---
title: "The IShowSpeed Method: Testing a Marketing Funnel With AI Personas"
date: "2026-11-03"
video_id: "7704006769579805966"
tiktok_url: "https://www.tiktok.com/@angusthenontechnical/video/7704006769579805966"
creator: "@angusthenontechnical"
category: "Marketing"
tags: ["ai personas", "marketing funnel", "sub-agents", "vibe coding", "ishowspeed"]
excerpt: "A TikTok uses IShowSpeed's stream prep to pitch AI persona agents that test your funnel. What his team really checks, how the loop works, and what is unproven."
---
The "IShowSpeed method" is a TikTok pitch from @angusthenontechnical: give AI sub-agents customer personas, have them walk your marketing funnel, let coding agents fix the friction they report, and repeat. The video says it "can be used to predict the success of any marketing." The IShowSpeed part is half right, since his cameraman does check the stream route beforehand, but that check is about internet bitrate. The video shows the persona loop running once and offers no evidence that it predicts real results.

## What does IShowSpeed's team test before a stream?

The video says "he tests everything first" and that "his team will walk the route beforehand to make sure the stream will hold." A [WIN.gg article on his IRL preparation](https://win.gg/ishowspeed-explains-irl-preparation/) reports that his cameraman sometimes covers the whole stream route on foot or by car to check the bitrate. If the connection holds, they go ahead. If not, they reroute.

So the route walk is real. The source does not say his team tests "everything", and the purpose is internet quality, not predicting how an audience will react. The jump from there to predicting marketing success is the creator's analogy.

## How does the AI persona funnel loop work?

The video describes this loop, and a four-step diagram on screen matches its order:

1. **Your customers.** A prompt "will scour all of your data. Like your past sales calls, CRM notes, signups" and produce customer personas.
2. **Walk the funnel.** Sub-agents role-play those personas, browse your ads "and the web", and report friction to a main agent.
3. **Fix it.** The main agent spawns coding agents that change the funnel.
4. **Walk again.** The loop returns to step 2 and repeats "until the persona sub-agents report that they had little to no friction."

## What do the personas look like?

Two persona cards appear on screen:

- "Agency / AI-services operator", with the goal "Land more clients; sell AI as retainers"
- "Content marketer", with a first-person quote beginning "I make all the content for my brand and I want a st[...] posting rou[tine]"

Both fit the creator's own audience of agencies selling AI services, so the example funnel appears to be his. The screen shows no source data and no count of personas produced.

## Can AI personas predict marketing success?

The video shows that a run happened. One task row reads "P1 Agency revised retest" with a duration of "1m 51s", and a code diff shows "5 files changed, 25 insertions, 5 deletions".

It shows nothing about outcomes. There is no before-and-after conversion data, no real customer response and no comparison with actual sign-ups. The stopping rule is a judgment by the same kind of model that plays the personas, so it measures whether simulated users agree the funnel is smooth. It does not measure purchases. I found no primary source showing that language-model persona walkthroughs predict real marketing results.

The narrower claim holds up: the creator has a working loop that reads personas, retests the funnel and edits files.

## Can you copy the prompt?

The video says "With this prompt right here," but no prompt text appears on screen, the description contains no link and the transcript asks viewers to do nothing to get it. The tools running the agents are not named either, and the screen does not show what the agents browsed.

## How to use this idea responsibly

If you try a persona loop on your own funnel, treat its reports as a usability pass that catches confusing copy and broken steps. Check the changes against real behavior, such as sign-up and conversion numbers from your analytics, before you trust that friction has fallen.

## Bottom line

IShowSpeed's cameraman does walk stream routes, but to check bitrate, not to predict audience reaction. The AI persona loop in the video is a plausible way to find friction in a funnel, and the screen shows it ran once. Nothing in the video shows that the loop predicts marketing success, so measure real users before acting on what simulated ones report.

## Sources

- [iShowSpeed explains IRL preparation, WIN.gg](https://win.gg/ishowspeed-explains-irl-preparation/): the cameraman's route check for bitrate and the practice of rerouting.
- [The IShowSpeed method, TikTok by @angusthenontechnical](https://www.tiktok.com/@angusthenontechnical/video/7704006769579805966): the video reviewed here.


## Video description and verification notes

**The file is a 52.22 s container (MP4, isom) of 1080x1920 portrait video, HEVC Main profile, 30 fps, about 521.9 kbps, with HE-AACv2 stereo audio at 44.1 kHz and about 48.2 kbps. The container bitrate is about 576.7 kbps and the size is 3,764,303 bytes (about 3.6 MiB). The metadata names format bytevc1_1080p_576672-1. The post timestamp is 1793728890, which is 2026-11-03 18:01:30 UTC, from the handle @angusthenontechnical (nickname "Angus the Nontechnical"). The audio credit is "original sound" by the same nickname, and the description is the 50-character title repeated ("The IShowSpeed method #ai #buildwithai #vibecoding"). Engagement at capture (2026-11-03 21:02:02 UTC, about three hours after posting) was 944 views, 33 likes, 0 comments, 5 reposts and 26 saves. I read four frames from the 2 s sampling set: frame_001, frame_008, frame_015 and frame_024. Assuming the first frame sits at 0 s, these sit near 0 s, 14 s, 28 s and 46 s, but I did not confirm frame times. The transcript has 18 lines and about 190 words, and I read it as plain text only (no segment-timestamp file was supplied, so I could not check segment times). Frame 1 shows a photo of IShowSpeed in a red jersey in a crowd, under a white caption box reading "IShowSpeed is a prediction engine", with the presenter cut out in the lower right and a burned-in subtitle reading "Vibe coders are now". Frame 8 shows a four-step diagram on a white background with an orange numbered circle for each step: "1 Your customers" with a dark box reading "Agency / AI-services operator" and "Land more clients; sell AI as retainers"; "2 Walk the funnel" with a dark task bar reading "P1 Agency revised retest", "1m 51s" and "2 tasks"; "3 Fix it" with "5 files chan[ged]", "25 insertions", "5 deletions"; and a step 4 starting "Walk ag[ain]" with a large "Ro..." box, plus an orange arrow looping back up to step 2. Frame 15 shows a dark text panel headed "Content marketer" with the quote "I make all the content for my brand and I want a st[...] posting rou[tine]..." circled in orange and the subtitle "Role playing as a". Frame 24 shows IShowSpeed holding a gold trophy in a selfie, a small inset photo of him holding a red and white flag, and the subtitle "dog, go make I show". No frame I read shows a full prompt, a terminal, a browser session, a funnel URL, a repo name or a measured conversion figure. The spoken argument is that IShowSpeed's team tests streams beforehand, so builders should give AI sub-agents customer personas, let them walk the marketing funnel and report friction, have coding agents fix the funnel, and repeat until friction is low. The transcript runs about 52 s of speech across 18 lines; I could not check the last segment's end time against the 52.22 s audio duration, and the final line ("follow me on my mission to Vibe Code everything") reads as a natural sign-off rather than invented text.**

## The IShowSpeed premise: partly supported, but the check is about bitrate

At roughly the 10 s mark the transcript says "his team will walk the route beforehand to make sure the stream will hold." Frame 1 sets this up with the caption "IShowSpeed is a prediction engine." A WIN.gg article on his IRL preparation reports that his cameraman sometimes covers the whole stream route on foot or by car to check the bitrate, and that if the connection is fine they proceed, while if it is not they reroute. The article's page date reads 02/06/2026, which is ambiguous between 2 June and 6 February, so I cannot give an exact date. The video's statement that the team walks the route is therefore consistent with the source. The source does not say "everything" is tested first (the video says "he tests everything first" at about 4 s), and the stated purpose is internet quality, not predicting audience reaction. The link from there to "predict the success of any marketing" is the creator's analogy and has no source behind it. WIN.gg describes a connectivity check, which is a pass/fail technical test on a physical route. A marketing funnel test with simulated personas measures something different, namely whether a language model's role-play finds confusing copy.

## The transcript's central claim: simulated personas "predict" marketing success

At about 20 s the transcript says the method "can be used to predict the success of any marketing." The video offers no result that supports this. Frame 8 shows a "P1 Agency revised retest" task of 1 m 51 s and a code diff of "5 files changed, 25 insertions, 5 deletions", which shows that a run happened, but I read no frame with before and after conversion data, no real customer response and no comparison against actual sign-ups. The stated end condition (spoken near the end, "until the persona sub-agents report that they had little to no friction") is a judgment by the same type of model that is simulating the personas. That measures agreement among simulated users and does not measure purchases. I did not locate a primary source showing that language-model persona walkthroughs predict real marketing outcomes, so the prediction claim is **Unverified**. The diagram does support a narrower statement: the creator has a working loop that reads personas, runs a retest and edits files.

## The data-ingestion step: described, not shown

Between roughly 15 s and 25 s the video says the prompt "will scour all of your data. Like your past sales calls, CRM notes, signups." Frame 8 labels step 1 "Your customers" and shows a single persona, "Agency / AI-services operator" with the goal "Land more clients; sell AI as retainers", and frame 15 shows a "Content marketer" persona with a quoted first-person goal. These are persona cards, but the frames I read do not show the source data, the prompt text or how many personas were produced. The cards read as having been written for the creator's own audience (agencies selling AI services), which means the example funnel is the creator's. The prompt itself was not on screen in any frame I read and the description contains no link, so what viewers would receive could not be checked. The video says "With this prompt right here" at about 17 s, but I saw no prompt text and no call to comment for it in the transcript.

## The agent loop: matches the diagram, with one unshown piece

The transcript describes sub-agents that role-play personas, browse ads "and the web", report to a main agent, which spawns coding agents, and the loop repeats. The diagram in frame 8 matches that order: customers, walk the funnel, fix it, walk again, with an arrow back to step 2. One detail does not show up: the video says personas browse "ads and the web," but the one task row in frame 8 shows only a title and a task count, so I cannot say what the agents actually browsed. The tool names are also absent. Nothing in the frames identifies which AI product ran the agents, so I make no claim about the tooling. The "1m 51s" duration for a retest is the only run-time figure and covers one task.

## Key Takeaways

- **Partial correction:** The video says at about 4 s that IShowSpeed "tests everything first." WIN.gg reports that his cameraman sometimes travels the route to check bitrate and reroutes if the connection is poor, so the verified test is about connectivity.
- **Verified:** A route walk before a stream is a documented practice by his cameraman, per WIN.gg, which quotes IShowSpeed on internet being essential.
- **Unverified:** The claim that persona sub-agents can "predict the success of any marketing" has no supporting data in the frames I read and I found no primary source for it.
- **Clarification:** The diagram in frame 8 shows one retest of 1 m 51 s and a diff of 5 files, 25 insertions and 5 deletions. It shows the loop ran once, not that friction fell.
- **Context:** The video is a 52 s clip with 944 views at capture. The prompt it refers to was not visible and no link was provided, so viewers could not reproduce the method from the video alone.

## Resources

- [iShowSpeed explains IRL preparation, WIN.gg, page dated 02/06/2026](https://win.gg/ishowspeed-explains-irl-preparation/) establishes the cameraman route check for bitrate and the reroute-if-poor practice I cite.
- [The IShowSpeed method, TikTok by @angusthenontechnical](https://www.tiktok.com/@angusthenontechnical/video/7704006769579805966) is the video itself, which I analyzed from the staged artifact files and did not fetch.
