Employees at large companies feed externally published content into AI tools 11 times a week, and the outputs built on that material reach an average of 96 people each time, according to a survey of 570 knowledge workers that CCC and Outsell released on September 22, 2026. The people who know the copyright rules best, senior executives, generate the most exposure of all.

In Short

A copyright licensing company and a research firm asked 570 office workers at big companies how they use articles, reports and other published material, and found they put that material into AI tools about 11 times a week before sharing what the AI produces with around 96 people. That matters to publishers, whose work is being copied this way, and to companies, whose content licenses were mostly written for people reading and forwarding articles rather than for machines. Bosses turn out to be both the best informed about the rules and the biggest source of risk, which is why the study argues copyright has to become part of how companies govern AI.

Two channels instead of one

The 2026 Copyrighted Content Usage Trends Report is the latest edition of a series CCC, the Danvers, Massachusetts-based licensing intermediary also known as Copyright Clearance Center, has run with Outsell since 2007. The partners track a number they call the velocity of content: how often employees share work material, with how many people, and how much of it comes from outside the organization. Multiplied together, those three means produce what the report describes as "potential unlicensed sharing instances per employee, per week."

That figure now stands at 73. In 2016 it was 17.8, which makes the current reading a little over four times the earlier level, and in 2025 it was 66. According to the report, employees shared content 8.2 times a week with 19 people in 2026, against 8.1 times with 17 people a year earlier. Frequency barely moved. The audience for each share grew. The share drawn from external sources slipped from 48% to 46%, so the rise came from reach, not from greater reliance on outside content.

Where does that sharing happen? According to CCC, 75% of it runs through collaboration platforms or email attachments, and 2026 is the first year in which collaboration platforms rank as the preferred channel. The shift has been building for years. In February 2023, CCC reported that the share of respondents preferring tools such as Microsoft Teams, Slack and Google Workspace had doubled from 16% in 2020 to 33%, and that executives shared third-party content at higher rates than middle managers or individual staff, a pattern the 2026 edition repeats with sharper numbers.

Why did traditional sharing keep rising as AI spread? The report's hypothesis is easier discovery: employees now name 11.8 publications as critical to their jobs, a third straight annual increase. "The more varied the content, the less workable it is to clear rights work by work," the report states.

One caveat applies to the history. The text says the velocity eased "in the middle of the decade" before climbing again, yet the chart plots only 2016, 2025 and 2026, and the intermediate readings that would show the dip are not published.

The AI layer

The traditional measure captures content passing between people, not content passing into machines. To fill that gap, the 2026 edition adds an AI velocity of content, built from three numbers and one audience breakdown.

  • AI input velocity: externally published content is used with AI 11 times per employee per week.
  • AI output velocity: outputs built on that content are shared or automatically distributed 12.5 times per week.
  • AI output reach: each output reaches 96 people on average, roughly five times the 19 people reached by a traditional share.
  • Audience exposure profile: among respondents who share such outputs, 91% share internally, 51% with external parties and 26% with the public.

The output figure, higher than the input figure and absent from the press release, combines manual sharing with automated distribution without separating them.

Adoption is no longer the variable being measured. According to the report, 96% of respondents use AI tools and 73% use them daily. Eighteen months earlier, about a quarter said they never or rarely used AI at work; that group has fallen to 6%.

How content goes in

According to the report, 66% of respondents upload files or documents, so most content enters AI as full text, and 60% paste text directly into the tool. "Both are acts of copying: the work itself, in whole or in part, goes into the tool. Reproduction is the act copyright governs most directly," the report states. Summarizing articles and content sets is now the second most common AI task, cited by 46%, up from fourth place.

The framing echoes litigation at a very different scale. CNN's complaint against Perplexity, filed in May 2026, argued that a retrieval system copies content twice, on ingestion and again when serving an answer. The CCC data describes the same two stages inside ordinary companies.

External material rarely travels alone. Many tasks blend it with internal documents in the same prompts or outputs, and a work can "enter AI in one form and reappear in an output separated from the source that shaped it," according to the report. Even internal sharing may require permission, the report states, "particularly where the outputs include protected expression or effectively substitute for source material." That is a broader test than verbatim reproduction, the failure researchers call regurgitation, which has become central to the fair use fight with model developers.

A figure that reads two ways

The report's own summary contains an inconsistency. One page states that outputs "go out 12.5 times a week to 96 people, about half of them outside the company." The 51% figure does not support that reading. It is the share of respondents who typically send outputs to external parties, a multiple-response measure alongside the 91% who share internally, not the proportion of the 96 recipients who work elsewhere. That split is not published.

Awareness at the top

The sharpest finding concerns seniority. Executives are the group most aware of copyright policy, with 92% saying they know their organization's policy. They are also the heaviest AI users: 86% use AI at least daily. And according to the report, they generate 111 potential unlicensed sharing instances per employee each week, six times the rate of an individual contributor, use externally published content with AI 14 times a week, and produce AI outputs that reach 202 people on average, more than double the overall figure.

The report prints neither the individual contributor number nor a figure for middle managers. One sixth of 111 implies roughly 18 or 19, well below the 73 average for the whole sample, which suggests senior staff account for much of the overall exposure.

Why would the best-informed group carry the most risk? The report's answer is time. According to CCC, 76% of executives say they will forward everything when time is short, and executives are "the most likely to override the rules when time is short." Awareness and exposure rise together, the report argues, because both follow from handling a lot of content. "So, a high score can describe exactly the group carrying the most risk," it states.

The broader awareness data has stopped improving. Copyright awareness has risen from 46% in 2016 to 76% in 2026, but that is slightly down from 77% in 2025. Confidence that the organization ensures compliance slipped from 88% to 83%. Among all respondents, 76% know their organization has a copyright policy, 71% can confirm that policy covers externally shared content used with AI, and 68% agree they need permission to use such content with AI, essentially unchanged from 69% in 2025 even as AI use became near universal.

The contrast between stated and actual behavior is stark. According to the report, 81% say they weigh copyright before using externally shared content with AI, and 77% say they think about copyright before sharing AI outputs. At the same time, 67% believe free web content can be shared without permission. "Clearing rights 11 times a week, employee by employee, is inefficient and unsustainable," the report states.

What executives share has already reached the courts in another form. A federal judge in Los Angeles refused on September 2, 2026 to dismiss a contributory infringement claim against Tesla built on a single repost by its chief executive's account, finding the conduct plausibly attributable to the company. That was a public post, not an internal AI workflow, but the logic treats senior people's sharing as corporate conduct.

Governance on paper

Other research shows AI policies outrunning enforcement. A March 2026 survey of 112 marketers by ProGEO.ai found 76.8% working under a formal generative AI policy, but only 43.8% at organizations enforcing it with technical controls. LayerX telemetry found 47.11% of enterprise AI conversations running through personal accounts, outside corporate controls. Neither study measured copyright.

Machine-scale ingestion

The report's fourth finding concerns content that no employee handles directly. According to the report, 51% of AI users say externally published content reaches their AI tools through automated feeds or APIs, in addition to manual input. A further 43% use AI for research assistance, discovering and filtering information, and 39% for process automation and optimization.

"For twenty years, the ceiling on content use was how much a person could physically read in a day," said Grant Hunter, SVP of Outsell Intelligence, in the release. "Today, half of knowledge workers we surveyed already have externally published content entering AI through automated feeds and APIs, bypassing case-by-case review by design. This is now a major enterprise concern."

The release puts the 51% against AI users; Hunter speaks of all knowledge workers surveyed. With 96% using AI, the two bases differ by about two points.

The content feeding these systems spans the categories publishers sell. According to the report, users cite market research (47%), websites (46%), research papers (41%) and news (40%) among the content types they use with AI.

The report is explicit about its limits. It "does not separate agentic use from ordinary automation," so the share attributable to agentic AI specifically cannot be isolated. And although its opening page lists training, fine-tuning, retrieval-augmented generation and prompting as routes into AI, the survey counts only what employees knowingly put into a tool. "Actual use runs higher," the report states, describing its AI figures as a minimum.

Measurements elsewhere point the same way. OpenRouter data circulated in August 2026 showed AI agents consuming nearly five times more tokens per task than human users, and TollBit found 15% of AI page fetchers in Europe reaching URLs publishers had disallowed. Enterprise pipelines pulling news feeds into internal tools are a quieter version of the same shift.

What CCC is selling

The report is not a neutral document. Outsell designed the questionnaire "with input and final approval from CCC," and each finding closes with a "Take Action" panel pointing to CCC products. The Annual Copyright License now includes internal-only AI use of lawfully acquired content from participating rightsholders, according to the report, and the Multinational Copyright License extends coverage to employees in more than 180 countries because "copyright and most licenses are territorial."

"Most organizations are still licensing for a world where content moves between people," said Lauren Tulloch, Vice President and Managing Director at CCC, in the release. "Now with AI usage becoming much more prevalent in the workplace, content moves into AI systems at a pace and a scale no compliance team can track by hand. Getting ahead of it starts with mapping where content enters AI, closing the gaps between the licenses organizations already hold, and establishing appropriate governance."

The commercial frame does not invalidate the data, but it shapes the emphasis. The report's own figures show 51% of those who share AI outputs sending them to external parties and 26% to the public, while the enterprise license it promotes "covers internal use," as the report acknowledges. Its closing line poses the question directly: "You have licensed content for your employees to read and share. Have you licensed it for your AI systems to use?"

In March 2026 CCC expanded its AI licensing portfolio to four options, including pay-per-use AI Transactional Rights that began with summarization, the task the survey ranks second among AI uses. In May, it confirmed that its Annual Copyright License for Higher Education would carry internal AI reuse rights from July 1, 2026. The report's call to "license for machine scale," arguing that "per-seat models were built for human reading," reads as groundwork for pricing that follows consumption rather than headcount.

Why this matters for publishers and marketers

For publishers, the survey is a measure of demand they may not be paid for. News accounted for 40% of reported content types used with AI and research papers for 41%. If the figures are even roughly representative, material publishers sell is being uploaded, summarized and redistributed inside large companies at a rate subscriptions priced for human reading never anticipated.

AI licensing has so far been shaped by disputes with model developers. Anthropic's $1.5 billion settlement with authors in September 2025, roughly $3,000 per work across about 500,000 titles, remains the largest public price point. Penske Media argued in February 2026 that training, retrieval and republishing are separate markets, and direct deals include a News Corp agreement with Meta reportedly worth up to $50 million a year. The CCC data points to a fourth market with far less attention: use by the customers of AI tools rather than by the companies building them.

Collective licensing is one proposed route. The US Copyright Office's May 2025 training report gave it particular attention as a way to cut vast numbers of individual transactions, and a House of Lords committee report in March 2026 anticipated that rightsholders would split between direct deals and collective management organizations. The US legal baseline remains unsettled: OpenAI and Microsoft moved for summary judgment on September 4, 2026 in the case brought over 10.8 million articles.

For marketers and agencies, the relevance is operational. Market research, websites and news, the raw material of briefs and client reports, are the content types the survey found flowing into AI tools. Advertising, marketing and media companies are not among the 14 verticals surveyed, so the report offers no direct measure of agency behavior, although pasting a research report into an assistant and circulating the summary resembles routine agency work.

Methodology and limits

Outsell surveyed 570 full-time knowledge workers at companies with 1,000 or more employees, headquartered in the United States, Austria, Germany, the United Kingdom, Switzerland, the Netherlands, India and Japan. Responses were collected from April 10 to May 8, 2026.

The sample plan targeted roughly 250 responses from the United States, 80 from Germany and 40 from each of the other six countries. US respondents therefore account for around 44% of the total, and the three German-speaking markets together for about 28%. Headline figures are not broken out by country, and the 14 verticals were chosen based on CCC's rights and content inventory.

The stated margin of error is plus or minus 4.0% at 95% confidence overall, and 10% to 15% by vertical. Every figure is self-reported. Base sizes vary where questions were shown only to qualifying respondents, and some percentages come from multiple-response questions. The traditional velocity multiplies three means, and the rounded inputs published for 2026, 8.2 shares, 19 people and 46% external content, produce about 72 rather than 73, a gap consistent with rounding.

Terminology also shifts. The press release describes the 11 weekly AI inputs as "externally published content," while the report's infographic calls the same figure "externally shared content." The methodology uses both terms for the same measures, so the difference appears to be one of wording rather than definition.

Timeline

Summary

Who: CCC (Copyright Clearance Center), a voluntary collective licensing intermediary founded in 1978 and based in Danvers, Massachusetts, and Outsell, Inc., a research and advisory firm. The study surveyed 570 full-time knowledge workers at companies with 1,000 or more employees. Lauren Tulloch, Vice President and Managing Director at CCC, and Grant Hunter, SVP of Outsell Intelligence, commented on the findings.

What: The 2026 Copyrighted Content Usage Trends Report, which found 73 potential unlicensed sharing instances per employee per week through traditional channels, externally published content used with AI 11 times a week, AI outputs reaching 96 people on average, and 51% of AI users reporting content arriving through automated feeds or APIs. Executives registered 111 potential unlicensed sharing instances a week, six times an individual contributor, alongside 92% awareness of their copyright policy.

When: The report was released on September 22, 2026. Survey responses were collected from April 10 to May 8, 2026. The series has run since 2007, with AI questions first added in 2025.

Where: Respondents worked at companies headquartered in the United States, Austria, Germany, the United Kingdom, Switzerland, the Netherlands, India and Japan, across 14 industry verticals.

Why: Content licenses were largely written for people reading and forwarding material, while AI tools copy full texts, transform them and distribute outputs to larger audiences. The findings matter to publishers whose work is being used inside enterprises without clear licensing, and to companies whose governance of AI has not caught up with how employees and automated pipelines actually handle content. The report is published by a company that sells licenses addressing part of that gap, and its data rests on self-reported behavior.