Product news: ChatGPT, Claude Code and Gemini Classroom make source-grounded AI work easier to inspect

This product-news report follows a practical design shift across OpenAI, Anthropic and Google: AI work is moving from disposable prompts toward reusable, inspectable workflows. OpenAI's August 7 ChatGPT update makes documents and saved files easier to carry into a conversation. Anthropic's August 11 Claude Code release strengthens the integrity of work that crosses sessions, machines and skill packages. Google is expanding Gemini in Classroom and adding assignment-grounded rubric drafting. The common educational question is whether context remains visible enough for a learner or teacher to check, revise and defend the result.

OpenAI says formatted text pasted from Google Docs or another ChatGPT conversation now keeps headings, emphasis, links and lists on the web. Users can add files already stored in their Library without uploading them again, select several Library items and search folders, file names and conversation titles more easily. The same release says ChatGPT makes better use of attached documents for summarization, detail retrieval, review and drafting. These are workflow improvements, not guarantees of source fidelity. A reused file can be outdated, a preserved link can be ignored, and a fluent answer can still misread the evidence.

Anthropic's Claude Code 2.1.228 release focuses on less visible but consequential reliability. It fixes cross-session messaging that could start without an inbox, prevents project-memory contents from being removed during session cleanup and avoids deleting a plugin cache whose only version is a symlinked development checkout. It also hardens skills synchronized from Claude's hosted service so they cannot shadow local commands or MCP prompts, and shows cross-session message senders and bodies inline. For collaborative coding or research, those details affect whether provenance, configuration and handoffs survive a longer workflow.

Google supplies the explicitly educational layer. Starting August 10, Gemini in Classroom is expanding to K-12 and higher-education students of all ages when administrators have already granted the required Gemini services. Contextualized starter prompts provide a guided entry point. A separate Classroom update lets educators draft a rubric from an assignment's context, review and edit the proposed criteria, and then attach it to the assignment; that rubric feature is limited to users over 18. Gemini for Education more broadly frames its purposes as helping institutions teach, learn and work inside administrator-managed environments.

None of these vendor release notes demonstrates better learning. They describe availability and intended behavior, not independent accuracy, accessibility, workload or outcome evidence. Persistent context can also preserve a mistaken source, unsafe instruction or unsuitable criterion. Schools should therefore distinguish a product's ability to retain context from a learner's ability to understand it.

For Hong Kong schools and universities, a source-grounded pilot can require a visible source list, versioned inputs, named handoff points and an editable rubric before AI-assisted work is accepted. Students can annotate which file supported each claim, compare an AI draft with the original evidence and complete a short unaided explanation. Teachers can record what they changed in generated criteria and why. The useful product direction is not continuity by itself; it is continuity that remains inspectable, correctable and educationally purposeful.