The playbook, stated fairly
The clearest recent statement of it came from Swulinski on 20VC's 20Growth, and the argument runs like this: every impression should be tied to an action, you need hundreds of creative variations rather than a handful of good ones, you should be spending on day one rather than waiting, and the whole thing should be measured properly before a penny goes out.
He came to it from ecommerce and applied it at two products people have actually heard of. Everything below is a response to that argument rather than an alternative to it.
Ecommerce spent a decade building that operating model out of creator programmes and enormous creative volume. The claim is that AI companies should run the same system. We agree with essentially all of it, which may be surprising on a site that sells the other thing.
Why it is right
The strongest part of the argument is about speed of learning rather than about acquisition.
A week of paid spend answers questions that take a quarter to answer any other way. Which claim gets a response.
Which framing people repeat. Whether the landing page converts anyone who arrives cold.
Whether the thing you think is the headline is the thing that stops a stranger. You can run content, do interviews and go on podcasts, and you should, and none of it produces a signal that fast.
So the common advice to hold off on paid until the brand is established has the sequence backwards. Paid is how you find out whether the positioning survives contact with somebody who has never heard of you, and doing it early is cheaper than discovering it late.
The second part worth taking seriously is the creative volume. A handful of assets is not a campaign. Platforms with algorithmic targeting read the creative to decide who to show it to, which means variety is the targeting, and a set of near-identical assets tells the system almost nothing.
The three channels it runs on
Ask anyone running this system what the core stack is and the answer is consistent.
| Channel | What it does |
|---|---|
| Meta | High-volume video intent. Creative is the targeting, so the whole job is creative strategy. |
| Search intent, plus YouTube for education. Reaches people who do not know you exist but are describing the problem. | |
| Lifecycle | Email, SMS, push. The net that catches everything the first two send. |
That is the list. Add channels once those three are working, and adding ten at once means running ten badly.
The one that is always absent
Read that list again and notice what is not on it.
X is where the AI conversation happens. It is where launches trend, where the founders and engineers and investors in this category spend their attention, and where every AI company wants to be seen. It appears on nobody's paid channel list.
That absence is not an oversight and it is not about company size or deal value. Asked directly, Swulinski said he has yet to meet a head of growth or performance marketer who tells him X ads print, and invited anyone who has cracked it to correct him.
That is a person running seven-figure monthly budgets across Meta and Google, with no reason to talk the platform down.
His observation is the useful half. What we would add is the mechanism, because there is one, and it explains why the invitation keeps going unanswered.
We have the same experience and a structural explanation for it. X sells a subscription that reduces or removes advertising, and the people who buy it are disproportionately the people you are trying to reach. The budget still spends.
It reaches whoever is left, which is the part of the audience least likely to be evaluating technical products. The mechanism in full.
What X takes instead
Once paid is off the table on one platform, the question stops being paid versus organic and becomes narrower. A subscription removes advertising. It does not remove the accounts somebody chose to follow.
So there are exactly two routes in, and both of them are a followed account.
- Become one of those accounts. A founder posts consistently for years, builds an audience of the right people, and ends up as somebody your buyers already follow. This is the better outcome and it takes years, and it will not help the launch you are shipping this quarter. Why it outperforms, and where it breaks.
- Go through people who already did it. The trust was earned by somebody else over years of posting and is lent to you for a launch window. That is a purchase you can make in three to ten days, usually five. How a campaign runs.
They are the same mechanism. The only difference is whose account carries it and how long you waited to have one. The decision between them.
Two different creative economies
The volume logic that makes the paid playbook work does not transfer to X unchanged, and the difference is worth understanding before you budget for it.
| Meta and Google | X, through creators | |
|---|---|---|
| Volume needed | Hundreds of new assets a month to avoid plateau | One post per creator, across a bench of the right size |
| Why variety matters | The algorithm reads creative to decide targeting | Identical copy across accounts reads as a manufactured push and gets punished |
| Who writes it | Creators produce assets to a brand brief | Creators write in their own voice or the value evaporates |
| What is being bought | Impressions, priced on a CPM | A claim carried by somebody with their own reputation at stake |
Both systems want variety and they want it for opposite reasons. On paid, variety feeds a targeting algorithm.
On X, variety is the evidence that these are real opinions rather than one message distributed eleven times. Why we brief facts and never sentences.
How to split the budget
The useful framing is that these are different jobs rather than competing options.
Run Meta and Google for acquisition and for learning. Set up conversion tracking properly before spending anything, because a campaign optimised against a broken signal will teach you that paid does not work when what you actually proved is that your tracking does not. What to measure.
Run X for the moments that need to be seen. Launches, funding, the thing you want the category talking about.
That is a per-launch purchase rather than an always-on channel, and it is priced accordingly. What that costs.
Start the founder account regardless. It is the cheapest asset on the list, it is the one with a start date you control, and in three years it is the reason a launch costs less to move. Forty minutes a day.
The mistake worth avoiding is treating X as a paid channel that underperforms and pushing more budget at it. It is not underperforming. The inventory your buyer would have seen is not for sale, and the two routes that remain are bought differently.
Questions we get asked
Should an AI company run paid ads from day one?
On Meta and Google, yes. Paid is the fastest way to test messaging, positioning and funnels, and a week of spend answers questions that take months of content and interviews. The advice to wait on paid until the brand is built is usually wrong, and the people running the largest budgets in AI say so plainly.
Why is X missing from the standard paid channel list?
Because the platform sells a subscription that reduces advertising, and the people who buy it are disproportionately the technical audience most AI companies are trying to reach. Practitioners running seven-figure monthly budgets across Meta and Google consistently report that X ads do not perform, and the mechanism explains why.
So how do you reach anyone on X?
Two routes, and both are a followed account. Either you become one of the accounts your buyers already follow, which takes a founder posting consistently for years, or you go through people who already did it. A subscription removes advertising and does not remove the accounts somebody chose to follow.