and it is the only one that catches a name that passed the first 4 on technicalities. Run it last, run it every time, and the rest of the funnel gets easier. Qualifying guests before you invite them covers the mechanics underneath each check.

What Rubric Does the Grader Actually Run?

Two stages, ordered by what they cost. Free and deterministic first, paid and probabilistic second, and nothing expensive ever runs before something cheap has had a chance to kill the row.

Stage 1 is a title kill list. String matching on the title column, no model call, no credits. Coordinator, junior, intern, and assistant die here. It runs before email verification, so a name that was never going to buy never burns a verification credit and never reaches the grader at all.

That stage taught us the most expensive lesson in the file. We once had the word "coach" on the kill list. On one 22,103 row consulting pull it killed 267 rows, among them "Founder And Coach", "CEO & Executive Coach", and "Executive Coach, Owner". Those were the exact buyers the pull was built for.

The fix was a rule rather than a deletion: an owner marker outranks a soft kill word. Founder, owner, CEO, president, principal, managing partner. A bare "Career Coach" still dies. "Founder & Coach" lives. Intern never lives, because an intern is an intern.

Stage 2 is one model call per batch of names, grading each row 1 to 10 against the ICP written in prose. Four bands, and the band boundaries are the part worth stealing.

The floor sat at 6 until July 2026. We dropped it to 5 after reading the band back by hand. On one consulting pull the score 5 rows were 6 of 94 names, and every one of them was a real consultancy the grader had marked down for a weak high ticket signal rather than for being out of profile. Companies that are genuinely out of profile score 2 to 4. The floor was deleting a band of buyers to avoid a problem that lived a tier lower.

One line in the grading instruction does more work than the bands do: judge only from the fields given, and treat a missing field as neutral rather than disqualifying. Without it, a row with thin enrichment text reads as a rejection, and the names with the least public surface area are often the owner operators who reply the most.

The order of operations around the grade matters as much as the grade. Title kill list, then verification, then the grade, then personalization at scale. Verification before grading means we never pay to grade an address that bounces. Grading before enrichment means we never pay to research a name that was never getting an invite.

One more lesson from that file, because it is the exact failure this article is about. List tools report headcount as a band rather than a number, and "1-10" is the most common band by a wide margin. Parsing the bottom of that band and applying a floor of 5 deleted every row inside it, roughly 45% of a nightly batch. Nothing broke. The list just looked smaller, and smaller looks like selectivity. Reading the top of the band fixed it. Scaling past 10,000 invites a month is mostly a story about catching errors of that shape before they compound.

Which Names Should the Gate Disqualify Outright?

Most of the gate is a judgment call graded on a scale. A short list is not. These rejections are hard, they run ahead of the grade, and a strong company on every other axis does not overturn them.

• Non buyer titles. Intern, assistant, coordinator, junior. Free to catch, caught before any spend, and never worth a second look.
• Buyers who sell to government. Federal, state, county, municipal, public school districts. The procurement cycle does not care that you had a nice conversation.
• Buyers who sell to nonprofits, NGOs, or philanthropies. Good work, wrong budget shape for a high ticket engagement.
• Buyers who sell to consumers. Homeowners, job seekers, individuals and families. A recorded conversation with a consumer brand's founder is a content win and a revenue dead end.
• Patient facing healthcare providers. A clinic, a dental practice, a therapy practice, a home care operator. A vendor that sells to those practices is a clean fit, because its buyers are the executives running them. Read the description closely enough to tell the two apart, because the industry tag will not.
• Companies that are not visibly operating. No live site, no recent activity, no team page. Dead companies are the quietest source of list decay and they never announce themselves.

Notice what is missing from that list. No industries, no revenue bands, no headcount floors, no title string match on the seniority you were hoping for. Each disqualifier names a structural reason the person or their buyers cannot complete the journey the invite is for. That is the only kind of rule that earns a hard reject.

The cost of a soft disqualifier being too broad is worse than most operators expect, because the names it deletes never complain. Before you add one, write down the buyer it would have killed on last month's list, and go look at whether that name would have been a fit. Inviting C level executives and why executives say yes both cover who tends to survive a gate like this, and it is rarely who the org chart suggests.

Which Filters Should You Delete From Your Gate Today?

Most gates carry inherited filters that feel responsible and do damage. These 4 are the ones we removed and never rebuilt.

The follower count one deserves a sentence on its own. The seat you are handing out is the scarce asset, not the guest's reach. Podchaser's guest pitching data puts the median acceptance rate at 5% when publicists pitch shows for a slot, which is what the market looks like from the other side of the table. You are holding the thing everyone is chasing, so spend it on someone who can buy, not someone who can broadcast.

What Does One Bad Fit Cost You After the Invite Goes Out?

A loose gate does not fail loudly. It fails as 4 separate costs that each get charged to a different account, which is why nobody adds them up.

The slot. Gartner puts the average B2B buying group at 11 people and the share of buying time spent with all potential suppliers combined at 17%. A recording is 40 to 45 minutes of undivided attention with one member of that group, on the record, with their guard down. No other channel hands you that. Harvard Business Review describes the modern purchase as carrying a swelling raft of stakeholders, and the slot is your one shot at a real one.

The relationship with the buyer you wanted. Buying behavior research collected by Corporate Visions found that 73% of B2B buyers avoid suppliers who send them irrelevant outreach. An invite aimed at the wrong person is still outreach. Send enough of them and you are not filling the calendar with the wrong guests, you are removing yourself from consideration with the right ones.

The domain. Google's bulk sender guidelines tell senders to keep the reported spam rate under 0.10% and never to reach 0.30%. At 10,000 sends a month, 30 complaints puts you at the ceiling. An ungraded list burns that budget from both ends at once: the wrong people report the mail, and dead mailboxes bounce. Both signals land on the domain carrying every future invite, which is why domain reputation, spam placement, and bounce rate are gate problems before they are deliverability problems. Rebuilding from there means new domains and another warmup cycle, which costs weeks you were not planning to spend.

The week. Salesforce research found reps spend only 28% of their week actually selling. On an acquisition show the recording is the selling. A slot given to someone who cannot buy does not waste 40 minutes of content time, it spends the scarcest block in the week on the one activity the whole motion exists to protect.

Two more numbers make the arithmetic worse. ZoomInfo's work on B2B data decay puts job title turnover at 25% to 35% a year, so a fit you graded 6 months ago is partly fiction today. And 6sense's Buyer Experience Report found that 94% of buying groups order their shortlist before contacting a seller, with the vendor ranked first winning roughly 80% of the time. The recording is one of the few plays that lands you on that list before the ranking happens. Spending it on a non buyer forfeits the only early entry you get.

Seniority changes the math too. Across 7,530,489 emails, Belkins measured founders and owners replying at 0.57%, C level executives at 0.42%, and VPs at 0.32%. Founders answer roughly 1.8 times as often as VPs on the same campaigns, and they are also the ones who can say yes without a committee. A gate tuned toward owner operators buys you reply rate and cycle time in the same decision, which is the pattern behind getting decision makers to reply.

How Do You Run the Gate on 10,000 Names?

One model call per row, reading the enrichment text already sitting on that row. It returns a verdict and one line of reasoning. Nothing more elaborate is needed, and every extra layer we have tried added spend without moving the verdict.

1. Write the gate in one file. One paragraph of ideal buyer prose, no numbers in it. Version it, date every change, and treat an edit as a decision rather than a tidy up.
2. Grade in a single call per name. Pass or fail, plus the reason in the model's own words. The reason is what makes the gate auditable 3 weeks later when somebody asks why a name was dropped.
3. Append the verdict, never overwrite the row. New columns on the right of the file, original data untouched. A gate that destroys its input cannot be re-run when the criteria change.
4. Hand audit 50 rows a week. 25 passes and 25 rejects. If you disagree with more than 5, the gate prose drifted from what you actually want, and the prose is what needs the edit.
5. Re-run the borderline rows. Grading is probabilistic, so a name at the edge can land either way on a given pass. Re-running the near misses recovers real buyers at almost no expense.

The list tooling underneath this is ordinary. Clay against Apollo covers what each does well, and building a list from scratch covers the pull that feeds the gate in the first place.

How Do You Tell If the Gate Is Set Right?

Watch 3 numbers and cohort every one of them by gate verdict.

A pass rate over 60% means the gate is descriptive prose rather than a rule, and it is approving names on general resemblance. A pass rate under 15% usually means the pull query was wrong upstream, and the gate is quietly doing work that belongs in list segmentation.

Reply rate tells you whether the invite landed with someone it should have landed with. Apollo calls 3% to 5% a realistic baseline for a well run campaign and anything under 2% a list or deliverability problem rather than a wording problem, and lemlist's breakdown puts agencies and consulting at 4% to 7%, which is the healthiest band in their data. If you are inviting into that market and sitting below it, the gate is the first place to look, not the subject line. The reply rate benchmarks break that down per market.

The sharper read is the one number most shows never cohort. Take recordings from passing names and recordings from names you waved through, and compare how many of each produced a sales conversation. If replies look fine but almost nothing converts past the recording, the gate is passing the wrong seniority, and every surface metric will look healthy the entire time it happens. Guest to client conversion rate is where that shows up first.

False negatives are the expensive half and they are invisible by design. Nobody complains about a buyer you never invited. Pull 100 rejects a month, read the reasons the grader wrote down, and look for a pattern the gate is not built to see yet. The "coach" incident and the headcount band both surfaced exactly this way, and both were a week of reading rejects rather than anything clever.

Two tells say the gate drifted rather than tightened. The first is a rejection reason you disagree with twice in the same audit, which means the prose says something you did not intend. The second is a pass rate that moved without an edit, which means the input list changed and the gate is now grading a different population than the one it was written for. Re-read the prose against 50 fresh rows whenever a new list source comes online.

The volume math sits on top of all of this. How many invites it takes to book one recording tells you how much a narrower list has to earn back, and guest no show rate tells you how much of what you booked survives to the room.

The Practitioner Take on ICP Gates

A guest list is a revenue decision that looks like a data task, which is why it gets handed to whoever has the afternoon open. The gate is where you decide, once, in writing, who is worth 45 minutes of your show and the sales conversation that follows it.

Shows that never write the gate down end up with a calendar full of pleasant recordings and nothing to sell into. The episodes are fine. The guests are interesting. Not one of them could have bought anything.

We measure client work in recorded conversations for this reason. 30 recorded conversations with your ideal buyers in 90 days, or your money back, where ideal buyer is the phrase the gate is holding up. Drop the gate and that number is still reachable, and it stops meaning anything, which is the version of this failure nobody catches until quarter end.

Write the gate before you write the invite. The copy can only reach the people your list decided to reach, and by then every decision that mattered has already been made.