The Chart Is Not the Asset. The Data Licence Is.
Teams generating infographics with AI ask one question — can we own this. It is the less dangerous of the two questions. The visual is usually worth little and protected barely; the numbers inside it arrived under terms someone else wrote, and those terms travel with the file you publish.
A Chart Is Three Legal Objects Wearing One Coat
Most AI-copyright coverage treats a generated image as a single thing you either own or do not. A data visualization does not behave that way. Pull it apart and there are three distinct legal objects stacked in one PNG, each with a different owner and a different failure mode.
The ownership question people ask about applies to the middle layer only. The claims that actually get made against publishers come from the layers on either side of it.
Why the Ownership Answer Matters Less Than It Sounds
Copyright Office practice registers the human-authored contributions in a work that contains AI-generated material, with the generated portions disclaimed. For a chart, the human contribution is often real — you chose what to measure, how to bucket it, what to call out — but it is narrow, and it protects your specific presentation rather than the underlying finding.
That thin protection is usually survivable, because the business value of a chart is rarely exclusivity. It is attribution and reach: a competitor redrawing your data with different colours is generally free to do so, and the citation back to you is the point. Teams that plan to license visuals as a product, or to enforce against redraws, are the ones who need to know the claim is narrow before they build on it.
The asymmetry is worth stating plainly: weak protection in your own output costs you an enforcement option you probably were not going to use. A missed term in an upstream data licence costs you a takedown, a retraction, or a breach-of-contract claim from a vendor whose terms you agreed to at signup.
Where the Data Actually Came From
Ask an AI tool for a chart of a market and it will produce one. The provenance of the numbers is the question that has to be answered before the file leaves the building, and there are only a few possible answers.
- You supplied it. Cleanest case — but check whether your own data includes customer information subject to contractual confidentiality or privacy commitments before it becomes a public aggregate
- A licensed provider supplied it. Read the redistribution clause. Many analyst and market-data licences permit internal use and limited citation but prohibit publishing derived visuals, or require a specific attribution string
- A public or open dataset. Open does not mean unconditional — share-alike terms can attach to derivatives, and attribution requirements survive into every republication of the chart
- Scraped from the web at generation time. Terms of service, and in some cases anti-circumvention exposure, apply to the collection step regardless of what you do with the result
- The model produced it from training data. This is the dangerous case, because it looks identical to the others in the output and the figures may not correspond to any real source
The Fabricated-Number Problem Is Not a Copyright Problem
A model asked for "market share by vendor" will return plausible percentages that sum to one hundred. Rendered as a clean chart with a title and a source line, it reads as research. Published on a commercial site, it is a factual claim about a market — and if it favours your product, it is an advertising claim about your product.
Advertising-substantiation rules do not have a visualization exemption. The test is whether you had a reasonable basis for the claim at the time you made it, and a chart you cannot trace to a source fails that test in exactly the way a sentence would. This is the single most common way an AI-generated visual creates real liability, and it has nothing to do with who owns the picture.
Publishing Checklist for AI-Generated Visuals
- ☐Require a source citation for each series in the chart before it is approved for publication — not a source for the chart, a source per series
- ☐Treat any figure the model produced without a retrievable source as unpublishable, regardless of how plausible it looks
- ☐Keep the working file and the source links together, so the substantiation exists when someone asks a year later
- ☐Check whether your data licence permits publishing derived visuals externally, or only internal use and quotation
- ☐Capture the exact attribution string the provider requires and place it in the image itself, not only in surrounding page text that gets stripped on republication
- ☐Flag share-alike terms early — they can attach conditions to the derivative work, which for a marketing asset is usually a reason to pick a different source
- ☐Confirm the licence of icons, typefaces and map tiles composited into the output, especially for assets destined for a downloadable or resold template
- ☐Avoid using a generated visual as a logo or brand mark — the thin copyright and the unresolved asset chain are both worst-case there
- ☐Watch for style imitation of an identifiable illustrator or publication, which raises separate claims regardless of the data
- ☐Read the output-ownership clause of the specific plan tier used to generate, and prohibit consumer-tier generation for assets that go into products
- ☐Document the human contribution — the selection, arrangement and annotation — if you intend to register or enforce anything
- ☐Add an AI-use disclosure where your industry, platform or client contract requires one, and check client agreements for AI restrictions before delivering generated visuals
Frequently Asked Questions
We rebuilt the chart ourselves from the AI's layout. Does that give us copyright?
It gives you a claim in your own selection and arrangement, which is the protectable layer in any visualization. It does not resurrect protection in the generated elements you kept, and it does nothing about the data licence. Redrawing is a fix for the ownership question only.
Our source dataset is public government data. Are we clear?
On copyright, largely yes for US federal works. On accuracy, no — you still own the substantiation for what the chart asserts, including any aggregation or projection the model applied on top of the raw series. Public-domain input does not make a derived claim self-supporting.
The AI tool says we own the outputs. Is that enough?
It resolves the vendor-versus-customer question and nothing else. The vendor can only assign what it holds, and it holds nothing in third-party data or in material that copyright does not protect in the first place. Output-ownership language is a contract term between you and the vendor, not a clearance of the upstream chain.
A competitor redrew our chart with their own branding. Do we have a claim?
Probably a weak one on the data, since facts are not protectable, and a narrow one on the presentation depending on how much of your specific selection and arrangement they copied — and how much of that was human rather than generated. Trade-dress or attribution-based theories are sometimes stronger than the copyright claim here.
Do we need to disclose that a chart was AI-generated?
There is no universal requirement, but three places create one: platform policies for synthetic media, sector rules in regulated industries, and client contracts with AI-use clauses. The last is the most commonly missed — agencies deliver generated visuals into agreements that prohibit them.
Ask for the Source Line First
The cheapest control here is a rule that no visualization publishes without a source citation per series. It catches the fabricated-figure case, it forces the licence question to surface before publication rather than after a vendor notices, and it costs one line in a review checklist.
Run it against the charts already live on your site. The ones that cannot produce a source are the ones to fix, in that order.