Customer-service automation and human oversight: a practical guide for online retailers
How to use chatbots, AI-drafted replies and triage without losing control: escalation design, quality measures, AI Act disclosure, GDPR Article 22 and liability.
Third-quarter filings from Chegg, Duolingo and Nerdy show AI squeezing homework-help revenue and adding running costs, while usage at the largest of the three keeps growing.
Third-quarter filings from Chegg, Duolingo and Nerdy show artificial intelligence (AI) squeezing older homework-help revenue while adding a new running cost for the apps that build it in. Read with survey data on student use, they offer a view of the study-app business based on reported figures rather than market forecasts.
The sharpest change is at Chegg. In its third-quarter earnings release, filed with the US Securities and Exchange Commission on 3 November 2026, the company reported net revenues of $46.5 million, down 40% from $77.7 million a year earlier.
Executive chairman and chief executive Dan Rosensweig said “AI hit our business hard, and we had to reinvent the company”. He said Chegg was evolving “from a homework help company into an employability company”, while chief financial officer David Longo said the “legacy homework-help business continued to decline as expected”. Chegg ended the quarter with $43.6 million in cash and investments after repaying its remaining convertible debt.
Duolingo, a language-learning company rather than a dedicated exam-preparation business, shows a different pattern. Its third-quarter shareholder letter, dated 3 November 2026, reported revenues of $309.0 million, up 14% year on year. Paid subscribers rose 14% to 13.1 million and daily active users rose 29% to 64.9 million.
AI appears in Duolingo’s margins as well as its product. Gross margin was 72.9%, ahead of the company’s expectations, which it attributed mainly to “lower feature-based AI costs” and hosting trends. It expects 72.0% in the fourth quarter “as we expand the rollout of AI-based features on our platform”.
Online tutoring company Nerdy reported a smaller and shrinking business. Its third-quarter release, filed on 9 November 2026, showed revenue of $33.2 million, down 10% from $37.0 million, and 31.8 thousand active members, down 7%. Its net loss narrowed to $12.8 million from $18.7 million, and it said its annualised AI cost run-rate fell about 60% from a June and July peak to September.
Survey data points to the demand behind these numbers, although public figures lag the market. Pew Research Center found that 26% of US teenagers aged 13 to 17 said they used ChatGPT for schoolwork, up from 13% in 2023. The survey covered 1,391 teens and ran from 18 September to 10 October 2024.
Because ChatGPT is a general-purpose chatbot, that figure also illustrates the competitive problem for paid study apps: students can get help without buying a specialist product. One response is to build around a student’s own course material.
Serona, an exam prep app, says it listens to classes and builds study tools around what professors say about upcoming exams, and lists Pro and Max subscription plans on its website.
For operators and investors, the filings suggest AI is neither a simple threat nor a simple advantage. Chegg’s results show how far a paid homework-help service can fall when AI changes the market, while Duolingo and Nerdy both discuss AI as a cost that has to be managed.
For smaller app businesses, the practical questions follow from those disclosures: what a subscriber pays, what each AI feature costs to serve, and whether users stay once a course or exam is over. Nerdy’s reported fall in AI run-rate and Duolingo’s margin guidance both suggest that serving costs can move quickly from one quarter to the next.
Fourth-quarter results will test the guidance already given. Chegg expects revenue of $38 million to $40 million, Nerdy $37 million to $39 million and Duolingo $322 million, with Duolingo’s gross margin forecast to slip as more AI features roll out.
These figures measure revenue, subscribers and margins rather than learning, and each company reports them on its own definitions. Duolingo and Nerdy also sit in adjacent categories, language learning and tutoring, so they are indicators for the wider sector rather than direct comparisons with exam-preparation products.
AI study tools can make revision material quicker to produce and easier to reach, but convenience is not evidence of better outcomes, so students and parents should treat any vendor’s claims with care.
How to use chatbots, AI-drafted replies and triage without losing control: escalation design, quality measures, AI Act disclosure, GDPR Article 22 and liability.
Where AI product copy saves time, where it fails (invented specs, green and health claims), what Google says about AI content, and a review workflow that works.
Define conversion properly, find where shoppers drop out, use site search and Core Web Vitals data, and run A/B tests that are big enough to mean something.