Viral video analysis with Jev, Virlo and Claude: a content engine that judges every video

Updated

Automated viral video analysis splits the job in two: one model judges, another writes. In this content engine, a Virlo marketing agent collects the videos outperforming in your niche, TypeSafe's Jev judges each one (worth scripting or not, hook type, format, score against the creator's own baseline), and Claude turns only the videos that pass into ready-to-film scripts. Five prompts set it up, and a sixth runs it on a schedule.

  • Why a judging model and a writing model beat one chatbot for video analysis
  • The 5 prompts that wire Jev, Virlo and Claude together
  • How to run the analysis every morning on its own
TikTokInstagram

How do you analyze viral videos at scale?

Analyze viral videos at scale by giving each step to the tool built for it: a data source that collects the videos, a reasoning model that judges them, and a writer that scripts the ones worth copying. A chatbot asked to "analyze this video" describes it; this engine makes a call on it.

The content engine, stage by stage
StageWho does itOutput
1. Watch the nicheVirlo marketing agentThe videos outperforming in your niche
2. Judge every videoJev (TypeSafe), across 80 signalsWorth scripting or not, hook type, format, angle and score
3. Write the scriptsClaudeData-backed scripts for the videos that passed
4. Film, edit, postYouAn iPhone, CapCut and a short keyword-relevant caption

The videos come from Virlo's database of 23.3M viral TikTok and Instagram videos, the same pool behind Outliers. For the manual version of this analysis, see outlier videos.

What is Jev and how does it judge a video?

Jev is a reasoning model from TypeSafe that judges instead of chatting or writing. It answers in only three ways, which makes its output easy to filter in a pipeline.

Jev's three answer types
Answer typeWhat it returnsUsed here for
NoulA 0-1 callIs this video worth scripting?
ChoicePick one optionThe hook type, the format and the angle
ScoreA numberHow the video performed against that creator's own baseline

Get a Jev API key by joining the waitlist at typesafe.ai. Some early users report same-day approval; results vary. You also need a Virlo account and API key, which feeds Jev the videos: create your Virlo account, then copy a virlo_tkn_ key at dev.virlo.ai.

How do you set up the viral video analysis engine?

Paste these prompts into Claude in order. Claude does the connecting, scripting and running; Jev does the judging. The Virlo for Claude page covers the connector if you want to add it by hand.

Prompt 1: wire Jev into your pipeline (one-time setup)
Goal:
- connect my TypeSafe Jev API key so every video from my Virlo marketing agent
  gets judged before it reaches me

Keys:
- Jev (TypeSafe) API key: [PASTE YOUR TYPESAFE KEY]
- Virlo MCP server: https://dev.virlo.ai/api/mcp/mcp
- Virlo API key starts with virlo_tkn_

Process:
- walk me through connecting both
- once connected, confirm by running one real video through Jev and showing me the judgment

Prompt 2 sets up what watches your niche. Skip it and Jev has nothing to judge.

Prompt 2: set up your marketing agent
Goal:
- set up a Virlo marketing agent that tracks what's outperforming in my niche

Process:
- ask me for my niche, my brand name, my competitors' names, and which platforms
  (TikTok, Instagram)
- suggest the keyword mix, then confirm it with me before saving

Save the agent ID when it's done.

Prompt 3 is the step most people skip: judging every video instead of only collecting them.

Prompt 3: get Jev's judgment on every video
Goal:
- run every video my marketing agent [AGENT ID] surfaces through Jev

Process:
- for each video, get Jev's judgment directly, not a description:
  1. Noul: worth scripting, 0 to 1
  2. Choice: the hook type, the format, and the angle
  3. Score: how it performed against that creator's own baseline
  4. Confidence attached to each call
- only hand me the ones that pass the Noul threshold

Jev hands back a judgment, not a script, so Claude takes over for prompt 4.

Prompt 4: turn it into a script
Goal:
- turn the videos that passed Jev's judgment into ready-to-film scripts

Process:
- pick the hook that fits my brand best, adapted under 12 words
- name the format to shoot it in and why
- build out the rest of the script (body + CTA) around it
- keep the original video linked so whoever's filming can watch the reference

Can the viral video analysis run on its own?

Yes. Claude can run the whole chain on a schedule: pull the newest videos from your marketing agent, judge each with Jev and script only what passed, skipping the day when nothing does.

Prompt 5: make it run on its own (optional)
Goal:
- run this automatically every day or week, at 7am [MY TIMEZONE]

Process:
- pull the newest videos from my marketing agent [AGENT ID]
- run every one through Jev
- script only what passed
- skip the run entirely if nothing passes that day

The marketing agent itself already runs daily inside Virlo; see marketing agents.

Why do videos go viral, and what should the analysis look for?

Judge a video against its own creator's baseline, not its raw views: a video that beats what that creator usually gets is the signal, whatever the follower count. That is why prompt 3 asks Jev for a score against the creator's baseline, plus the hook type, format and angle that explain the result.

Hook type is usually the first thing to check. See which hook types win most often on types of hooks, and borrow proven lines from viral hooks. For a no-code way to break down a rival's videos, use competitor content analysis.

What happens after the scripts are written?

  • Filming: an iPhone is enough.
  • Editing: CapCut yourself, or a low-cost freelance editor.
  • Posting: TikTok and Instagram Reels, with short, keyword-relevant descriptions.

Each script keeps the reference video linked, so whoever films can watch what they are copying.

Start with the data: point a marketing agent at your niche and get the outperforming videos every day.

How this resource was made

Source
The original "content engine that finds, watches and analyzes viral videos (with Jev)" resource, September 2026: its prompts and Jev's three answer types as described there. The 80-signal figure is TypeSafe's, as stated in that resource.
How this page is made
Written by the Virlo team from the original resource; prompts are word for word except the platform list and one word in prompt 2.
Limitations
Jev access is waitlist-based and its details belong to TypeSafe; check typesafe.ai for current terms.
As of
(first published September 18, 2026)

Source: Virlo, Viral video analysis with Jev, Virlo and Claude: a content engine that judges every video, November 2026, virlo.ai/resources/viral-video-analysis

Want the full list?

A marketing agent collects the TikTok and Instagram videos outperforming in your niche every day. Virlo is social listening for TikTok and Instagram Reels: start a 7-day free trial at app.virlo.ai and see real videos from your niche before you pay.

Frequently asked questions

What is viral video analysis?

Working out why a video beat its creator's usual performance: its hook type, format, angle and how far it outperformed. This engine automates that judgment with Jev.

How do you analyze why a video went viral?

Compare it with the creator's own baseline, then name its hook, format and angle. Prompt 3 asks Jev for exactly those calls, with a confidence on each.

What is Jev?

A reasoning model from TypeSafe that only judges. It answers with a Noul (a 0-1 call), a Choice (pick one) or a Score.

How do I get a Jev API key?

Join the waitlist at typesafe.ai. Once approved, TypeSafe issues your API key; some early users report same-day approval.

Why not let Claude judge the videos too?

A chat model tends to describe a video rather than make a call on it. Jev returns a decision you can filter on, and Claude writes from the ones that pass.

Where do the videos come from?

A Virlo marketing agent that tracks what is outperforming in your niche, drawing on Virlo's database of 23.3M viral TikTok and Instagram videos.

What is a content engine?

A pipeline that finds, judges and scripts videos on repeat. Here: a marketing agent finds them, Jev judges them and Claude writes the scripts.

Can the analysis run automatically?

Yes. Prompt 5 runs it daily or weekly at 7am in your timezone and skips the day if nothing passes Jev's threshold.

What do I need to film the scripts?

An iPhone, CapCut or a low-cost editor, and a short keyword-relevant caption for TikTok and Instagram Reels.

Run this on your own niche

A marketing agent collects the TikTok and Instagram videos outperforming in your niche every day.

7-day free trial, easy to cancel.