- Members usually own their posts. What you can license depends on the license your terms of service took from them, and whether it can be passed on.
- Reddit takes a perpetual, sublicensable license; Stack Overflow takes one under CC BY-SA 4.0, which requires attribution. Read your own terms against those.
- The FTC warned in February 2024 that starting AI training and telling users only through a quiet, retroactive terms change may be unfair or deceptive.
- Reddit disclosed $203.0 million of data licensing contracts in its 2024 IPO filing. No buyer has published a price for a small or niche forum archive.
Who owns forum and community content, and what did members grant?
Start with your terms of service, because they are the license you hold. On most forums the member owns what they write and grants the site a license to host and show it. Whether you can license those posts to an AI company turns on three words in that grant: is it sublicensable, transferable and broad enough in use? The table sets the main rules side by side.
| Rule or source | What it says | What it means for you | Date checked |
|---|---|---|---|
| Reddit User Agreement | Members "retain any ownership rights" and grant a "worldwide, royalty-free, perpetual, irrevocable, non-exclusive, transferable, and sublicensable license" | An example of terms written to allow licensing posts onward. Compare your own grant word by word | 20 Nov 2026 |
| Stack Overflow public terms | Posts are licensed to the site "pursuant to Creative Commons licensing terms ( CC BY-SA 4.0 )" | A share-alike license allows reuse but requires attribution, which an AI license has to respect | 20 Nov 2026 |
| Discord Developer Policy | "Do not use message content obtained through the APIs to train machine learning or AI models" without Discord's permission | A Discord server owner cannot export messages by bot and license them for training | 20 Nov 2026 |
| FTC staff post on terms changes | Telling users of new AI training use only "through a surreptitious, retroactive amendment" may be unfair or deceptive | Changing your terms quietly the week before a deal is a risk. Give real notice | 20 Nov 2026 |
Sources: the Reddit User Agreement, Stack Overflow terms, Discord Developer Policy and the FTC staff post, read on 20 November 2026. This page is general information, not legal advice.
Can a forum owner license posts to AI companies?
Yes, when three things line up. Your terms take a license from members that you are allowed to pass on. The use you offer, such as training, falls inside that license. And the personal data in the posts can lawfully go along, or comes out first. Most hobby forums, trade communities and Q&A sites were set up with short terms that never mention third parties, let alone AI. That does not end the matter, but it means the first job is a lawyer's reading of every version of your terms since launch.
Versions matter because posts were made under the terms in force at the time. A forum that added a sublicensing clause in 2022 may hold a clean grant for later posts and a weaker one for earlier posts. Keep an archive of every version and the date it took effect, and be ready to license by date range.
Members in the UK or the EU add a data protection test. Personal data must be collected "for specified, explicit and legitimate purposes" and not further processed in a way incompatible with them (UK GDPR Article 5; the EU GDPR has the same principle). Licensing posts for AI training is a new purpose, so a forum with European members needs a lawful basis and real notice, or it removes the personal data first.
Content you created yourself is the easy part. Staff articles, moderators' guides written as paid work, and official answers by your own team usually belong to the site and can be licensed on the same basis as any company document.
A forum owner who wants payment each time an AI bot reads a thread can set that rate through TollBit; Redpine applies per-use pricing to research content instead.
Who owns customer community forum posts, and can they be licensed?
A customer community is the support forum a software or hardware company runs for its users. The company usually wrote the terms, moderates the threads and posts official answers, so it often holds a broad license. The catch is what customers wrote. Support threads carry configuration details, error logs, account names and sometimes confidential information about the customer's own business.
Two checks apply on top of the forum terms. First, your customer contracts: an enterprise agreement may treat anything the customer shares with you as confidential, wherever it was posted. Second, personal data in threads, such as names, emails and signatures. The guide to NDAs and customer contracts in data licensing covers the first; de-identifying company data before licensing covers the second.
The most useful part of a customer community is often the official answer marked as the solution, written by your own support engineers. That text is yours, and a thread with a question, failed attempts and an accepted fix is the shape AI companies want.
Can we sell our Q&A data to AI companies?
Q&A data has the clearest public precedent. On 6 May 2024 Stack Overflow announced that OpenAI would use its OverflowAPI product "to improve model performance for developers who use their products". What makes Q&A valuable is structure: a question, several answers, votes, an accepted answer and edits over time. That is a ready-made record of which answer experts judged correct.
The same deal shows the risk. The Verge reported on 8 May 2024 that members who changed their posts in protest found the site "reversing the alterations and suspending the users who carried them out". Stack Overflow's terms let it do that, because the license members grant is perpetual and irrevocable. A smaller community with weaker terms and closer ties to its members should expect pushback and plan the announcement before the deal, not after.
Attribution is the other Q&A issue. Content under a share-alike license such as CC BY-SA carries an attribution duty. Agree with the buyer how that duty is met, for example by keeping post URLs and author handles in the delivered data.
What is a forum or community archive worth to AI companies?
The largest public figures come from one company. Reddit's IPO prospectus, quoted by TechCrunch on 22 February 2024, said it had entered "data licensing arrangements with an aggregate contract value of $203.0 million and terms ranging from two to three years". TechCrunch also noted reports of one agreement "worth about $60 million on an annualized basis". Reddit sells scale: a very large archive and constant new activity.
No buyer has published a price for a small or niche forum archive. A forum with a few million posts is not a small Reddit, and its price will not be a fraction of Reddit's. What it can offer is depth on one subject that general web text lacks: years of threads from machinists, nurses, pilots, growers or repair technicians solving real problems.
What moves value: how specialized the topic is, the share of threads with an accepted or upvoted answer, how much text is original rather than quoted or pasted, years of history, and how clean your rights are by date range. A license limited to evaluation can also suit a forum that does not want its content trained on. See data licensing pricing models for how deals are usually structured.
How do you export Discord data, and can a Discord server be licensed?
Discord's own export is for one person's account. A member requests it in User Settings under Data & Privacy, and Discord says "the data request process can take up to 30 days to complete". It is not a server-wide export, and it is not a tool for collecting other members' messages.
For servers, the binding rule is Discord's Developer Policy, which says not to "use message content obtained through the APIs to train machine learning or AI models (including large language models) unless express permission is granted by Discord". A server owner who pulls history with a bot and licenses it for AI training is outside that policy. Treat Discord community content as not licensable for training unless Discord itself agrees.
Self-hosted forums are different. On Discourse, site admins can create a backup that, in Discourse's words, includes "the full site database, which contains everything on the site: topics, posts, users, groups, settings, themes". That file also holds members' emails and IP addresses, so the licensed export should be built from it, not handed over as it is.
Sources: Discord data requests, the Discord Developer Policy and Discourse backups, read on 20 November 2026.
How do you license forum and community content for AI training?
We help you license your data. Origin Data Partners is an independent data partner that helps business owners value and license their company records: the company pays Origin nothing, and a buyer may pay Origin when a deal completes. A community owner's side of the work usually looks like this.
- Collect every version of your terms of service and privacy policy, with effective dates, and have a lawyer read the license grant in each.
- Decide what stays out: private messages, deleted and edited-away posts, posts by members who closed their accounts, and anything flagged for personal data.
- Give members real notice of the new use before any delivery, and consider an opt-out by member. The FTC has said a quiet, retroactive terms change may be unfair or deceptive.
- Build the export from your database: thread, post, timestamp, votes, accepted-answer flag and a stable post ID, with emails, IP addresses and signatures removed.
- Offer a sample of a few thousand threads in one subforum so a buyer can judge quality. Check any offer against what to review in a data licensing offer.
When you know what you hold, see what your community content may be worth. The estimate starts from a description of the forum, not an export. Not the owner? Advisers can introduce the company.
How we checked this
We read every source on this page on 20 November 2026: the platforms' own terms and developer policies (Reddit, Stack Overflow, Discord), Discord's and Discourse's official help pages, the FTC staff post on terms changes, the UK GDPR on legislation.gov.uk, the companies' own deal announcements, a law firm's report of the Reddit remand order, and reporting by AP News, TechCrunch and The Verge. The FTC page returns an error to scripted checks, so we saved the text we read.
What we could not verify: the price of any forum license other than Reddit's disclosed totals, the terms of Reddit's individual deals, and whether any small or niche forum has licensed its archive, because we found no public record of one.
Which published deals and disputes should forum owners know about?
Three public facts set the market for community content. Reddit has licensing agreements with Google, OpenAI and other companies, as AP News reported. Stack Overflow licensed API access to OpenAI in May 2024. And Reddit has gone to court against a company it says used its content without a deal: AP News reported that Reddit sued Anthropic, "alleging that it is illegally 'scraping' the comments of millions of Reddit users to train its chatbot Claude". That case is still open: on 30 March 2026 a federal court sent it back to California state court, holding that the Copyright Act does not preempt Reddit's state-law claims, and we found no final judgment as of 20 November 2026.
The pattern is the same as in news: licensing and lawsuits side by side, and buyers who want proof of rights. Names here are dated market facts, not recommendations. For the full list of announced deals across data types, see the AI data licensing deals tracker; for how content licensing works generally, see AI licensing explained.
Sources: Stack Overflow, TechCrunch on Reddit, AP News on the Reddit lawsuit and Loeb & Loeb on the remand.
Origin Data Partners works with companies that own data and helps get it bought. We are independent of the services on this page, and a buyer may pay us when a deal closes (how Origin is paid). See what your data is worth.
Common questions
Can a forum or online community license its posts to AI companies?
Often, yes, within the limits of its terms of service. Members usually keep copyright and grant the site a license, so the site can license posts to an AI company only if that grant can be passed on and covers the use. Private messages, deleted posts and personal data need separate handling. Give members real notice before any delivery rather than changing terms quietly.
Can an app developer sell user-generated content to AI companies?
Only within the license users granted in the app's terms, and only with privacy law satisfied. Reviews, comments and community posts are usually owned by the users who wrote them. If the terms take a sublicensable license broad enough to cover AI training, the developer can license that content; if not, it needs new consent. Data collected for one purpose cannot simply be repurposed without notice.
What is a user-generated content license?
It is the permission users give a platform, usually in the terms of service, to host, copy and show what they post. Reddit's version, for example, is worldwide, royalty-free, perpetual, irrevocable, non-exclusive, transferable and sublicensable. The words "sublicensable" and "transferable" decide whether the platform can pass the content to someone else, such as an AI company, under its own license.
Can we include private messages or deleted posts in an AI license?
Leave them out. Members wrote private messages expecting only the recipient to read them, and deleted posts reflect a choice to withdraw. Including either invites privacy complaints and loss of trust, and most terms of service never covered that use. Build the licensed export from public, live posts only, and apply deletions that happen after the deal to future deliveries.
Do we have to tell members before licensing the forum to AI?
Telling them is the safer course. FTC staff wrote in February 2024 that adopting a new practice such as AI training and informing users only through a surreptitious, retroactive change to the terms may be unfair or deceptive. Clear notice before delivery, an explanation of what is shared, and an opt-out for members who object reduce both legal and community risk.
Can members opt out of an AI license of the forum?
They can if you build it in, and many communities should. An opt-out list keyed to member IDs lets you drop those members' posts from the export before each delivery. Under a perpetual, irrevocable grant like Stack Overflow's, members cannot withdraw their content, but a smaller community depends on goodwill, and an opt-out costs little against the trust it keeps.
Sources and further reading
- Reddit User Agreement
- Stack Overflow Public Network Terms of Service
- Stack Overflow: Stack Overflow and OpenAI partner (6 May 2024)
- The Verge: Stack Overflow users protest the OpenAI deal (8 May 2024)
- TechCrunch: Reddit says it has made $203M so far licensing its data (22 February 2024)
- AP News: Reddit sues Anthropic over scraping
- Loeb & Loeb: Reddit, Inc. v. Anthropic PBC, case update (N.D. Cal., 30 March 2026)
- Discord Developer Policy
- Discord Support: Requesting a copy of your data
- Discourse Meta: Create, download and restore a backup of your Discourse database
- FTC Office of Technology: Quietly changing your terms of service could be unfair or deceptive (13 February 2024)
- legislation.gov.uk: UK GDPR, Article 5, principles relating to processing of personal data
