
Results
On the ElevenLabs marketing site I designed heroes that let visitors start the actual job: drop in a file to transcribe, or type a prompt and pick a video model. The pattern lifted sign-ups, sent people into the product already working, and became infrastructure for the creative funnel.
+48% sign-ups on Speech to Text
Sign-up rate rose from 12.0% to 17.8% on English /speech-to-text, and the hero rolled out to 100%
Sign-ups from the page itself rose from 9.6% to 14.0%
+70% transcriptions and +48% activation
More visitors arrived in the app already doing the job: users creating a transcription went from 9.9% to 16.8%
Credits used per user rose 68%
+28.5% subscription rate on /video
The prompt-and-model hero lifted sign-ups 10.3% and subscriptions 28.5% across 61.5K exposures on the highest-traffic creative page
One hero, reused across the funnel
The video hero became the template for external model pages like Seedance 2.5 and Gemini Omni Flash, which now publish without engineering
It anchors the paid AI Video Generator landing page, where the body below it added 12% more new-subscription MRR per user
The Problem
Our highest-intent pages explained the product instead of letting people use it.
A headline, then a generic input
/speech-to-text led with a title and the shared Omnibox, the same prompt box used across the site
/video led with a static headline, a Create video button, and a model grid below the fold
Sign-ups that didn’t activate
Visitors signed up without a task in hand, so many reached the app and stalled before their first transcription or video

The Idea
Start the job in the hero. When the primary action on the page is the product action, sign-ups and activation should move together.
Input first
Put the upload, the link field, and the prompt box in the first screen
Add playable samples for people who don’t have a file ready
Sign up to finish, not to start
Let visitors begin the transcription or choose a model, then ask them to sign up to download or generate
Build it once, reuse it everywhere
Design the hero as a component the web team can drop onto model pages and paid landing pages

The Action
Prove the pattern on long-tail pages
Tested compact, input-first, and all-in-one heroes on the audio-to-text, video-to-text, and TikTok transcript pages in August and September
Ship it on the main pages
Video: designed the prompt-and-model hero with the Marketing Web team and tested it in August
Speech to Text: designed the compact audio hero with upload, link, and samples; tested October 16 to 26 with about 12,500 visitors per arm

Carry it into the product
Designed the in-app companion: on the Speech to Text page, upload, link, and record now sit inline instead of behind a Transcribe files button
Added a recording option to the marketing hero

Turn it into infrastructure
The video hero became the model-page template, and stayed constant in the paid video tests so we could isolate the page body
A filmstrip hero variant tested against it in late October lost on sign-ups (−11%, −32% on mobile) and was stopped after a day
What I Brought
Design
Designed the hero components, the sample players, and the in-app inline input
Explored compact, input-first, and all-in-one variants before committing
Strategy
Framed “start the job in the hero” as a pattern to roll out, not a one-off test
Measured activation alongside sign-ups so a win meant people actually used the product
Collaboration
Worked with Marketing Web engineers on the site, the Growth Product team on the in-app version, and Growth Data on the readouts
Reflection
Interactive heroes beat descriptive ones on paid conversion, not just sign-ups. The video hero moved subscriptions almost three times as much as sign-ups.
Words alone weren’t enough. A version of the audio-to-text page that only changed the vocabulary was a draw; changing the structure so people could start the job is what won.
The best outcome was the reuse. A winning hero became a template the team publishes without engineering, and the starting point for our paid landing pages.
More screens

In the app before: transcription behind a button

The same hero reused on the Seedance 2.5 model page

Live /video with the prompt hero, which beat a filmstrip hero in late October


