Our last piece argued that distribution problems are usually channel problems. This raises the obvious question: which channels should you choose in 2026?
Here is what the people actually working on this are saying, and what each piece of advice means in practice.
Buyers are starting somewhere you don’t control
Wynter’s third annual CMO buying survey covered 101 mid-market B2B SaaS CMOs in January 2026. 84% now use LLMs somewhere in vendor discovery, up from 24% a year earlier and effectively zero the year before that. 68% start a category search inside an AI tool before a search engine.
In practice: ask ChatGPT, Claude and Perplexity for the best tool in your category. Whether you appear, and whether what they say about you is accurate, is now a distribution metric. Most companies have never checked it.
The part usually left out: the same survey found 65% of those CMOs also start in peer communities, 42% rank word of mouth as their single biggest influence, and cold outreach came last, with 2% ranking it first. Sample of 101 at companies above $50M in revenue, so treat the trajectory as the finding and the percentages as directional.
Agents are already in the buying process
Kyle Poyar’s essay is titled “Your next customer might be an AI agent, whether you’re ready or not”. Agents aren’t making autonomous purchases yet, but the infrastructure is being built: Ramp’s agent cards, verification standards from Mastercard and Google, Stripe’s agentic commerce protocol.
The point isn’t really about who signs the contract. It’s that you’ve spent years writing for humans, and humans have stopped reading. They ask something else, and that something else goes looking. If it can’t find an answer, it comes back with nothing, and the buyer loses interest before they ever reach you.
We tested this on a well-known SEO platform, asking an AI to find its enterprise pricing. It couldn’t, because the company doesn’t publish it.
What came back instead: average annual spend figures from procurement data brokers, contradictory accounts of what the current tiers even are, add-on prices sourced entirely from aggregators, and one third-party page actively advising buyers to skip the enterprise tier because the cost gap rarely justifies itself below a certain scale.
That last one is now part of the answer a buyer gets. The company had no say in it.
This isn’t a criticism of anyone. Gating enterprise pricing was reasonable when a person did the research. They’d hit the wall, fill in the form, and a salesperson would get the chance to build the case. That still works. It just happens later now, and only if the buyer initiates it. If the shortlist forms before anyone makes contact, the best sales team in the world never gets the meeting.
In practice: about half of B2B tech companies still gate pricing behind a contact form. You don’t have to publish a number you’d rather not publish. But publish a starting point, publish what moves the figure, and make the contact path completable without a phone call. If you leave the space empty, someone else fills it.
Your product may already have non-human users
Wes Bush frames the shift in three stages: “PLG 1.0 = user-led. PLG 2.0 = agentic. PLG 3.0 = headless.” His test for the third stage is whether an agent can use your product with no human in the loop. He already puts Netlify there, citing roughly 80% of signups coming from agents. Netlify’s co-founder has said publicly that the majority of new signups are now agents.
In practice: find out what share of your own signups are agents. Standard analytics can’t tell the difference, so most companies run that channel blind.
Whatever channel you pick is on loan
Brian Balfour’s The Next Great Distribution Shift describes a cycle every platform runs: identify a moat, open the gates so developers and creators build it for free, then close and monetize once it can’t be displaced. Facebook did it in about two years, Apple in four, Google over two decades, LinkedIn in under four. His harder claim is that AI has so far destroyed distribution channels rather than creating them, and that ChatGPT is “a destination, not a distribution channel.”
In practice: budget for the channel closing. Anything you build on rented ground should feed something you own.
What these have in common
There’s a hierarchy in all of this, and it’s worth naming.
Peer communities are the layer you can’t touch at all. Someone asks a private Slack what they use, a member answers from a support ticket they had eight months ago, and you never know it happened.
Published chatter is the layer you can influence but not control. A comparison thread, a Reddit answer, a review left in 2024. This is the one that matters most, because it works on both audiences at once: a buyer lands on that thread when they search, and a model ingests the same thread and builds its picture of you from it. Same artifact, two readers, neither of whom checked with you.
That’s what the pricing test actually showed. The answer wasn’t wrong, exactly. It was assembled entirely from third parties, including one telling buyers not to bother. Answer engines and agents sit downstream of that layer, largely relaying what other people already wrote about you.
Which changes what there is to optimize. Not persuasion. What’s on the record about you. Every one of these systems works from what’s already written, so the only real question is whether the answer they find came from you or from someone else.
Siege Media has a practical piece on auditing and shaping how AI describes your brand, including a protocol for running buyer-intent prompts across ChatGPT, Perplexity, and Gemini and tracing which third-party sources drive the answers, and there’s a lot more out there worth reading.
