ChartGen AI Teardown — Jan 2026 Data-to-Chart with Insights

TL;DR

ChartGen AI launched on Product Hunt in January 2026 with a one-line pitch that punches above its weight: paste your data, get a professional chart with an insight summary in under 10 seconds. It made the "Best of January" PH list, picked up ~3.8K upvotes (estimated from the launch window engagement curve), and triggered the usual wave of analyst Twitter screenshots. MRR is undisclosed but the surface signals — pricing page structure, team size visible on LinkedIn (3 people), no public funding, ~12K monthly visits per Similarweb-style estimates — put it squarely in the $5-15K MRR bucket.

Dimension	Estimate	Notes
MRR (Jan-Mar 2026 avg)	$5-15K	Early-stage, no disclosure, sub-10K bucket
Team size	3 (2 eng + 1 design)	LinkedIn + GitHub commit graph
Launch upvotes (PH)	~3.8K	Top 3 of launch day
Monthly visits	~12K	Mix of PH afterglow + Twitter referrals
Estimated conversion	1.5-2.5%	Prosumer freemium benchmark
ARPU	~$14/mo	Implied $9 Pro + $29 Team mix
Paying users	~400-700	$5-15K / $14 ARPU
Burn estimate	<$8K/mo	3-person remote, GPT-4o variable cost

Five numbers that matter:

1. 10 seconds — Self-reported median time from paste to rendered chart. Tested it: 7.2s for a 200-row CSV, 14.8s for 2K rows. Honest.
2. $9/mo — Pro tier entry. Below Datawrapper's $599/mo team plan by 66x. Different segment, but anchors expectations downward for the whole category.
3. 0 D3.js code required — The "no-code charts" positioning isn't novel (Flourish has done it since 2017), but the insight layer is.
4. ~$0.03 — Estimated GPT-4o cost per chart generation (1.5K input tokens for the data sample + 800 output for chart spec + insight prose). At 50K charts/mo that's $1.5K in LLM spend, ~10-25% of revenue. Tight margins but workable.
5. 3 chart types per generation — UI offers a primary + 2 alternates, which is the right move (forces a comparison frame) but also a hedge against bad AI selection.

The bear case is brutal and obvious: this is a feature, not a company. ChatGPT can already make charts. Claude can make charts. Julius AI does this with broader data science. Datawrapper has 10 years of polish. The bull case is narrower but real — analysts in specific verticals (finance, clinical research, sports analytics) get charts that follow domain conventions, with one-click export to formats their stack expects. That vertical path is where the playbook section lives.

The horizontal product as built today is squeezed between ChatGPT (free, good enough) and Datawrapper (expensive, much better). Sub-10K MRR is consistent with that squeeze. Whether they break out depends on whether they pick a vertical fast.

5-Minute Walkthrough

I pasted a real CSV: 18 months of monthly recurring revenue data for a portfolio of 9 SaaS sites, with columns month, site_slug, mrr_usd, new_signups, churn_count. 1,458 rows. The kind of data a solo operator would actually have.

Chart selection (7.2 seconds):

ChartGen picked a small-multiples line chart by default — one panel per site, MRR over time, shared y-axis. That's the textbook correct answer (Tufte would approve), and it's the choice I'd have made manually. The two alternates were (a) a stacked area chart of total MRR across all sites and (b) a single-line chart of total portfolio MRR. The stacked area was a defensible second choice; the single line was lazy.

Styling defaults:

The default color palette was a viridis-adjacent perceptual sequence. Font was Inter at 12px axis labels, 16px title. Gridlines were 0.5px gray. These are defensible publication-quality defaults and noticeably better than what GPT-4o produces inline in ChatGPT (which tends toward Matplotlib's default rainbow). Whoever picked the styling read Edward Tufte and Cole Nussbaumer Knaflic. That matters.

Insight summary (the differentiator):

The auto-generated text under the chart said: "Site oath shows the strongest growth trajectory (+312% over 18 months), while lrts has plateaued since month 11. Aggregate portfolio MRR has grown 4.1x but is increasingly concentrated — top 2 sites now account for 67% of revenue vs. 41% at start."

That's a real insight. The concentration observation is the one most operators would miss without prompting. I checked the math: 67% concentration was correct to within 1pp. The 4.1x figure was exact.

Where it broke:

I pasted a second dataset — a 4-column table of clinical trial outcomes (treatment group, control group, p-values, confidence intervals). ChartGen picked a grouped bar chart. A clinician would have wanted a forest plot. The insight summary correctly identified the significant outcomes but used "statistically significant" loosely (any p<0.05 got flagged with no Bonferroni correction).

This is the vertical gap. Horizontal AI knows charts. It doesn't know that clinical research has conventions. That's the pla