Dymocks Tutoring Closes Sydney Centres, Tells Parents to Use AI

A tutoring company that blamed AI for its own closure

Tutoring businesses usually tell parents that their children need more help. Dymocks Tutoring and Talent 100 did the opposite. Before closing its five Sydney centres in late September, the company told parents to put their money into general-purpose chatbots such as ChatGPT and Gemini, because students were already using AI for exam preparation and quick feedback on their study2. The tutoring arm belongs to the 140-year-old Dymocks brand. Its message was that its own product was no longer good value, with fees reaching about $900 per subject1.

Dymocks Education chief executive Mark Buckland was blunt in comments to the Australian Financial Review. He said most customers would do better spending on cheaper options than on "overpriced tutoring." He also admitted that AI would not match the company's human tutors, but noted it costs about $30 a month1. He went further and argued that the tutoring industry does not do anything unique. In his view, what it mostly sells is reassurance, and AI can supply that too1.

The outlets broadly agree on the facts but differ on some details. Most describe the $900 figure as a per-subject fee. One report says it is charged per term4, and a Japanese report puts the range at A$800 to A$900 per subject per semester3. The company's closure FAQ, as reported, set the last business day as 27 September. It said the problem students face has changed from getting help with learning to using AI well3. Students were not simply dropped. Parramatta and online classes are moving to Uplift Education, students at the other four centres are being offered discounts with Art of Smart, and about 38 students facing final exams will keep one-on-one online sessions with their tutors until the exams are over3.

The industry pushes back

Other tutoring operators did not accept that the sector is finished. Mohan Dhall, chief executive of the Australian Tutoring Association, said the exit surprised him. He suggested the company could have gained more by staying in the market1. David Ramagge, chief executive of Matrix Education, said his business teaches about 8,000 students around Sydney each term and that the number is rising. Matrix offers AI study tools of its own, but Ramagge still expects high-achieving students to want in-person teaching3.

This disagreement is the most useful part of the story. Dymocks' argument is about price, not quality: AI is worse, but it is cheap enough to win anyway. Commenters picked up on this. Some said Australia's cost-of-living squeeze, not the technology, is what is pushing families away from paid tutors. Others pointed out a basic problem: a student often lacks the expertise needed to notice when a chatbot gets something wrong6. Parent reactions quoted in coverage also included strong objections that human connection cannot be replaced1.

One analysis expects specialised tutoring, such as scholarship exam preparation and support for students with learning difficulties, to stay with humans. It sees general homework help and routine revision as the part of the market most exposed to automation4. That reading looks right. Dymocks may simply have been operating in the segment AI threatens most, not proving that the whole industry is doomed.

What the research actually says about AI tutors

Buckland's position relies on an empirical claim: a cheap AI tutor is good enough. The research from 2025 and 2026 does not settle that question either way. Results depend heavily on how the tool is designed.

The optimistic evidence is real. A randomized trial at Harvard found that physics students learned more than twice as much, in less time, from a custom AI tutor than from an in-class active-learning lesson. They also said they felt more engaged26. That tutor, however, was built around the same teaching practices as the course. It was not an off-the-shelf chatbot26. Stanford's National Student Support Accelerator has summarised two trials of AI built into live, chat-based maths tutoring. In one, students working with Google's LearnLM fixed their mistakes as well as students with human tutors did, and did better on later, harder topics: 66 per cent succeeded, compared with 61 per cent for human-only tutoring21. In the other, the Tutor CoPilot tool raised mastery rates by four percentage points. The largest gains came for students whose human tutors were less experienced21. Coverage of the Dymocks closure also mentioned a GRE-preparation experiment in which one hour with an AI tutor improved scores by about as much as an hour with a human expert3.

A large new study points the other way. A randomized trial at a big U.S. public university, released as a working paper in October, covered 2,379 undergraduates and 30 instructors. It found that giving students a course-integrated AI tutor lowered final grades by up to 0.37 standard deviations. It also sharply reduced participation in instructor-designed online activities24. First-generation students lost more than twice as much as their peers24. The authors place this alongside earlier evidence of a "learning penalty": general-purpose AI can improve performance on assisted tasks while weakening independent performance later, though tutoring guardrails reduced that harm in high-school maths24. They conclude that AI's effects depend on how the tutor is configured, how instruction is structured and how much students actually engage24.

This contrast is central to judging Dymocks' advice. The good results mostly come from purpose-built systems, AI working alongside human tutors, or tightly structured settings2126. Dymocks pointed parents to consumer subscriptions to ChatGPT and Gemini2. That is the kind of open-ended, unsupervised use the more cautious research warns may replace real thinking rather than support it24. In practice, the company was citing the best case for AI tutoring while recommending the setup with the weakest evidence.

Australia's school AI policy is going the other way

Australian schools have moved toward controlled, enclosed AI tools rather than open consumer chatbots. States banned ChatGPT soon after it launched, mainly out of fear of undetectable plagiarism16. The national Framework for Generative AI in Schools, released in December 2023, replaced those bans with a set of principles15. One principle states that AI tools should support teaching and students' own skill development, not replace them11. Another warns against harm from over-reliance11. A June 2025 review found the framework fit for purpose. It is policy rather than law, and it carries no penalties11.

The states have filled in the detail. NSW public schools use NSWEduChat, a department-owned tool hosted in Sydney and open to students in Years 5 to 12. Other free AI tools are not endorsed, and students' access to outside AI on department networks is restricted unless the tool passes an ethics assessment12. Queensland uses its own tool, Corella. Victoria requires opt-in parental consent before any AI tool collects personal information beyond a school email login12. South Australia led this approach with EdChat, built with Microsoft. Its education chief has said teachers saw students questioning and cross-checking AI answers rather than simply accepting them1519.

Here the Dymocks advice runs into policy. The approach most Sydney families now see at school is supervised AI, kept in Australia, with human oversight. The company recommended general commercial tools with no supervision. One analysis notes that NSW's education standards authority has not issued comprehensive guidance on using commercial AI for private tutoring outside school4. Without such guidance, parents are left to judge on their own whether these tools teach well4. Concerns about children's data are concrete, too. Character.AI, the AI companion app most popular with Australian children, has filed for bankruptcy in the U.S., and users' private conversations are now among the assets for sale9.

The academic integrity question

Cheating with AI remains a live issue. NESA's 2024 policy treats unapproved AI use in assignments as a breach of academic integrity, and education departments around the country are stepping up enforcement19. Tasmania sets no take-home exams in its external Year 11 and 12 assessment program19. In higher education, the regulator TEQSA moved in 2026 from guiding universities to regulating them. It now reviews each provider's plan for protecting the integrity of its awards as part of normal oversight11.

This matters for parents weighing Dymocks' advice. A human tutor is accountable to someone. A chatbot will write an essay on request, and the line between getting feedback and handing in the AI's work depends entirely on the student's restraint. High-stakes exams like the HSC are sat without AI. The research on learning penalties suggests that students who lean on chatbots throughout the year may be poorly prepared for that moment24.

The verdict

The best way to read the Dymocks closure is as a business decision presented as an education argument. The price gap is real, and Buckland was frank that AI is worse but cheaper1. It is a sign that mid-market, homework-help tutoring is under pressure. It is not proof that AI has caught up with human tutors. The closure also came as PISA results showed Australian teenagers at their lowest literacy and maths levels in two decades1. Swapping $900 human tutoring for unsupervised $30 chatbots is not an obvious fix for that. The research most likely to help children involves structured AI with a human still involved, and that is the model schools are adopting and Dymocks has just left.