Demonstration
Supportedcontextdemonstrationresult
121 videos · 108 creators
Take ToucanGet the free skillFree AI script skill
For short-form creators and their AI
A hook is only a promise. Give your AI a structure that keeps it.
Attention Architecture studies 3,000 complete spoken TikToks—not just their first lines—to help your AI choose what a video should do next. Get the free skill and source research, or use the current system inside Take Toucan without copying files or maintaining prompts.
Attention atlas
Choose by communication goal
Story or reframe
Supportedpremisereversalpayoff
101 videos · 87 creators
Answer set
Supportedhookitemsaction
81 videos · 63 creators
The complete attention arc
0s → payoffAtlas updated 21 July 2026. The live repository is authoritative when the research changes.
The scale behind the skill
3,000 complete videos. Not 3,000 hooks clipped from context.
The atlas follows what happens after the opening: how a video establishes context, creates tension, demonstrates proof, reveals a result and earns its ending. That is the part generic hook lists leave your AI to guess.
- independent creators
- 1,321
- topic domains
- 12
- hours of spoken video
- 49.6
- captured public views
- 4.13bn
Captured public view totals describe the sample. They do not mean a structure caused reach or will guarantee it.
See the difference
Same idea. One follows generic advice. One has a job to do.
The idea is simple: show how a cheap desk light improved a video. The difference is not a louder hook. It is whether every line moves toward the promised result.
Without the skill
A loud opening with nowhere to go
“Stop scrolling if your videos look bad.
Here are three lighting tips: use natural light, buy a ring light and turn up your brightness.
Follow for more.”
Loud hook · disconnected middle · generic ending
With Attention Architecture
contextdemonstrationresult
“My videos still looked muddy even in daylight.
So I moved this cheap desk light from above my monitor to 45 degrees beside my face.
Here’s the same shot before and after. The light didn’t need to be brighter. It needed to create shape.”
Specific promise · visible proof · earned payoff
121
videos
108
creators
11
domains
87
exact variants
65s
median length
Evidence for the supported context → demonstration → result family. The scripts above are original illustrations, not source transcripts or promises of performance.
Purpose before template
The topic stays the same. The job changes the structure.
Demonstrating a result, telling a story and giving a set of answers ask different things of the viewer. The skill chooses for that job first, then adapts the pattern to your topic and duration.
Tell the story
The cheap desk light surprised me.
Without
I bought a cheap light, set it up and here’s what happened.
premise → reversal → payoff
Set up the expected compromise, overturn it with the placement discovery, then finish on what actually changed the image.
Supported evidence: 101 videos · 87 creators · 12 domains · 93 exact variants
Give an answer set
Three lighting fixes to try before filming.
Without
List three unrelated tips, then ask viewers to follow for more.
hook → parallel items → action
Promise a bounded answer, keep each fix comparable, then ask the viewer to change one variable and check the same frame.
Supported evidence: 81 videos · 63 creators · 10 domains · 67 exact variants
Give your AI the decision system
The skill, source research and creator email series are free.
Why the scale matters
One viral creator is not a system.
A big number is only useful when the sample makes imitation and easy bias harder. Each research choice protects a different part of the script your AI eventually writes.
- 01
Cap each creator
No creator contributes more than four selected videos.
One prolific account cannot quietly become the whole playbook.
- 02
Compare like with like
Relatively high- and ordinary-performing videos are compared within the same creators using age-normalised views.
Fame and time online have less opportunity to masquerade as structure.
- 03
Balance the sample
The collection targets variety across topic, duration, creator size and recency.
Advice is less likely to be a trend, niche or format presented as a universal rule.
- 04
Group by function
Videos match only when their complete functional progression appears in order.
The skill borrows a useful shape, not somebody else’s hook, wording or persona.
Research with restraint
The 1,773 videos it refused to turn into a formula matter too.
A video only enters a family when all three phases appear in the right order. Anything incomplete stays unassigned instead of being squeezed into a cleaner marketing claim.
32candidate families16routable families
- matched a complete canonical family
- 742
- left deliberately unassigned
- 1,773
- excluded from structural evidence
- 464
Structural coding was automated and videos were not individually human-audited. Findings are observational, not causal, and no pattern guarantees reach.
Packaging, measured properly
The caption should package the idea, not perform hashtag theatre.
Attention Architecture studies the description and hashtag layer separately, so generic discovery tags do not get mistaken for a substitute for a coherent spoken script.
Without
#viral #fyp #trending
Broad tags ask distribution to create relevance that the script and description have not earned.
With the skill
Name the subject and the person it is for.
Or leave tags out when a clear description already does the packaging work. Relevance is the goal, not decoration.
1,699
focused stacks
15
broad discovery stacks
269
no-hashtag examples
The AI script skill is free
Stop collecting hooks. Start building videos that know where they are going.
Confirm your email to get the skill, inspect the source research and receive seven practical creator emails. When you want the simplest route, Attention Architecture is ready inside Take Toucan from the first idea to the first take.