Critics of language model copyright protection often make the argument that such content production by artificial intelligence harms existing content market suppliers.
But some of us might note that this also is a trend and theme quite familiar to all digital content processes and their impact on existing content business models. One example is the impact of AI summaries on search traffic volume for content suppliers.
Phenomenon / study | What is happening | Evidence of substitution / economic impact | Implication for profitability |
Google AI Overviews / AI search | Search engines increasingly answer questions directly rather than simply providing links | Reuters Institute/Chartbeat data show Google organic-search traffic to 2,500+ sites fell 33% globally and 38% in the U.S. between Nov. 2024 and Nov. 2025. Publishers expect search traffic to fall another ~43% over three years. (reutersinstitute.politics.ox.ac.uk) | Negative. Less referral traffic means fewer ad impressions, subscriptions and affiliate conversions. Particularly damaging to publishers dependent on search. |
Wikipedia + Google AI Overviews | AI summaries can satisfy the user's information need without a click | A 2026 causal study of 161,382 article-language pairs found Google AI Overview exposure reduced English-Wikipedia traffic by about 15%, with larger effects for topics where short answers are sufficient. (arXiv) | Strong evidence of actual content substitution, rather than merely correlation. |
French publishers / Google AI summaries | Publishers argue AI-generated summaries are replacing visits to original articles | The French press association says regulator Arcom data indicate AI summaries have produced a 33–38% decline in traffic to media sites. The association is seeking competition action and compensation. (Reuters) | Potentially serious threat to advertising-supported journalism; also creates pressure for AI licensing revenue. |
LLMs as news destinations | Consumers increasingly obtain news directly from chatbots | Reuters Institute reports that weekly generative-AI use in six markets rose from 18% to 34% between 2024 and 2025. Its 2026 research finds a growing, though still minority, use of chatbots for news. (reutersinstitute.politics.ox.ac.uk) | Potentially more important over time: AI moves from being a distribution intermediary to being the destination. |
Publisher AI production | Publishers increasingly use AI for back-office automation, newsgathering, coding and content production | In the 2026 Reuters Institute survey, 97% of publishers regarded back-end AI automation as important; 82% cited newsgathering and 81% coding/product development. But only 44% said AI initiatives were showing promising results, versus 42% calling results limited. (reutersinstitute.politics.ox.ac.uk) | Positive cost effect, but so far not a demonstrated profitability windfall. |
AI actually increasing publisher content volume? | One might expect near-zero-cost generation to create enormous increases in articles | A study of large publishers finds no evidence that publishers increased text volume following GenAI adoption. Instead they increased rich content, advertising and targeting technologies. (arXiv) | Important counterexample to the simple "AI = infinite content" thesis. Professional publishers may be recognizing that additional generic text has little economic value. |
AI and publisher traffic | LLM bots consume publisher content while potentially sending fewer readers back | The same study finds a moderate decline in publisher traffic after August 2024. Interestingly, publishers that blocked GenAI bots subsequently experienced 23% lower total traffic and 14% lower real-user traffic than comparable publishers that did not block them. (arXiv) | Shows the relationship is complicated: AI can be both a threat and a discovery mechanism. |
AI-assisted social-media creation | AI makes it much cheaper to create posts, comments and other social content | A controlled experiment with 680 U.S. participants found some AI tools increased engagement and content volume, but also reduced perceived quality/authenticity and generated negative spillovers in conversations. (PubMed) | Supply explosion is real, but more content does not necessarily mean more economic value. |
AI-generated social posts | AI can produce content that competes directly with human-created material | A 2025 study found GPT-4-generated social-media posts could outperform human-written posts in engagement; another cross-platform study examines comparable performance on Facebook, Instagram and X. (ScienceDirect) | Potentially disruptive to the labor economics of content creation, especially routine marketing/PR content. |
AI disclosure / authenticity | Consumers don't necessarily value AI-created material as much as human-created material | A 2026 study found labeling content as AI-generated or AI-enhanced reduced affective and behavioral engagement relative to human-created content, especially for emotional content. (DOI) | Creates a possible scarcity premium for human/original content as AI content becomes abundant. |
AI use in newspapers | AI-generated material is already entering professional media | An audit of 186,000 articles from 1,500 U.S. newspapers estimated about 9% were partially or fully AI-generated in summer 2025. (arXiv) | Demonstrates that substitution of human content production is already occurring, particularly in smaller/local outlets. |
AI licensing | Publishers are attempting to turn substitution into a new revenue stream | Reuters Institute found 36% of publishers expected licensing income from technology/AI companies to become significant. Brookings describes AI licensing as a new layer of the media economics historically dominated by search platforms. (reutersinstitute.politics.ox.ac.uk) | Could partially offset lost advertising/search revenue, but licensing revenue is not yet comparable to the scale of displaced traffic. |
Premium vs. commodity content | AI has much greater ability to substitute for routine informational content than differentiated reporting | Reuters Institute finds subscription/membership-oriented publishers with strong direct traffic have a clearer path to profitability, while advertising-dependent publishers are much more worried about AI search. (reutersinstitute.politics.ox.ac.uk) | Suggests bifurcation: commodity information gets cheaper; original reporting, brands, personalities and communities become more valuable. |
Much of the argument about regulating artificial intelligence training and output has to do with the efficiency with which computers work, compared to biological limitations humans have doing the same things.
That might strike some of us as an odd argument. Of course, the real issue is not about how humans or machines learn, or even how quickly they can produce original new work.
The issue, as often is the case, is about the effect on markets for content. And that is what the authors of a new paper suggest is the case.
source: Tuhin Chakrabarty, Xinyue Liu , Jane C. Ginsburg, Paramveer Dhillon
Comparing best-selling books with no AI content to books with light AI content or wholly-AI produced on Amazon, the authors suggest the AI books are having an impact on non-AI book sales and revenues.
source: Tuhin Chakrabarty, Xinyue Liu , Jane C. Ginsburg, Paramveer Dhillon
When a human author reads hundreds of mystery novels, internalizes their mechanics, and writes a new mystery novel using those structural lessons, copyright law views this entirely as lawful inspiration and learning.
Supporters of applying stricter copyright rules to AI model content typically are based on the argument that humans are relatively slow learners, while computers are fast.
Just as a human author synthesizes everything they have ever read to draft a novel, an AI synthesizes the patterns learned from its training data to generate new text.
Critics essentially argue AI models should not receive the same level of protection because they are “too efficient,” which is a new argument in the copyright domain.
If copyright law were applied with strict, absolute functional consistency, the legal outcomes for human and AI generation would look remarkably similar:
Process Equivalence: Both human brains and AI models consume existing works to extract abstract patterns, rules of grammar, and stylistic conventions.
Output Evaluation: If an AI generates a completely novel story that merely employs general tropes and stylistic patterns learned during training (without plagiarizing specific passages), a consistent legal framework would view it the same way it views a human-written work.
Proponents of AI model “freedom to create” argue that training is inherently transformative. The AI is not being trained to reproduce the books it reads; it is learning how language works. So that is fair use.
Critics argue AI should not be protected in the same way humans are because the machines are so much more efficient.
Dimension | Human Authors | AI Models |
Legal Status of "Reading" | Lawful (cognitive processing falls outside copyright). | Contested (involves digital copying; subject to ongoing fair use litigation). |
Legal Status of Output | Protected, provided it avoids literal copying or plagiarism. | Contested, with questions regarding authorship, originality, and market substitution. |
The point is that AI-produced content will probably have a similar impact to existing content suppliers as we have seen with both digital and internet content markets.
There will be some amount of disruption; severe disruption in at least some instances.
Consider what happened to business-to-business content businesses such as specialized trade media.
Historically, trade journals relied on a print-centric, monopoly-like model where niche B2B advertisers had virtually no other way to reach specialized professional audiences. But what happened was more than a shift from physical media to online and digital formats.
Advertising budgets massively migrated away from print. In 1995, specialized print trade journals commanded nearly the entirety of B2B advertising budgets.
By the late 2000s, digital channels achieved parity, and today, digital and online formats capture the vast majority of B2B marketing spend.
Year | Legacy Print Share (%) | Digital Online Share (%) |
1995 | 98% | 2% |
2000 | 88% | 12% |
2005 | 65% | 35% |
2010 | 35% | 65% |
2015 | 15% | 85% |
2020 | 8% | 92% |
2026 | 4% | 96% |
AI-enabled changes might have a range of effects, some quite negative for legacy content providers but also some positive changes as well for others.
Business/content type | Effect of AI content abundance | Likely long-term economics |
Commodity news | Very high substitution | Worse |
Weather, sports scores, financial quotes, basic facts | Very high substitution | Much worse |
SEO articles / "10 best..." content | Very high substitution | Worse |
Generic marketing copy | High substitution | Lower labor cost, potentially higher margins |
Social-media posts | High substitution | More supply; declining value per post |
Local routine journalism | High production substitution | Lower costs but potentially weaker differentiation |
Original investigative journalism | Low direct substitution | Scarcer and potentially more valuable |
Celebrity/personality content | Low-to-moderate | Human identity becomes an asset |
Video/entertainment | Moderate initially | More content, but attention remains scarce |
Strong media brands | Moderate substitution, strong defensive value | Potentially resilient |
Communities / memberships | Low substitution | Potentially increasingly valuable |
Proprietary data/research | Low substitution | Potentially more valuable |
Human-authored expertise/authenticity | Potentially negative supply effect but positive scarcity effect | Could command a premium |
No comments:
Post a Comment