Chegg's 388-Role Plan and the Free-Answer Trap

In October 2025, Chegg planned to cut 388 roles—roughly 45% of its workforce. The company built its business on making homework answers easier to reach; now chatbots gave students another way to find them.

In the coverage record, Chegg had 3.7 million subscribers after 69% year-over-year growth during the remote-learning surge of 2020. Its Q1 2025 count was 3.2 million after a 31% year-over-year decline, when the company planned to cut 248 employees. In October 2025, Chegg also returned Executive Chairman Dan Rosensweig as CEO, citing the “new realities of AI.” The two subscriber snapshots do not show that demand for tutoring disappeared. They show that demand no longer reliably reaches Chegg in the form its business was built to serve.

General-purpose AI makes homework responses cheap, while Google’s AI search summaries can intercept the visit that once brought a student to Chegg. To justify a subscription, Chegg would have to show where help came from, how it fits the assigned material, and what the learner did rather than the agent. Chegg has yet to demonstrate that shift.

Chegg priced access when finding an answer took work

During the pandemic, Chegg depended on thousands of freelance tutors to work through students’ difficult questions. That arrangement answered a practical question: when a student was stuck outside class, who could supply a useful explanation quickly? Remote-learning demand lifted subscriptions, and the human work behind each response helped make rapid access worth paying for.

By 2021, reporting described more than 70,000 freelance tutors and a problem already built into that design: students could use the service to obtain test answers. The same worked solution could be legitimate help during practice and answer substitution during an assessment. Chegg could distribute the solution, but the transaction alone could not establish which use had occurred.

Generative AI changed the cost of supplying the next plausible response. Chegg suspended its full-year outlook in May 2023, saying ChatGPT’s popularity was pressuring subscriber growth. By November 2024, Wall Street Journal coverage described more than 500,000 subscribers lost since ChatGPT’s launch as students turned to free AI tools. Even when a generated explanation was poor, a student seeking a quick answer could try another without first paying Chegg.

Chegg and other education companies argued in 2023 that AI could improve their products even as cheap AI threatened them. Both claims can be true. An AI tutor can ask for an attempt, identify a mistaken step, and adapt its explanation. A system that merely produces another finished solution competes on the task general assistants already perform.

Google moved the answer onto the road to Chegg

Chegg also depended on students finding its pages. A chatbot can replace a visit when a student starts with the chatbot; a search summary can replace one when the student starts with Google. Chegg sued Google in February 2025, alleging that AI Overviews harmed its traffic and revenue while it explored strategic alternatives, including a possible acquisition.

Alphabet said AI Overviews had reached 1.5 billion monthly users by April 2025. That figure measures the scale of Google’s answer surface, not the number of visits Chegg lost. It does, however, identify the changed handoff. Google can now present a response at the point where a student once chose a result and followed it to an answer site.

On October 2, 2026, a federal judge dismissed Chegg’s antitrust lawsuit over AI Overviews and web-traffic claims. The dismissal does not establish that Google caused Chegg’s subscriber decline, or that search summaries left its traffic untouched. It does weaken a recovery plan that depends on litigation restoring the old route from search query to specialist page.

A cited answer still cannot prove who learned

Anthropic’s Citations API made source-level attribution a product capability: a developer can ground an answer in supplied documents and show the supporting source. In education, that could let a student or instructor inspect whether an explanation follows an assigned passage rather than an unrelated, fluent account. It addresses a real weakness of models that can sound right while being wrong.

A citation answers a narrower question than a teacher must ask. It can identify the material behind a response; it cannot show whether the student tried the problem first, understood the cited passage, or could repeat the reasoning unaided. A licensed repository could settle permission to use a source—the market for authenticated, queryable knowledge shows why access terms matter—without settling what happened after the student received the answer.

A tutoring system could make those steps inspectable: record the student’s attempt, tie a hint to the assigned material, pause before supplying a full solution, and give an instructor a checkpoint to review. The relevant test would not be whether the chatbot can respond fluently to a homework prompt. It would be whether it serves defined learning tasks, including the moments when withholding an answer helps more than delivering one. Chess has the same distinction: an engine can supply an excellent move, but the move alone cannot demonstrate the player’s skill.

Chegg's history makes accountability a design requirement

Chegg would enter that work with an unusually visible integrity history. Reporting on its tutor network described students obtaining test answers through the service. In 2022, the FTC accused Chegg of security lapses after personal information and passwords from about 40 million users had been stolen in 2018; Chegg agreed to strengthen its data security. A service that records attempts, assigned materials, and learning progress would have to protect more than a library of solved problems.

Schools also face conflicting evidence about the scale of AI cheating. A UK survey found nearly 7,000 proven cases in 2023–24, or 5.1 per 1,000 students, up from 1.6 per 1,000 the year before. A Stanford survey of more than 40 US high schools, meanwhile, found no rise in cheating rates in 2023 over prior years. Different populations and measures cannot establish a universal increase. They do establish why a teacher needs visibility into a particular student’s work, rather than an assumption about how students use AI.

That visibility requires durable human review, not merely a person summoned when a model makes an error. Teachers must be able to decide when assistance is appropriate, inspect what the system supplied, and assess the student’s own reasoning.

Schools offer distribution, but not inherited trust

Students have already demonstrated broad appetite for AI study help. In May 2024, five of the top 20 education apps in the US App Store were AI agents helping with school assignments. By 2025, coverage described technology companies investing in AI adoption across US schools and colleges, including partnerships with major teachers’ unions. Financial Times coverage in 2026 described frontier-model providers offering free or discounted tailored learning tools through schools and education companies.

Those partnerships put model providers nearer to the instructor who sets the assignment and the institution that decides what counts as work. Chegg’s earlier strength was getting a useful solution to a student outside that room. The instructor’s harder question begins before the solution appears: a citation at the foot of a finished answer cannot show who held the pencil.