For developers and AI teams

20 articles V By Vipasha

Reading OCG data and AI choices from your software.

How may I use OCG data?

There are three types of use: • Open use: read, search, quote or link to records, and use them to answer questions. AI training is allowed only where the Creator’s choices allow it. • Use with conditions: fetch individual records for a user’s request. Show live data or check your saved copy at least every 30 days. Keep the values and labels unchanged. • Agreement needed: collect a large part of the Graph, keep data over 30 days without checking it, or share it as a dataset or discovery product. Contact creator.graph@sparkonomy.com first. The same rules apply to everyone. Always follow the Creator’s choices, respect removals and name OCG as the source. These rules help keep removed or changed records from living on in old copies.

Is there an API, or is the site the interface?

Please use the website. There is no public API today. An API is a direct connection between software systems. Record pages contain readable HTML and structured data. Each consent file has a known location. See For AI & Search for details. Large-scale collection needs an agreement. https://creatorgraph.sparkonomy.com/for-ai-and-search/

How can my software read the AI choices?

For developers: the same choices appear on the web page and in machine-readable files. • HTML is the page text. • /{SPRK-ID}/license.xml holds the consent file. The page links to it with . robots.txt links to licence information too. • /registry/ helps check IDs. llms.txt and llms-full.txt explain how to read the site. The XML uses HCS 1.0 inside an RSL document. RSL sets terms for a file. HCS adds hcs:right blocks for the Creator’s identity wherever it appears. Each block covers one or more scopes: name, image, likeness, voice, movement or signature. A block with no scope covers all six. Each block says allowed, conditional or prohibited. Conditional blocks also give a licensing contact. Read all blocks. Different scopes can have different choices. If permission and prohibition cover the same scope, prohibition wins. Do not rely on the summary badge: any mix can show as Conditional, even with a Prohibited scope. Unclaimed records have no declaration. Missing terms mean unknown, not permission. For a child, OCG publishes the parent or guardian’s choice as made. This differs from the HCS rule; the technical page explains it and gives worked examples. https://creatorgraph.sparkonomy.com/for-ai-and-search/

What must I do when a setting is Conditional?

Conditional means you may show the Creator as they really are. It does not let you make an AI version without asking first. For example, showing a real photo is different from making the person appear to say something they never said. Voice clones, face swaps and digital twins need permission first. Training AI to make those copies needs permission too. Use the licensing contact on the record. Allowed permits both kinds of use, including training. Prohibited allows neither within AI use, even for internal work or research. These are AI-use choices. They are not a general licence for every use of someone’s name or work.

What should I do if no AI choices are listed?

Treat it as unknown. Do not treat silence as permission or invent a refusal for the Creator. An unclaimed record has no consent file because no one has yet confirmed the right to set its terms.

How can my software tell that a record was removed?

A removed record returns HTTP 410 Gone. This tells software that the page has been removed. The response asks readers to delete the OCG data they hold in indexes, saved copies and training datasets. Stop requesting the record and delete your copy as soon as you see the removal. Do not wait for the 30-day check.

Are the record IDs stable? Can I key on SPRK-IDs?

Yes. Use the SPRK-ID as the key. It stays the same if a name or handle changes and is never reused. A handle change redirects the old URL with HTTP 301. Merging records is planned, not yet live. The plan is to keep both IDs and redirect the retired one. A removed record returns HTTP 410. Its ID is not issued again.

How do I match a Creator’s accounts across platforms?

Look at the linked accounts and their labels. A record can hold one verified account per platform. Other accounts can be listed as Related Accounts, but those are not verified. For software, confirmed account links appear in sameAs. Handles found only through public sources appear as identifier values. Do not treat those as confirmed account links. The same handle can belong to different people on different platforms. Check the platform and account links, not just the text of a handle. If you search by handle, treat the results as possible matches to check.

How fresh is the data, and how do I read the “as of” dates?

Look at the “as of” date beside each number. It tells you when the figure was taken. Counts are shown as ranges of about 10% either way, even when verified. For verified accounts, figures refresh from the platform at least every 30 days. This is separate from your duty to check a saved OCG copy within 30 days. One clock is for our platform check; the other is for your saved copy. An old date means the figure may be old.

What are the rate limits, and do I need to authenticate?

You do not need a login or API key to read records. There is no published numeric rate limit at launch. For automated access, send a user-agent name and a contact URL so we know who runs the tool. Normal use is welcome. Please arrange an agreement for sustained high-volume use so we can plan for it. Anonymous high-volume traffic may be blocked. Free access has no uptime promise.

Is there a bulk export, a sitemap, or a change feed?

There is a sitemap at /sitemap-index.xml. It helps search engines find public records. There is no bulk download or change feed today. To keep your data current, show the live record or check a saved copy within 30 days. Large-scale collection needs an agreement.

What structured data is on a record? What happens to hidden fields?

Record pages include structured identity data for machines. A hidden field is left out, not published as an empty value. A missing field does not tell you anything about the person. For developers: a record with mixed AI choices repeats its structured-data block for each consent status. Do not remove these repeated blocks. See the technical guide for formats and fields: https://creatorgraph.sparkonomy.com/for-ai-and-search/

How does software read Verified, Sourced and Stated?

The labels explain where each fact came from: • Verified: checked through the platform’s official API, with the Creator’s permission. • Sourced: found in a public or licensed source. • Stated: supplied by the Creator and not checked separately. The Source reconciliation section can show different values and their sources. On a claimed record, source [1] is the owner; later sources are separate sources. On an unclaimed record, numbering starts with the first non-owner source. A platform blue tick is shown separately. Keep these labels with the data when you display it.

Do you promise the site will always be available?

We aim to keep the free service available, but we do not promise a set uptime. If your service needs a formal uptime agreement, please contact us to discuss it.

How do I report a record that is wrong or is the wrong person?

Use Correct this record on the page. There is no reporting API in this version. Our systems and team review reports before making changes. A record about the wrong person may be held back while we check it. If your system finds many related errors, email support@sparkonomy.com with the pattern. You do not need to send a separate form for every example.

Is there a tool or example to test my software against?

There is no hosted tool to validate declarations yet. The For AI & Search page has worked examples you can test against. They cover Allowed, Conditional, Prohibited, mixed choices, an unclaimed record and a removed record. https://creatorgraph.sparkonomy.com/for-ai-and-search/

Do I need to name OCG as the source?

Yes. Put “The Open Creator Graph by Sparkonomy” beside the data. Add a live link to the record. Keep the confidence labels and AI consent status with each value. This helps readers see what was checked and what the Creator stated.

What should I do with my saved copy when AI choices change?

If you show live data, you see the current record. If you keep a copy, check it at least every 30 days. Act sooner if you learn that the Creator has made their terms stricter. If you see HTTP 410 Gone, remove your copy at once. The 30-day rule is not a reason to ignore a change you already know about.

Can I get notified when a Creator’s consent changes?

There are no change alerts, webhooks or subscription feeds yet. Show the live record or check your saved copy within 30 days. If you need the latest terms for each use, read the live record.

Will the data fields change? How will I know?

Fields and allowed values may change as OCG grows. The public changelog records changes and versions. We add new fields where possible rather than changing existing ones. We give 90 days’ notice before removing an existing field or changing its purpose. Check the changelog: https://creatorgraph.sparkonomy.com/changelog/