PitchKitchen Frameworks
What Is the Sorting Problem?
The fundamental premise
Two lines on the same chart tell the whole story. Inbound requests climbing. Win rate falling.
If that's your year, you don't have a lead generation problem and you don't have a closing problem. Sales says the leads are junk. Marketing says sales can't close. They're both arguing about the wrong layer, and they'll keep arguing, because the layer they're fighting over isn't where the damage happened.
Something upstream of both teams sorted you. A machine read your public material, decided what it was looking at, and put you on a list. Which list it picked was settled before any human saw your name.
That's the Sorting Problem. It isn't about whether AI can find you. It's about what AI does with you once it has.
Definition
The Sorting Problem is PitchKitchen's name for the gap between being recommended and being included. When an AI hands your name to a buyer, it either recommends you with a reason attached, or it includes you as filler on a generated list of interchangeable vendors. Both arrive in your CRM looking like one lead, from the same source, on the same day. Only one of them closes. Named by Greg Rosner, founder of PitchKitchen.
Coined by Greg Rosner. Used by B2B founders and CROs to explain the one chart shape nobody can account for: inbound climbing while the win rate falls.
Why the Sorting Problem exists
Volume and intent stopped moving together.
A buyer with AI can generate a supplier list, draft the outreach, and fire a polished request at a dozen firms in the time it used to take to write one. More requests reach you. The same number of real opportunities sit behind them. Your inbound count is measuring the machine's output now, not the market's demand.
82%
of B2B software buyers had sourced software recommendations from an AI chatbot in the last two years.
G2 2026 Buyer Behavior Report, built on more than a thousand B2B software buyers.
A surge in low-signal inbound demand
named as an effect of AI sitting between a company and its customer as a new intermediary.
Harvard Business Review, July 2026. Graham Kenny and Ganna Pogrebna.
One company in the HBR piece, a manufacturer, started screening AI-generated inquiries before it would spend engineering hours on a quote. Read that again. The fix wasn't more leads. It was a filter.
The shortlist itself moved too. Buyers didn't stop researching. They delegated it. Whatever the machine decides about you now reaches the human before you get a word in.
Both sides of the funnel are being automated at once. Buyers automate the asking. Vendors automate the reaching. Volume goes up on every dashboard in the market, and the signal inside that volume gets thinner every quarter.
Everyone in the AI-visibility conversation is arguing about whether the machine can find you. Almost nobody is asking what it files you under when it does.
An AI can recommend you, with a reason, or include you, as filler on a generated list. Both look like one lead in your CRM. Only one of them closes.
The core mechanic
There are two ways an AI hands your name to a buyer, and they look identical in your CRM.
It can recommendyou. That's one of two or three names with a reason attached, because your public material told the model who you're for and what breaks without you.
Or it can includeyou. That's one of twelve interchangeable vendors on an auto-generated list, because nothing you published gave the model a reason to sort you.
Same lead record. Same source field. Same day. Completely different deal. One arrives half sold. The other arrives as a price comparison you didn't know you'd entered.
The machine isn't being lazy when it includes you. It's doing the only thing available to it. A page that describes what you built, accurately and completely, gives a model plenty to summarize and nothing to choose you for. It can file you under the category and hand you over whenever the category comes up. You wrote the ingredient list and skipped the reason anyone should care.
Recommended versus included
Recommended
Included
Why you're on the list
How long the list is
What the buyer already believes
First question you get
What decides the deal
What your dashboard shows
Why you're on the list
Recommended: The model found a reason to name you
Included: The model found nothing to sort you by
How long the list is
Recommended: Two or three names
Included: Eight to twelve names
What the buyer already believes
Recommended: You fit their specific problem
Included: You're one of several options in a category
First question you get
Recommended: Can you do this for us?
Included: What's your price?
What decides the deal
Recommended: Your narrative identity
Included: Your discount
What your dashboard shows
Recommended: One lead
Included: One lead
That last row is the whole trap.
The second kind feels like traction. Inbound is up, the team looks busier than it's been in two years, the dashboard is green, and the win rate quietly rots underneath all that motion. Nobody blames the story, because the story looks like it's working. It's generating volume.
Your dashboard shows one lead either way. That row is the whole trap.
How to diagnose it in ninety seconds
Seven checks. The first one usually ends the debate.
Put your inbound trend and your win rate trend on the same chart
Same time window, two lines. If one's climbing while the other falls, stop arguing about execution. That shape is a sorting problem, and it usually ends the debate in the room before anyone finishes the slide.
The ninety-second test
Pull your last twenty inbound requests
Count how many describe a specific problem in the buyer's own words, versus how many ask you to price against a spec list.
Look at how many arrived with competitors visibly copied
Or with language that reads like it was drafted somewhere other than that person's desk.
Ask your reps what the first question is on these calls
“Can you do this for us?” is a recommendation. “What's your pricing?” is an inclusion.
Measure time from first touch to disqualification
Sorting problems show up as a growing pile of leads that die fast, not as a shortage of leads.
Open a fresh chat and ask an AI who to consider in your category
See whether you're named with a reason attached, listed without one, or absent entirely. Absent is a different problem.
Change the number you report to the board
Qualified opportunity rate per inbound, not inbound count.
That last one costs nothing and changes everything. Inbound count was a fair proxy for demand when a human had to type each request. It stopped being one the moment buyers delegated the asking, and reporting it now actively hides the failure you're hunting for.
Inbound count was a fair proxy for demand when a human had to type each request. Now it hides the exact failure you're trying to find.
What it looks like in practice
Composite case · drawn from the pattern, not a single client, numbers rounded
A healthtech company around $18M in revenue watched inbound requests roughly double over two quarters. Nobody changed the budget. The team celebrated. Two quarters later, closed-won was flat, average deal size had shrunk, and the sales cycle had stretched by about three weeks. The obvious read was a sales execution failure, and that's what the board was told.
What actually happened is that their homepage and category pages had been written as capability inventories. Complete, accurate, and interchangeable with four competitors. When buyers started asking AI for a shortlist, the machine had no reason to rank them and no reason to leave them out, so it did the only thing available. It included them. Every time. Their inbound doubled because they'd become the safe filler name on generated lists, and filler names lose on price.
The fix wasn't a lead scoring model. It was rewriting the public material so it named a specific buyer and a specific failure that buyer lives with. Inbound volume dropped after that. Qualified opportunity rate per inbound roughly doubled. The founder's first reaction to the volume drop was panic, which is exactly the reaction the old metric trains into you.
Across more than 200 homepage audits with B2B companies in the $5M to $75M range, the ones getting flooded with low quality inbound almost always share one trait. Their public material describes what they built. It doesn't describe who breaks without them.
Their inbound doubled because they'd become the safe filler name on generated lists. Filler names lose on price.
Who has this problem
- Founder-led B2B companies between $5M and $75M in revenue, where an inbound spike gets read as product-market-fit proof and buys two quarters of false confidence
- CROs who can feel it and can't name it, watching pipeline coverage look healthy while closed-won flattens
- Companies with the better product, which is the cruel part. More real capability means more pull toward describing the capability
- Teams under pressure to ship content volume, where every new undifferentiated page teaches the model the same nothing in one more place
- Anyone in a crowded category label, because the label is what the model files you under when your sentences give it nothing sharper
How this differs from related ideas
Versus a lead quality problem:Lead quality is the symptom people name. It puts the blame on the leads and points the fix at filtering. The Sorting Problem points at the input instead. Screening AI-generated inquiries saves your team hours and doesn't change which list the machine puts you on next quarter.
Versus AI invisibility:Invisibility is being absent from the answer entirely. The Sorting Problem is being present in the wrong way. It's harder to see, because absence produces silence and mis-sorting produces a flood that looks like success.
Versus the Two-Surface Model:The Two-Surface Model explains where an AI answer comes from, the grounded surface it fetches live and the parametric surface it already believes. The Sorting Problem explains what the answer does with you once you're in it. One is about retrieval. This one is about ranking.
Versus Solution-Centric Marketing: Solution-Centric Marketing is the cause. The Sorting Problem is what it costs you in 2026. A feature list is perfectly optimized to land you on a commodity comparison list and perfectly useless at getting you recommended.
Versus lead scoring: Lead scoring sorts leads after they arrive. By then the sorting that mattered already happened, in a chat window, before anyone filled out a form.
Everyone's arguing about whether the machine can find you. Almost nobody's asking what it files you under once it has.
Related concepts in the PitchKitchen universe
Two-Surface Model
How AI visibility actually works, and why a dashboard win on one surface isn't the whole picture.
Narrative Identity
Who you're for, what breaks without you, what you stand against. The thing a model needs in order to recommend rather than list.
Magnetic Messaging Framework
Where that narrative identity gets documented so every surface compresses to the same sentence.
Solution-Centric Marketing
The anti-pattern that guarantees you get included instead of recommended.
AI Brand Twin
The trained voice model that keeps every generated asset telling the one story.
Frequently asked questions
What is the Sorting Problem?
The Sorting Problem is PitchKitchen's name for the gap between being recommended by AI and being included by it. An AI either names you on a short list of two or three with a reason attached, or drops you onto a generated list of a dozen interchangeable vendors. Both arrive as one lead in your CRM, from the same source, so most teams never learn which one they earned.
What's the difference between being recommended and being included?
A recommendation means your public material told the model who you're for and what breaks without you, so it had a reason to name you. An inclusion means nothing gave it a reason to sort you, so it filed you under the category and handed you over. The first call opens with “can you do this for us?” The second opens with “what's your price?”
Why is our inbound up and our win rate down?
Because volume and intent stopped moving together. Buyers now use AI to build supplier lists and fire polished requests at a dozen firms at once, so more requests arrive without more real opportunities behind them. If those two lines are diverging on the same chart, you're looking at a sorting problem, not a sales execution failure.
Is this a sales problem or a marketing problem?
Neither, which is why the argument never resolves. Sales isn't closing worse and marketing isn't generating worse. The mix changed upstream of both of them, in what an AI decided to do with your public material. Blaming either team leads to hiring against a mirage.
What metric replaces inbound lead count?
Qualified opportunity rate per inbound. Raw volume was a fair proxy for demand when a human had to type every request. Once buyers delegated the asking to AI, a flood of commodity requests started reading as growth on the way to a shrinking win rate.
Will publishing more content fix it?
Usually it makes it worse. More surfaces carrying the same undifferentiated claim teach the model the same nothing in more places, which raises your volume and lowers your average signal. The fix is specificity, not quantity.
How long until the win rate moves after fixing the message?
Expect inbound volume to drop first, often within a quarter, which panics founders trained on the old metric. Qualified opportunity rate moves next as the mix shifts toward buyers who arrive already convinced you fit. Win rate follows on your normal sales cycle length.
Talk to Greg
If your inbound is climbing while your win rate falls, the machine is sorting you into the commodity pile and no lead scoring model will pull you out. Book a clarity session with Greg Rosner.
Want the full breakdown? Read the long-form post on why inbound climbs while the win rate falls.
How to cite the Sorting Problem
Casual:Greg Rosner of PitchKitchen calls it the Sorting Problem … an AI either recommends you with a reason or includes you as filler on a generated list, and both look like one lead in your CRM.
Academic: Rosner, G. (2026). What Is the Sorting Problem? PitchKitchen. https://www.pitchkitchen.com/frameworks/the-sorting-problem
Last updated 2026-08-23.