The Attribution Crisis

We are confronted with a paradox of the digital age: we share more information than ever, yet almost nobody remembers where it came from. In an era of likes and retweets, ideas flow unchecked through social platforms as if unowned. LinkedIn’s official policy still assures creators that “you retain ownership” of what you post, and their transparency reports show thousands of DMCA notices resolved each half-year. But the missing piece is provenance. When a human author’s insights are casually copied, lightly reworded, or embedded in an AI prompt, platforms neither enforce attribution nor publicly connect the derivative content back to its origin.

This isn’t simply a quirk of LinkedIn; it’s a systemic problem in modern knowledge sharing. Social networks reward content distribution — likes, engagement, “views” — far more strongly than they reward intellectual lineage. In practice, this means the first publisher of a high-value insight often cedes audience to dozens of others who simply echo the idea. Nobody gets a “knowledge badge” for being the originator of a viral thought. In fact, one LinkedIn poster, stunned to see their account locked after reposting someone else’s content, observed that “X allows and even encourages people to repost” — and yet enforces copyright arbitrarily when users complain.

The result is a world where ideas drift without credit, and provenance disappears. We may trace a Git commit or a scientific citation precisely back to its root, but who can trace a pithy LinkedIn post in April to its true ancestor in January? In practical terms, knowledge loses its accountability and audit trail.

Copyright, not credit

Every platform we use today has some handle on copyright. LinkedIn, X, Medium, YouTube and others all accept DMCA takedowns and publish transparency reports. LinkedIn’s latest transparency report shows nearly 3,000 infringement requests per half-year (January–June 2025), with roughly 94% resulting in content removal. YouTube’s Content ID system quietly generated 722 million copyright claims in the first half of 2021. Even GitHub — used mainly for code — insists on explicit licenses or defaults to “all rights reserved”.

But legal ownership is not the same as academic or ethical attribution. On these platforms:

• LinkedIn offers a “share” button that shows the original author when you use it. But if someone copy-pastes your post text into a new post, LinkedIn will not alert you or flag it. Nor will it penalize the copier unless a DMCA notice is filed.
• X (Twitter) supports retweeting with author credit, yet millions of users also screenshot or copy text and re-post it without credit. Technically every one of those is copyright infringement, but enforcement is sporadic, and users routinely profit in attention from it.
• Reddit has no universal built-in share mechanic. Cross-posting or quoting often relies on manual curation, and moderators may remove posts reused without credit — but this is ad hoc.
• Medium allows authors to canonically link a story they republish via its import tool, which preserves search ranking. Other authors, on or off the platform, can still copy text without attribution.
• YouTube actively hunts reuploads via Content ID, and embeds always show the original channel. Short clips or transcripts still slip through.
• GitHub, by contrast, is almost the ideal model: code sharing requires a license declaration, and the platform provides transparent version control. Every fork and merge tracks commit authors. Copying someone’s code without attribution is a clear violation of license terms.

In short, the law cares about ownership, not about honor, credit, or trust. Platforms ensure copyright holders can issue takedowns, but they do not actively cultivate a culture of giving credit where due. The few built-in mechanisms — retweeting, sharing, canonical linking — are easily bypassed by copy-paste or by reformatting the idea. The outcome is visible: on many networks, copies of original content can out-engage the original. The person who first built the insight often disappears from sight.

Incentives and engagement

This attribution gap exists for a reason: modern social networks are optimisation machines. They tune metrics to maximize engagement above all else. As I’ve argued elsewhere, social media platforms optimise engagement instead of healthy discourse. Every one of these systems behaves exactly as it was incentivised to behave. Attention is the currency, and any content strategy that attracts eyeballs will be rewarded. Copying a popular post verbatim or with a catchy headline is easier than creating something new — and it can be just as effective, or more, in generating likes.

Ironically, Instagram recently announced it will prioritise original content and down-rank straight reposts, precisely to break the cycle of “clickbait villages” where recycled posts get disproportionate reach. Networks like LinkedIn or X have not made such moves in any systematic way. In practice, viral recyclers gain attention, while originators pay the cognitive and time cost of creation.

The optimisation trap is real: platforms don’t distinguish between “original story” and “aggregator repost,” so long as both keep users scrolling. Creators have no easy way to signal “I’m the originator,” nor does the system highlight that fact. Conversely, there is no incentive in the model to cite sources or link back. In a Goodhart’s Law scenario, the metric — engagement — became the target, and attribution gets discarded as noise.

When AI eats your ideas

This problem is magnified by AI. Generative models train on massive pools of Internet text, typically without revealing who wrote each sentence. The Sovereign Context Protocol paper describes exactly this: LLMs consume vast quantities of human-generated content, yet the creators of that content remain largely invisible in the value chain. Even if an AI regenerates or paraphrases your blog post, you see no citation, and the user may not even know the text is AI-derived.

We already witness forks in idea provenance: an original blog → LinkedIn post → tweet thread → newsletter → AI recap → new LinkedIn rant. By the time it comes back around, nobody recalls the starting point. One consequence is cultural: people do blame AI for plagiarism, without recognizing that the AI itself learned from uncredited human work. One study found participants judged plagiarism from an AI — or a friend — as less immoral than copying from an unknown blogger, especially when they perceive the AI as providing silent permission. In other words, the source changed: copying a ChatGPT output feels looser morally than copying a person.

The legal battles underscore how urgent the attribution gap is. The New York Times and the Authors Guild have sued AI firms for scraping copyrighted journalism. The pending EU AI Act transparency obligations even mandate some disclosure about training data. But courts and regulators move slowly. Meanwhile, the actual pipeline of knowledge delivery runs wide open, untethered.

Provenance in other domains

Every serious knowledge system has built-in lineage. Academic publishing frowns on a fact or idea without citation. Every math proof, every scientific result, and even patent applications all trace back to prior work. The financial world uses auditable ledgers and blockchains; journals require references; engineers use version control. These systems share a key insight: information is only as reliable as its source.

By contrast, social media is source-blind. No forensic trail exists for a viral post or a clever tweet. Even worse, the platforms doing the amplification have a vested interest in crowding out clear authorship. As one contributor put it in a discussion of social-media incentives: social platforms often behave as if they share a teenager’s concept of copyright — which is to say, they don’t enforce it thoughtfully.

Learning from successful provenance systems, we see potential solutions. Git’s model — branches, commits, merges — explicitly logs every contributor. Digital libraries and Wikipedia require sources, enabling checks on truth. Even blockchains conceptually prove who issued a transaction or token. Could we imagine something analogous for social media ideas? Some researchers propose cryptographic signatures or invisible watermarks for text, or protocols such as the Sovereign Context Protocol, where every access to a creator’s content is logged with identity and terms. These ideas aim to make attribution a default property of data access, rather than an afterthought.

Toward provenance-aware platforms

We need to treat ideas like assets with lineage. That means combining policy, design, and technology.

Platform changes

Networks should make attribution easy and visible. A “share” or “cite” button should be as prominent as reposting text. If you repost a LinkedIn post, the original author’s name could appear by default — like nested comments do — rather than burying it in a link. Algorithms could boost signals of originality: if a post matches another earlier post exactly or nearly so, rank the original higher, or at least display both side by side. Instagram’s test is one model: its feed will label and even replace reposts with originals. LinkedIn or X could do similar cross-checking.

Automatic detection tools

Platforms already run virus scanners and content filters. Why not text-scanning tools? Just as YouTube’s Content ID looks for video matches, we could develop a “Content ID for words” that flags probable plagiarism — exact matches or paraphrases from known sources — and prompts attribution. This could even be community-driven, like duplicate detection on Stack Overflow, to protect high-quality posts.

Support open protocols

The Sovereign Context Protocol idea invites an open standard. Social networks and AI services could adopt a Content Provenance Protocol akin to Anthropic’s MCP, where each piece of content carries metadata: original author, license, and a usage log. LLMs could be required to include provenance strings — like emerging watermarking research — whenever they generate text from training data. This is a systems-level intervention, but it aligns with trends in regulation around data transparency.

Creator best practices

On the user side, authors must guard their own provenance. Like scientists, they can explicitly license content — Creative Commons BY, for example — to require attribution. They might embed signatures, timestamps, or steganographic codes in images or text. Networks of journalists have begun inserting hidden copyright lines or requiring clients to register posts. While imperfect, these actions set norms: please credit the author and link when resharing. A clear attribution culture, even if enforced only by peer pressure, matters.

Education and norms

Finally, we can’t underestimate the role of norms. Guides such as the Michigan Online Handbook remind creators that copying without credit damages trust and is essentially intellectual theft. Campaigns to raise awareness — as Wikimedia does for contributors — could teach professionals that every LinkedIn post is somebody’s IP and deserves citation if reused. If even a fraction of users started adding a courtesy link or author tag, momentum could shift.

How could we measure success?

To know if these ideas work, we need metrics. Possible experiments and KPIs include:

• Original-to-copy ratio — track the fraction of high-engagement posts that match earlier content. A drop after an intervention (for example an algorithm change) would indicate more credit to originators.
• Engagement redistribution — run A/B tests where one feed orders posts strictly by timestamp (rewarding originality) and another by popularity. Measure whether original authors get a larger slice of likes and views.
• Creator participation surveys — just as scientists cite more when credit matters, measure whether creators publish more or less when they feel protected. A longitudinal study could ask top LinkedIn writers: since the new policy, do you feel safer sharing novel content?
• Plagiarism incidents — platforms could report not just copyright takedowns, but flagged plagiarism cases. An increase in detected copies resolved by attribution instead of removal might even be good evidence that people are now crediting sources properly.
• AI-usage metrics — if LLMs adopt a provenance layer, one could track how often a given user’s content is retrieved by the model. Similar to views on Wikipedia’s edit history, creators could see real-time counts of their idea’s usage. Growth in these numbers — and creator compensation — would signal a functioning system.

A timeline of an idea’s journey would expose the drift graphically: original post → reposts → AI mentions → a later article — all without credit — until finally someone asks, “Wait, who first said this?”

Recommendations

1. Build attribution into the UX. Platforms should make credited share as easy as copying. Show the original author prominently when users share. Downrank or label near-duplicate reposts. Instagram’s originality push is a proof of concept.
2. Detect and nudge. Deploy similarity checks. When a post matches earlier content, prompt the poster: this looks like content from [Author] — want to credit them? For automated pipelines, flag networks of copying.
3. Open provenance protocols. Adopt standards like the Sovereign Context Protocol across services. If every content fetch requires logging source and time, we build a shared audit log for ideas. Encourage or require LLM vendors to use such APIs for training and data retrieval.
4. Creator safeguards. Writers should watermark or license their work. Use Creative Commons BY or other terms that require attribution, and consider hidden digital signatures. In LinkedIn posts, one might even add: © [Name], share with link back.
5. Measurement and incentives. Platforms should publish attribution metrics — for example the percentage of posts with clear sources, or engagement distribution between originals and copies. Run experiments to see whether rewarding originators changes behaviour. Reserve badges or status for verified originators only if they consistently post original work.

We can’t fix what we don’t measure. Possible metrics include original-author share of engagement over time, user-reported satisfaction, and the number of automated attributions. Experiments could temporarily enable citation prompts for a subset of posts, or simulate a feed with watermarks against a normal feed.

Conclusion

In a knowledge economy, ideas are currency, and currency without a ledger is easily counterfeited. LinkedIn and its peers inherited the DMCA system — copyright — but not a scientific peer-review system — attribution. Until we build provenance into our platforms, we’ll continue to see the shadow of anonymous content generation: viral posts whose birth parents are long forgotten.

The fix is not censorship, but engineering: design features that trace and display authorship, incentives that reward originators, and protocols that log every share. Think of it as putting brakes and GPS on the runaway copy machine of social media. We will never go back to an era where complex systems hide all their decisions — we seek transparency in AI, finance, law, and even pipelines of electricity. Why should knowledge be an exception?

By reframing the issue, we see this as a systemic provenance problem, not merely one of etiquette. It asks us to extend the philosophy of evidence and lineage into the social web. A Library dedicated to understanding before technology demands we recognize this gap: ideas need lineage as much as data needs a checksum.

Ultimately, restoring provenance is not only fair, it’s practical: trust in information is eroded when authorship is obscured. Every scientific paper, every piece of code, every audited system relies on a chain of custody. Now our conversations and shared knowledge must catch up. If we fail to do so, we create fertile ground for misinformation, dispute, and disengagement — a world where we see trees of content but lose the forest of facts.