attention/dna Request access

Videos · Image ads · Copy · Landing pages

Your content
has
a DNA.

Attention DNA reads every video, image ad, line of copy and landing page you make, paid or organic, and breaks it into what holds attention and what loses it: the hook, the words, the delivery, the offer. Then it shows what your strongest work has in common.

Request access See how it works

Enterprise workspaces · pilot on content you can already score

Illustration of a strand: three rows of per-second cells for visual, language and auditory response.

The gap

Views, likes and run time tell you what happened. They never tell you why. Attention DNA reads the video, the image, the copy and the page itself, so you can see what your best work shares, and what the rest is missing.

  • 4Formats: video, image, copy, landing page
  • 40+Signals read per video
  • 5Language DNA styles, in every format

Two cuts. One holds.

Put two videos side by side. For videos run through the brain-encoding model, the strand shows where each one holds attention and where it lets go, second by second.

AFounder to camera

Hook /10
8
Held
74%

BProduct montage

Hook /10
6
Held
48%
Example readout — illustrative values.

How a readout gets made.

  1. Import01

    Example

    Bring it in by link.

    A creator's Instagram, TikTok or YouTube profile, a brand's ads from the Meta Ad Library, any landing page, or copy you paste. No downloads.

  2. Read02

    Example

    Read the way an audience meets it.

    An AI model watches, listens to and reads each piece: the hook, the pacing, the faces and sound, the words, the offer. For video, a brain-encoding model can add a second-by-second strand.

  3. Label03

    Example

    Name what is there, and when.

    Hook, story, proof and call to action, the Language DNA of the words, and for video the delivery, the emotion shown and the key moments.

  4. Compare04

    Example

    Learn from your best work.

    Split a folder of videos or ads into high and low performers and see which differences hold up under a strict test, with the pieces that show them.

The readout

One readout for every format.

Bring in a video, an image ad, a block of copy or a landing page. Each is read for what it is, scored on its own terms, and given the same Language DNA, so you can line them up.

Video

Comes in fromInstagram, TikTok and YouTube profiles · Meta Ad Library · direct video links

Hook and opening line
1–10
Pacing and cuts
counted
Words on screen and transcript
read
Delivery and emotion shown
labelled
Key moments
timed
Language DNA
five styles
Brain-encoding strand
when encoded

Scores

  • Hook
  • Scroll-stop
  • Share-worthiness
  • Rewatch

RUN · founder-to-camera.mp4complete

Brain-encoding strandwhen encoded

  1. 0:03Hook pays offHOOK
  2. 0:08First proof on screenPROOF
  3. 0:13Product on screenPRODUCT
  4. 0:17Call to action startsCTA
  • Hook8/10
  • Scroll-stop7/10
  • Share-worthiness6/10
  • Rewatch7/10
Example readout — illustrative values.

Image ad

Comes in fromStatic image ads from the public Meta Ad Library

Words on the image
read
People, product and CTA
read
Visual design
described
Language DNA
five styles

Scores

  • Scroll-stop
  • Hook strength
  • Clarity
  • Share-worthiness

IMAGE AD · Meta Ad Librarycomplete

  • Words on the image: ‘Built for long days’
  • CTA: Shop now
  • Scroll-stop8/10
  • Hook strength7/10
  • Clarity9/10
  • Share-worthiness6/10

Language DNA · words on the imagefeeling-led

Facts 15%Fun 5%Feeling 45%Opinionated 5%Action 30%

Example readout — illustrative values.

Copy

Comes in fromPaste an ad, caption, post, email or script, or a page's URL

Headline and opening line
read
Claims and proof
listed
Reading level and structure
labelled
Call to action and urgency
read
Language DNA
five styles
Biggest weakness
one line

Scores

  • Hook
  • Read-on
  • Specificity
  • CTA strength

COPY · launch caption, pastedcomplete

Twelve-hour shifts end at your feet.Feeling Three layers of foam, tested by nurses for six weeks.Facts Honestly, most insoles are an afterthought.Opinionated Your feet may send a thank-you note.Fun Try a pair for 30 nights.Action

Facts 26%Fun 14%Feeling 24%Opinionated 16%Action 20%

Biggest weaknessThe offer waits for the last line.

  • Hook7/10
  • Read-on8/10
  • Specificity9/10
  • CTA strength5/10
Example readout — illustrative values.

Landing page

Comes in fromAny public https page · one desktop capture

Headline and offer
read
Above the fold
checked
Form friction and social proof
read
Objections
listed
Language DNA
five styles
Biggest leak
one fix

Scores

  • Conversion likelihood
  • Above the fold
  • Offer clarity
  • Trust
  • Message clarity

PAGE · brand.com/offercomplete

  • Conversion likelihood6/10
  • Above the fold7/10
  • Offer clarity5/10
  • Trust6/10
  • Message clarity8/10

Language DNAfacts-led

Facts 38%Fun 6%Feeling 16%Opinionated 12%Action 28%

Biggest leakThe form asks for a phone number before the offer is clear.

Example readout — illustrative values.

Then, as a team

  1. 01

    Organise

    Folders, Spaces and shared Boards for brands, campaigns and clients, with a search across everything.

  2. 02

    Analyse

    Split a folder of videos or ads into high and low performers and see which differences hold up, with the pieces that show them.

  3. 03

    Check

    See whether the scores follow real results where you have them (views, how long an ad ran) and, for video, how often a person agrees with the AI.

  4. 04

    Brief

    Export a script brief, or print a readout for your client.

Method

Numbers you can defend.

Your workspace checks its own read. Nothing is called a difference until it holds up, and nothing untested is dressed up as one.

  • rho 0.41 · 95% CI −0.17 to 0.78 · n 14

    Does the score follow real results?

    Each workspace ranks its own content by the AI's scores, checks the order against what really happened (views, or how long an ad ran), and says so when it doesn't.

  • h 1.25 · 95% CI 0.45 to 2.05 · n 12 vs 12

    An interval on every difference

    Effect size, interval and sample size beside every finding. What only leans, or has too few pieces to say, is labelled as such.

  • 14 of 20 agreed · 70%, 95% CI 48–85%

    Checked against a person

    A reviewer labels tone, delivery and style blind, and the Analysis tab prints how often they agree with the AI once ten videos are labelled.

Example values.

Made for teams that publish.

  • 01

    Brand and content teams

    Put your videos, ads, copy and pages in one library, paid and organic, and learn what held attention for your brand.

    • Paid and organic in one library
    • High-versus-low analysis
    • Briefs from what held up
  • 02

    Agencies

    Settle the creative debate with evidence your client can see.

    • Competitor ads from the Meta Ad Library
    • Landing page readouts
    • Client-ready printouts
  • 03

    Creators and social teams

    See what your strongest videos share, and what the rest are missing.

    • Profile imports
    • Views beside the read
    • Video leaderboard
  • 04

    Researchers

    Per-second predicted activation for encoded videos, in the same readout as the transcript and the timed on-screen words.

    • Effect sizes with intervals
    • Blind agreement check
    • Leaderboard CSV

Imports

Straight from the feed.

Pull a creator's latest videos, a brand's video and image ads from the public Meta Ad Library, any landing page, or copy you paste. Your library stays put, ready to compare.

  • Instagramvideos
  • TikTokvideos
  • YouTubevideos
  • Meta Ad Libraryvideo and image ads
  • Landing pagesany https page
  • Copypaste or URL
  • CSVbulk creator import

IMPORTtiktok.com/@creator10 videos

IMPORTMeta Ad Library · brand9 video ads · 3 image ads

CAPTUREbrand.com/offerlanding page

PASTElaunch caption, draftcopy

✓ readhook · words · delivery · offer

✓ dnafacts · fun · feeling · opinionated · action

READOUTready

Example.

Good questions.

Is this real fMRI?

No scanner is involved. Videos can go through a brain-encoding model trained on fMRI recordings, which predicts how a brain would respond. Everything else is read by an AI model. Readouts are directional, not clinical.

What is a strand?

For a video run through the brain-encoding model, a per-second map of predicted activation across the visual, auditory and language channels. Brighter cells mean a stronger predicted response.

What can I bring in?

Videos from Instagram, TikTok and YouTube profiles or a direct file link; video and image ads from the public Meta Ad Library; any public landing page; and copy you paste or link. Everything comes in by link or paste; you never send a video or image file.

Paid and organic in one place?

Yes. Organic videos come with their views, and likes where the platform reports them; ads from the Meta Ad Library with how long they ran. Copy and landing pages are read on their own merits.

Does a strong readout mean it will perform?

Not on its own. The scores are the AI's ratings, not your results. Where real results exist, the Analysis view checks whether the scores follow them, and says so when they don't.

Can my whole team use it?

Yes. A shared workspace with admin, member and viewer roles, plus spaces, folders and boards for campaigns and clients.

Is my content private?

Your workspace's records are readable only by its signed-in members; the database enforces it with row-level security. Stored files (video copies, thumbnails, page captures and brain-encoding output) sit behind unlisted links.

Can I get the raw numbers?

Yes. Export the video leaderboard to CSV, and every finding shows its effect size, interval and sample size.

Decode your
next launch.

Attention DNA is an enterprise engagement, not a self-serve tool. We open a workspace for your team, load content whose results you already know, and show you what the read says about it.

Request access
  1. 01

    Validation pilot

    We analyse a slice of your past content, work you can already score, so you can judge the read against outcomes you know.

  2. 02

    Your workspace

    One shared library for your team, with roles. We import from social profiles and the Meta Ad Library, and mirror the media so a past study still opens.

  3. 03

    Review together

    We walk the readout with your team and agree what the analysis is allowed to claim, and where the evidence is too thin to claim anything.