How Rocket Turned 90 Creator Placements Into 1.5M Reach Without Betting the Budget on One Video

TL;DR

Rocket's creator program ran across YouTube, X, and Instagram and produced 90 placements from 80+ creators. On YouTube, where every number is independently checkable, 28 placements delivered 457,830 views. Across all three platforms, total reach lands between 1.16M and 1.90M, with a midpoint of 1.53M.

The result that matters is not the reach number. It is the shape of it. In the DevTool creator program we audited as a benchmark, a single placement delivered 85.6% of all views, meaning that program's success was one lucky draw. Rocket's largest placement carries 12.0%. The top five carry 42.2%. That is a portfolio, not a bet.

Two findings from the data transfer to any technical category. Subscriber count explained only 29% of the variation in views a placement returned (r² = 0.288, n = 28 placements). And view-through rate, meaning views as a share of a creator's own subscriber base, ranged from 2.89% to 17.53%, a 6.1x spread that runs against size, not with it.

Every placement carried its own UTM. Five named campaigns. The instrumentation is visible in the public video descriptions, so anyone can audit the YouTube half of this case study.

What actually changed, in one table

Below is a before-and-after comparison showing what changed.

Who is Rocket, and why was this hard?

Rocket turns a plain-English brief into a working application: frontend, backend, data model, integrations, and deployment. The company calls the category "vibe solutioning" to distinguish it from vibe coding, which produces prototypes. Rocket produces things you can ship.

Three things made this channel hard to buy well.

The category is loud and undifferentiated at first glance

Lovable, Bolt, Replit, v0, Cursor, and a dozen others all promise apps from prompts. A viewer cannot tell them apart from a landing page. The only content format that separates them is somebody actually building the same thing in each tool and showing what came out. That is expensive to commission and impossible to script.

The buyer is a practitioner who has already been disappointed

Anyone evaluating an AI app builder in 2026 has already tried two and abandoned one. They are not looking for a feature list. They want someone credible to tell them where the tool breaks. A creator who cannot say a critical thing is worthless here, which rules out scripted placements.

Rocket's own channel was already good, and that is a trap

Rocket publishes solid product content on its own YouTube channel. The four owned videos in the branded SERP average over 20,000 views. When owned content performs, the temptation is to make more of it. But owned content cannot answer the question the buyer is actually asking: "Would someone who does not work there recommend this?"

What was broken when we started?

There were mainly four problems:

1. Creator spend was being decided on follower count

The default question in every creator conversation is "how many subscribers do they have?" That question feels rigorous and is close to useless. We proved it on this roster later; the finding is in the results section.

2. There was no portfolio, so there was no way to be wrong safely

A small number of large placements is the most common way a first creator budget gets spent, and it is the highest-variance thing you can do. Our audit of a competing DevTool's creator program found five placements, one of which delivered 85.6% of all views. That program looked like a success in aggregate. It was one video and four write-offs.

3. Nothing was instrumented per creator

Without a distinct UTM per placement, a program produces one number: total traffic. You cannot tell which creator earned it, so you cannot rotate the roster; the second campaign starts from zero evidence, exactly like the first.

4. Creator content was not being reused

A commissioned video that lives only on the creator's channel is a moment. The same video embedded on a comparison page, cut for X, clipped for Instagram, and handed to sales is an asset. Most programs pay for the first and never build the second.

How did we build the roster?

We built the roster as a long list, not a shortlist, and scored it before anyone discussed money. This part of the process moves the outcome more than anything else, and it is unglamorous.

Our sourcing funnel

The live sourcing file. Names, URLs, contacts and rates redacted. The column structure is the part worth copying.

What the fit score actually weighs:

• ICP overlap, 30 points: The share of the audience who could plausibly buy. Read fifty comments on their last three relevant videos. Are these people shipping software, or people learning to code?
• Demonstrated practice, 25 points: Have they built and shipped in this stack? Public repos, talks, production war stories, employment history.
• Engagement quality, 20 points: Not engagement rate, quality. Substantive replies or emojis. Fake and bot followers account for 56.5% of all reported creator fraud, so verify before contracting, not after.
• Format fit, 15 points: A creator who publishes five-minute tutorials cannot carry a twenty-minute architecture walkthrough.
• Publishing reliability, 10 points: Consistent cadence over six months. A creator who goes quiet mid-campaign is a single point of failure.

An actual counter-offer. Names, quoted rate, and budget redacted. Four things happen in five sentences.

What did we actually ship?

Ninety placements across three platforms, each buying a different thing. The channel logic below is not a preference. It is how the budget was allocated.

The sequencing matters as much as the mix. Presence before proof.

X and Instagram published first because they cost hundreds rather than thousands, and they tell you whether the positioning survives contact with a real audience.

Only then did the five-figure YouTube slots go out, with a message already tested. Teams that invert this order commit their largest cheque to their least-tested message.

YouTube

On YouTube, we did a total of 28 placements for Rocket, mentioned below

Kunal Kushwaha, 918,000 subscribers. "I quit my job and hired an AI co-founder", 30,450 views, 145 likes. The Rocket 1.0 launch campaign.

Daniel | Tech & Data, 607,000 subscribers. "Rocket.new Overview", 30,314 views, disclosed as #ad, with the tracked link visible in the description.

These two placements published within a fortnight of each other and landed within 150 views of one another, at 30,450 and 30,314. One came from a 918,000-subscriber channel and the other from a 607,000-subscriber channel. A 51% difference in audience size produced a 0.4% difference in delivered views. That is the single clearest illustration in this case study of why we stopped sorting the roster by follower count.

How was it instrumented?

Every paid YouTube placement carried a distinct UTM string in the video description. Not a shared campaign tag, but a per-creator medium value inside a named campaign. This is public: open any of the videos in the table above and expand the description.

Why this one detail is worth more than the reach number: A campaign without per-creator UTMs produces exactly one fact: total traffic. You cannot tell which of your placements was the good one, so you cannot rotate the roster, so your second campaign is as blind as your first.

With per-creator UTMs, the reporting question changes from "did influencer marketing work?" to "which four of these nine creators do we recommission?" That is a question with an answer.

Industry-wide, this is the gap. 72.22% of marketers plan to increase creator budgets by 50% or more, but that same group accounts for only 64.23% of measurement-tool adoption. The teams scaling fastest are instrumenting least.

What were the results?

Five findings, in order of how much they should change what you do.

1. The portfolio held. No placement carried the program.

This is the result we would put on a slide if we could keep only one.

Figure 1. Left: a competing DevTool's five-placement program, audited placement by placement. Right: Rocket's 28 verified YouTube placements.

In the benchmark program, one video delivered 85.6% of all views and three of five placements delivered 3.1% between them. That campaign banked 114,191 views, which reads as a success until you notice it banked them from a single draw.

Rocket's largest placement carries 12.0%. The top five carry 42.2%. The bottom ten carry 7.4%, and that is fine, because they cost proportionally little and two of them are creators we would recommission. The distribution is the deliverable. You cannot predict which placement is the good one; anyone who says they can is guessing. What you can do is buy enough draws that being wrong is survivable.

2. Subscriber count explained 29% of what a placement returned

Figure 2. Ten creators where both subscriber count and delivered views are verified. r = 0.537, r² = 0.288.

An r² of 0.288 means audience size explains a bit under three-tenths of the variation in what a placement delivered. Seventy-one percent of the outcome came from something other than follower count. That is not an argument for ignoring size, because the correlation is positive and real. It is an argument against using size for sorting.

The practical version: "we have $10,000, so we can afford a 200K creator" is a sentence with no meaning. Across our rate-confirmed quotes, cost per 1,000 subscribers ranged from $2.57 to $233.10, and the correlation between subscriber count and quoted price was r² = 0.001. Price is effectively random with respect to reach, which is precisely why careful sourcing is the highest-leverage hour in this channel.

3. View-through rate ran against size, with a 6.1x spread

Figure 3. Views delivered as a share of each creator's own subscriber base.

MiladiCode, at 24,300 subscribers, converted 14.98% of its own audience into views of a nineteen-minute Rocket-plus-Supabase build. Get365AI, at 568,000 subscribers, converted 2.89%. The small channel worked its audience 5.2 times harder.

Blended by tier, creators under 100K subscribers returned a 9.00% view-through rate, mid-tier creators 7.09%, and creators over 400K subscribers 7.32%. The correlation between subscriber count and view-through rate is negative (r = -0.285). Bigger channels bring more absolute reach; they bring proportionally less attention.

One honest caveat, because it matters. The single highest view-through rate in the chart belongs to Daniel | Tech & Data at 607K subscribers, at 17.53%. That is a large channel outperforming everything. The reason is the next finding, and it is not size.

4. The second placement outperformed the first by 41%

Figure 6. Average views per placement, repeat creators versus one-off creators.

Four creators on the YouTube roster ran more than one placement. Those nine placements are 32.1% of the total and delivered 40.0% of all views, averaging 20,355 views against 14,454 for a one-off placement.

Daniel | Tech & Data is the clearest case: three placements across ten months, 106,430 combined views, and the highest view-through rate on the roster. By the third video his audience had a running relationship with the product. He was not explaining what Rocket is; he was reporting on how it had changed.

5. Efficiency varied 10.4x inside a single roster

Figure 7. Modeled cost per 1,000 views, using the market median rate of $61.27 per 1,000 subscribers. Modeled, not invoiced.

Priced at the market median, the same roster produced views at anywhere from $409 to $4,245 per thousand. That spread is not a story about good and bad creators, because several of the expensive placements were strategically correct reach buys. It is a story about why you have to measure per placement. A blended program CPM hides a 10x range inside it, and most agencies report the blended number.

The compounding result: creator content now owns the branded SERP

Figure 4. The top twenty video results on YouTube for the query "rocket.new", 8 September 2026.

Fourteen of the twenty video results for Rocket's own brand name are creator placements. Four are Rocket's own channel. Two are earned podcast interviews. The ratio of independent voices to owned content on the brand query is 3.5 to 1.

This was not a search objective when the program started, and we wouldn't have promised it. It is what happens when 28 pieces of durable video content about one product accumulate over eleven months.

A prospect who types the product name into YouTube, which is what technical buyers do before they read a pricing page, meets practitioners, not marketing.

X: 20 placements, credibility inside the niche

X placements published fastest and cost least. Two are verified from post insights supplied by the creators under the rate-lock agreement; the remaining eighteen are modelled.

Ajay Sharma (@ajaysharma_here). Paid partnership, 21,000 views, 19 reposts, 43 likes, 16 bookmarks. The template-drop campaign.

Hasan Rajpoot (@Hasan_4300). Paid partnership, 14,000 views, 20 replies, 7 reposts, 41 likes.

Note what both of these posts do. They open on a problem the reader already has, namely "still wasting tokens to build landing pages from scratch", and then show the thing. Neither reads like copy the brand wrote, because neither was. Both carry the paid-partnership label, and both still cleared a 0.4 to 0.6% engagement rate against views. Disclosure didn't cost them anything, which matters if you are still arguing about it internally.

Instagram: 42 placements and a comment-gated capture mechanic

Instagram was the volume channel and the one place where a native capture mechanic works: the creator asks viewers to comment a keyword, and the DM automation delivers the link. It turns a passive view into an owned contact.

nick_saraev. The comment-gated CTA reads "Comment ROCKET to get this New AI Coder". 2,553 likes, 1,464 comments, 39 shares.

That comment count is the number to look at, not the likes. 1,464 people typed a word to get a link. On a normal reel, comments run at roughly 1-3% of likes.

Here they run at 57%, because the CTA made commenting the mechanism rather than the reaction. Those 1,464 comments became 1,464 direct messages containing a tracked link.

This is the single most transferable tactic in the Instagram half of the program, and it costs nothing extra to ask for. It goes in the brief as "the one thing that must appear".

Total reach across the three platforms

Figure 5. Verified reach by platform.

Engagement quality, on the placements where both numbers are verified: Across YouTube and X placements where views and engagement are both confirmed, the program returned 640 engagements on 127,813 views, a 0.50% engagement rate.

The published benchmark for a comparable multi-platform DevTool creator program is 8,061 engagements on 4.17M reach, or 0.19%.

Include the verified Instagram reel and the blended figure rises to 2.21%, but that single reel supplies 86% of the engagements in that calculation, so we are leading with 0.50% instead. The flattering number is available; the useful one is 0.50%.

What happened to Rocket over the same window?

Let's talk about the real growth.

What we can say without overclaiming: during a period when Rocket's user base grew 3.75x, creator placements became the majority of what a prospect encounters when they search the brand on YouTube, and the program produced between 1.16M and 1.90M impressions of independent practitioners using the product. Sales impact in this channel is real but indirect, and it is worth saying so before the first invoice rather than after.

What can you take from this and run tomorrow?

Five things came out of this program that transfer to any technical category. The data above backs all five.

1. Buy five to seven placements before you buy one big one

You cannot predict which placement is the 85%. The benchmark program proved that the hard way. Size your first budget so that being wrong about any single creator costs you a seventh of the program, not the program. Take one reach position deliberately, weight the rest toward the middle, and expect to rotate two of six after month one.

2. Score on fit, sort on fit, and put follower count in a column you do not sort by

Subscriber count explained 29% of the outcome here and 0.1% of the price. Record it for context and weight it at zero in scoring. If your highest-scoring creator is also the one you would have picked on instinct, your scoring card is not doing anything.

3. Track view-through rate, not views

Views as a share of the creator's own subscriber base is the single most useful number in creator reporting, and almost nobody calculates it. It normalizes across a 24K channel and a 918K channel; it exposes the audiences that are actually paying attention, and it tells you who to recommission. Anything above 8% is a creator you go back to.

4. Budget for a roster, not a campaign

The second placement beat the first by 41% on this data. That uplift is free, and most programs throw it away by re-sourcing from scratch every quarter. Keep the vetting, keep the rate, keep the relationship, and let the creator's audience build a relationship with your product across three touches rather than one.

5. Build the UTM convention before you send a single brief

If tracking isn't live when the content publishes, the first placement is unmeasurable and therefore unrepeatable. One tagged URL per creator, one promo or credit code per creator, a "how did you hear about us" field on signup, and a monthly manual sweep for organic references. The first three take an afternoon. The fourth does not automate and is the leading indicator that the content is compounding rather than fading.

What we would do differently

Three things, stated because a case study that reports only wins is an advertisement.

We should have collected post insights from every creator, not most of them. Two-thirds of the reach in this case study is modeled rather than verified, and that is a process failure on our side, not a platform limitation. The ask goes in the rate-lock message, and on this program it did not go into every rate-lock message.

Instagram was under-instrumented relative to its volume. Forty-two placements produced the largest reach contribution and the thinnest evidence. The comment-gated mechanic proved out on one reel; we should have briefed it into all forty-two.

We under-used repeat commissioning. The 1.41x uplift on second placements is one of the clearest findings in the dataset, and only four of twenty-three YouTube creators ran more than once. In the next cycle, the default is a two-placement commitment, with an option for a third.

Want the system, not just the case study?

Everything above, including the 100-point fit score, the counter-offer template, the rate-lock message with the two data requests, the channel-by-channel job list, the 30-day run sheet, and the attribution ladder, is written up in full in our playbook. It is free, 9 strategies long, and every number has a source.

Download the B2B Influencer Marketing Playbook

If you would rather we ran it, we source, vet, negotiate, brief, and report on creator campaigns across YouTube, X, LinkedIn, Instagram, and Discord. Every number in the playbook and this case study came from running them. Nothing publishes without your approval.

Get in touch | Explore the B2B influencer marketing agency service | Read more case studies