How to make an editorial data explainer video with AI
One chart, one argument, one minute. Animated time series with the dataset cited, kinetic captions in time with the voiceover, signature Vox grey + paper-grain background. ~60s, 9:16 for TikTok and Reels.
why this format works
Data explainers work when one chart does the argument's load-bearing work. Housing vacancy rates over 70 years, inflation since 1990, electricity prices by source, college tuition vs wages — anything where the line on the chart is the story. The format earns trust by sourcing the dataset on screen (Federal Reserve, Census Bureau, BLS, Urban Institute), drawing the line as you watch instead of cutting to a static image, and letting kinetic captions advance the narration. Use this when your topic is quantitative and the chart is the proof. Use a different format if the argument is qualitative or doesn't have a single chart that does the work.
what ivy does
What's automated, scene by scene.
Reads the article or dataset
Paste an article URL with the chart embedded, or a dataset CSV, or both. Ivy pulls the dataset's source, time range, what's being measured, and the central trend. For named datasets (FRED series, Census tables, BLS series IDs), ivy fetches the canonical version directly so the chart matches the source exactly.
Draws the chart, doesn't paste it
The chart animates. Line draws left-to-right across the time range, an axis label fades in as the line crosses it, the current value appears at the line's tip with a small square marker. Sourced charts get an attribution line ("FEDERAL RESERVE ECONOMIC DATA, US CENSUS BUREAU") below the title in the muted Vox grey-green. No screenshots — every chart is rebuilt in SVG.
Layers a second chart for the second argument
Many data explainers have a setup chart ("here's the trend") and a payoff chart ("and here's why"). Ivy supports up to three charts in one video — the second-act chart is usually a stacked area, a histogram, or a small-multiples grid that breaks the headline number into its components.
Builds explanatory illustrations between the charts
Charts alone are dry. Between them, ivy builds simple SVG illustrations — a row of houses with one swapping to a multi-unit building (zoning), a yardstick with tick marks (magnitude), a US map with one state shaded (geographic specificity). Every illustration is in the video's palette, never stock-photo-style.
Writes editorial kinetic captions
Captions advance mid-scene with the voiceover. Single short factual phrases — chyron-length, not paragraph-length. The full transcript is generated from the voiceover and segmented into caption phrases that fit the screen.
Returns a brief you can edit before rendering
Re-time a beat, swap the dataset's framing, change the highlighted year, soften a caption. After the brief, render is ~5 minutes for a 60s explainer.
make one in five minutes
Six steps. Five minutes.
Pick one chart-shaped argument
"Why are starter homes disappearing?" "What's actually happening with rent vs wages?" "Where does US electricity come from now vs 2000?" One question, one chart that answers it. Vague topics produce vague explainers.
Paste the article or dataset source
An article URL with a chart embedded works. A FRED series ID, a Census table, or a CSV upload also works. Ivy needs to know what dataset to draw and what to cite as the source.
Confirm the chart type and time range
Brief shows: "Will draw a line chart of US homeowner vacancy rate, 1956-2023, sourced from FRED (HOWN_RATE), with the current low highlighted." Override if the framing or time range is wrong.
Press Render
Vertical 9:16 by default. Ivy returns an MP4 in ~5 minutes. About 55-65 seconds.
Refine via chat
"Highlight 2008 instead of 2023." "Add a second chart of price tier breakdown." "The third caption is wrong — change it to: ..."
Post to TikTok / Reels / Shorts
Download the MP4 or post directly to X from inside ivy. For TikTok and Instagram, download and upload there.
what to watch out for
Hard-won lessons from shipping these.
Cite the dataset on screen
Every chart needs a source line in small caps under the title ("FEDERAL RESERVE ECONOMIC DATA, US CENSUS BUREAU"). This is what separates an editorial data explainer from a TikTok hot-take. Without the citation, the chart reads as fabricated and the video stops being a data explainer — it becomes opinion. Don't disable the citation.
One chart, one argument
The strongest data explainers are about a single trend. If your topic needs three charts to hold up, it probably needs to be three videos. The most-viewed chart-driven shorts land at one headline chart + one breakdown chart, not five.
Time range matters
Cherry-picked time windows are the fastest way to lose viewer trust on a data explainer. If the dataset goes back 70 years, show 70 years. If you're zooming in on the last decade, say so on screen and explain why. Ivy will flag time-range choices that look cherry-picked.
Caption pacing
Captions advance mid-scene with the voiceover. Each caption is one short factual phrase that fits in roughly two lines. Long captions read as bad subtitles and tank retention. Default ivy caption length is good — only override if you have a specific reason.
Illustrations between charts, not instead of them
House icons, ruler bars, map overlays bridge between charts and keep the video visually moving — but they are not the argument. Don't replace charts with illustrations because the data is awkward. If the chart is awkward, the argument needs work.
What ivy won't do
Won't fabricate data points. Won't extrapolate beyond the dataset's time range. Won't show a chart with a fake source attribution. If your prompt asks for a chart with no underlying data, ivy will push back and ask for the source.
try one of these prompts
Steal these.
faq
Common questions.
Federal Reserve Economic Data (FRED), US Census, BLS, BEA, Urban Institute, OECD, World Bank, and EIA — these are first-class sources. For other datasets, paste a CSV or link to the source and ivy will use that. Every chart cites its source on screen.
Your first one's on us.
Paste a URL, get a video back in three minutes. Two free credits to start.