Google Gemini Skills Replace Gems, but the Migration Starts Before Feature Parity
Google is moving Gemini Gems into skills, despite several Gem capabilities remaining unavailable when the migration begins in November 2026.
The Google Gemini skills rollout gives eligible free users reusable instructions that can work across ordinary chats. Users can call a skill with a slash command, let Gemini select one automatically, or combine several skills within one task.
That sounds like a rename, but the underlying product model is changing. Gems behaved like separate custom assistants with their own entry points and conversations. Skills act more like portable workflow components that Gemini can insert wherever they are relevant.
Google also adopted the Markdown-based SKILL.md format used elsewhere in the AI market. That decision places Gemini closer to the reusable workflow systems offered by OpenAI and Anthropic. It also makes portability, rather than assistant branding, a central part of the competition.
The immediate benefit is broader access. Skills can be used without a Google AI subscription, according to Google’s support documentation. The immediate risk is an uneven handoff because skills do not yet support every tool, context source, or organizational surface that Gems support.
The Gemini Gems migration has several different deadlines
Google is replacing Gems through a staggered transition, not a single worldwide shutdown on one date.
Google introduced Gems in 2024 as customized versions of Gemini. A user could give a Gem persistent instructions, attach relevant materials, and return to it for repeated tasks.
A marketing team might create one for campaign briefs. A student might use another to review course materials. An analyst could create a Gem that follows a preferred research structure and response format.
Skills retain the recurring-instruction concept but change how users access it. Google defines a skill as reusable instructions and additional context that teach Gemini how to complete a particular type of task.
Instead of opening a separate Gem before beginning work, a user can enter / in a chat and select a skill. Google says an @ invocation will follow later.
A skill can also remain active for automatic selection. Gemini evaluates the request and applies an enabled skill when it considers that skill relevant.
The company’s migration guide identifies three broad retirement windows. Personal Google accounts begin losing Gems support in November 2026. Business, enterprise, and nonprofit Workspace accounts follow in March 2027. Education accounts follow in June 2027.
Google’s Workspace schedule adds important precision. Skills are scheduled to begin reaching Workspace users on October 5, depending on the release channel. The Gemini app rollout is scheduled to begin October 13 and continue through mid-November.
On November 17, Gems move into the Gemini app’s Settings panel. Workspace users can still create, edit, and use them during the transition.
Google says the final removal for business and enterprise accounts will happen no sooner than March 1, 2027. Education removal will happen no sooner than June 1, 2027.
Those dates matter because some early reports described November 17 as the universal end of Gems. Google’s official schedule shows a longer overlap for managed organizations.
Personal accounts face the quickest handoff. Google says it will automatically recreate remaining Gems as skills when Gem support ends for those accounts.
Users do not need to rebuild everything immediately. They can continue using a Gem until its migration, or manually recreate it as a skill sooner.
The automated process does not make the two products identical. Migrated items will arrive as draft skills, and users should review their instructions and attached resources before depending on them.
Google is also retiring related personal-account experiments on the November schedule. Its help documentation says Opal mini apps and Gems by Google Labs will go away alongside Gems for those users.
The safest interpretation is straightforward. November begins the consumer transition, while managed organizations receive a longer migration runway.
That distinction prevents two common misunderstandings. Gems are not disappearing everywhere overnight, and every existing Gem is not becoming a fully equivalent skill immediately.
Google Gemini skills turn custom assistants into reusable components
The strategic change is composability: Google wants instructions to travel between chats, tasks, and Workspace applications instead of remaining inside isolated assistants.
A Gem bundles a role and its behavior into a named destination. A skill separates the reusable process from the conversation where it runs.
That separation lets users combine instructions. An organization could pair a vendor-evaluation skill with an executive-writing skill, then ask Gemini to assess proposals and draft a recommendation.
An educator could combine newsletter rules with institutional brand guidance. A product team could apply a research-summary skill and a release-note style guide to the same source material.
This is how Gemini skills work at the product level. Each skill contains a name, description, instructions, and optional reference files. Gemini uses that package when a task matches its purpose.
Users can create one manually, ask Gemini to create it, start from a template, or upload a compatible folder. The folder must contain a SKILL.md file at its root.
Google supports plain-text documents, code files, configuration files, PDFs, and images within uploaded skills. The total uploaded package cannot exceed 100 MB.
The company does not support every file type. Microsoft Word and Excel formats are excluded from direct skill uploads, although plain-text formats such as Markdown, CSV, and JSON are supported.
Uploaded scripts also face boundaries. Google’s skills documentation says scripts cannot make requests to external websites.
Those restrictions make current skills closer to structured instruction packages than unrestricted software extensions. They can guide Gemini and supply context, but they cannot independently perform arbitrary network operations.
The SKILL.md structure is strategically important. Google calls it an open, Markdown-native standard and says users can copy compatible skills from other platforms into Gemini.
OpenAI also describes SKILL.md as the playbook for a reusable workflow. Its workflow guidance recommends defining the task, required inputs, steps, output format, and final checks.
Anthropic has likewise published Agent Skills as an open standard. The shared format creates the possibility of moving workflow knowledge between assistants without rewriting it from scratch.
Portability will still have practical limits. A skill that expects one platform’s tools, permissions, or file system might require changes before it works elsewhere.
A writing-style skill should travel relatively well because it relies mostly on instructions and examples. A workflow that depends on product-specific connectors will be harder to transfer.
This changes what users are being asked to preserve. Under the assistant model, the durable object was a named bot. Under the skill model, the durable object becomes a documented method.
That shift favors smaller components. One enormous Gem might contain research rules, writing preferences, formatting requirements, and approval checks.
The same process can become several skills. A user can activate only the pieces needed for a particular task, which reduces irrelevant instructions and makes each component easier to update.
It also creates a different organizational challenge. Teams need clear ownership, version control, testing, and naming conventions once reusable instructions become operational assets.
A poorly written skill can spread inconsistent work across many conversations. A well-governed one can preserve a team’s process without requiring employees to paste the same prompt repeatedly.
This is where skills overlap with knowledge blending. Instructions define how work should be performed, while trusted source material supplies the facts and context used during that work.
Neither element substitutes for the other. A polished workflow can still produce unreliable results when its sources are incomplete, stale, or poorly selected.
The open format therefore matters less as a file extension than as a governance opportunity. Teams can inspect plain-text instructions, compare revisions, and review changes before distribution.
Free access widens adoption and pressures rival workflow platforms
Making Google Gemini skills available without a subscription turns the migration into a distribution move, not merely a product cleanup.
Google says skills are available without a Google AI subscription. Eligible users must be at least 18, use a personal Google Account, and keep Gemini Apps Activity enabled.
Availability is still rolling out. A support page can describe a feature as available while an individual account has not yet received its interface.
That explains some conflicting early experiences. Publications testing free accounts reported that Gems remained visible while the Skills page had not appeared.
The distinction is between policy and delivery. Google has established free eligibility, but regional and account-level rollout timing can still delay access.
Google’s eligibility choice removes a significant source of anxiety around the Gemini Gems migration. Users who created Gems without a paid plan should not lose the underlying instructions simply because the replacement originated inside Gemini Spark.
Spark remains a separate layer. It manages broader tasks and schedules, including work that can continue in the background.
Skills describe how Gemini should perform recurring work. Tasks define what it should accomplish, while schedules define when an action should occur.
A user who cancels an eligible Google AI subscription can lose Spark, tasks, and schedules. Google says that user can still access skills within ordinary Gemini chats.
This separation gives Google a broad free entry point and a paid automation path. Free users can standardize recurring chat work, while subscribers receive more agent-like orchestration.
The competitive pressure falls on companies that treat reusable workflows primarily as managed workplace features. OpenAI’s ChatGPT skills are described as reusable, shareable workflows for eligible Business, Enterprise, Healthcare, and Edu users.
OpenAI allows a skill to contain instructions, examples, code, and supporting resources. It can select one or more skills automatically when they fit a request.
Google offers a similar conceptual model inside consumer Gemini accounts. That gives the company a large testing and adoption funnel before organizations complete their Workspace migrations.
The comparison is not perfectly symmetrical. Product availability, execution environments, connectors, administration, and supported code differ across platforms.
Still, the direction is shared. Major AI providers are moving beyond one-off prompting toward reusable packages that encode how a person or team works.
Google’s advantage is its reach across consumer accounts and Workspace applications. Its scheduled expansion covers the Gemini app and products such as Gmail, Drive, Docs, Sheets, and Slides.
The company says a skill created in the Gemini app will not automatically appear in Workspace. Users must recreate it in Workspace if they want to apply it inside those applications.
That division weakens the promise of universal reuse during the first rollout. Personal and managed surfaces remain separate even when they use the same terminology.
Administrators also need to consider how workflows cross account boundaries. A personal skill might contain instructions or examples that an employee should not move into a managed environment.
Conversely, an organizational skill might encode confidential processes that should remain inside controlled Workspace accounts. Portability increases flexibility, but it also increases the need for policy.
Free access therefore serves two purposes. It reduces migration friction for existing Gem users and teaches a wider audience to think in reusable workflows.
That education can influence procurement later. Employees who standardize tasks with free consumer skills may ask employers for sharing, administration, connectors, and audit controls.
Competitors are responding to the same demand. Custom assistants, projects, memory, and workflow skills all attempt to reduce the cost of explaining a task repeatedly.
Google is betting that modular instructions will become the most reusable layer. Gems placed customization in a destination, while skills place it inside the flow of work.
That bet pressures rivals to make their own workflow formats easier to move, inspect, and share. It also pressures Google to prove that free distribution does not come at the expense of reliability.
The migration begins before skills match every Gem capability
Google’s main risk is not the idea behind skills; it is asking users to move before the replacement reaches full feature parity.
Google acknowledges that skills do not yet work with several familiar Gemini features. The current exclusion list includes Canvas, Deep Research, Guided Learning, video creation, and music creation.
Gems can also use default tools and established context patterns that do not transfer cleanly today. A user who built a specialized Gem around those functions should not assume immediate equivalence.
Google says sharing links and adding Google Drive files or Gemini Notebook materials will arrive over the coming weeks. Until then, users may need to upload supported reference files directly.
That is a meaningful limitation for knowledge-heavy workflows. A manually uploaded copy can become outdated when its original Drive document changes.
Drive integration promises a more natural relationship with living material. Its absence during early migration creates extra maintenance and raises the chance of using stale context.
Skills also lack a dedicated page showing recent chats associated with a particular skill. Those conversations appear in the ordinary Gemini side panel because skills can operate across chats.
That design supports composability, but it removes one organizational cue that Gem users may value. A named Gem offered a recognizable workspace and a clearer history boundary.
Automatic skill selection introduces another tradeoff. It saves users from remembering which workflow to activate, yet it requires Gemini to infer intent accurately.
A mistaken activation can change tone, format, or task steps without the user requesting that behavior explicitly. Google lets users deactivate automatic use for individual skills, which provides an important control.
Teams should test that inference rather than treating it as invisible convenience. The most useful questions are simple: Did the correct skill activate, and did unrelated skills remain inactive?
Stacking creates its own conflict risk. Two skills might issue incompatible formatting rules, define different approval steps, or expect competing source hierarchies.
Google says users can combine skills, but it does not eliminate the need to resolve contradictory instructions. More components can produce more flexibility and more unpredictable interactions.
Security also deserves attention because uploaded skills can contain scripts, reference files, and operational instructions. OpenAI advises users to review skills from outside sources before installing them.
Google limits external web actions from uploaded scripts, which narrows the immediate attack surface. However, users still need to inspect instructions that might redirect behavior or expose sensitive material.
The open format makes inspection possible. It does not make every package trustworthy.
Automatic migration is another point requiring verification. Google says it will recreate remaining Gems as draft skills, but a converted workflow might behave differently under Gemini’s new selection model.
A Gem designed as a self-contained persona may not map neatly onto a modular process. Its instructions might be too broad, rely on an unsupported tool, or conflict with other enabled skills.
Users should preserve copies of important Gem instructions before migration. They should also download any attached knowledge files and document the output characteristics that matter.
After conversion, they can run representative tasks through both versions while overlap remains available. Useful checks include factual coverage, formatting, tool use, citations, tone, and refusal behavior.
A critical early account found that free eligibility did not guarantee immediate interface access. That report does not contradict Google’s policy, but it highlights rollout uncertainty.
The same caution applies to feature announcements. “Coming in the next few weeks” establishes direction, not a guaranteed date for every account or region.
Google also has to reconcile two product experiences. The consumer Gemini app emphasizes easy creation and automatic use, while Workspace customers require administrative control and predictable deployment.
The later enterprise and education deadlines reflect that complexity. Those customers have integrations, shared processes, and classroom experiences that cannot be switched as casually as a personal assistant.
Education receives the longest runway. Gems remain connected to Google Classroom and supported learning-management systems, making a rushed removal especially disruptive.
The Gemini Gems migration will succeed only if users experience continuity at the task level. Preserving a title and instruction block is not enough when tools, context, and conversation organization change.
This is the core reversal behind the announcement. Skills promise a more flexible system, but the first stage can deliver less capability for some established workflows.
What happens next will determine whether skills become a standard
Three signals will show whether Google has built a durable workflow layer or simply renamed customization before the replacement was ready.
The first signal is successful migration quality for personal accounts. November provides the earliest large test because Google plans to recreate remaining consumer Gems automatically.
Users should watch whether instructions, reference materials, and expected behavior survive conversion. Reports of missing context or substantially changed outputs would weaken Google’s continuity claim.
The most revealing cases will involve complex Gems rather than simple style prompts. A short proofreading Gem is easy to convert. A research workflow using files and specialized tools is a harder test.
Google should also make the draft status clear. A converted skill needs review before users rely on it for consequential work.
The second signal is the arrival of promised parity features. Sharing links, Drive files, and Gemini Notebook support will determine whether skills can replace knowledge-rich Gems.
A broad rollout of those features before personal Gem removal would strengthen the migration case. Significant delays would force users to maintain manual file copies or redesign their workflows.
Canvas and Deep Research deserve separate attention. Google currently says skills do not work with them, and it has not promised that every unsupported tool will arrive on the same schedule.
If skills gain access to those experiences, they become a stronger universal layer. If exclusions persist, users may need different customization systems for different parts of Gemini.
The third signal is cross-surface portability. Google says Workspace skills and Gemini app skills do not automatically synchronize, even though both use the same concept.
That fragmentation matters more than file compatibility. A genuinely reusable workflow should move from chat to Gmail, Docs, and Sheets without extensive rebuilding.
The Workspace rollout will reveal how much administration Google provides. Organizations need sharing controls, ownership rules, lifecycle management, and visibility into what instructions employees use.
Google’s Workspace schedule gives business customers several months to evaluate the replacement. Education customers receive even more time before June 2027.
Those periods should be treated as testing windows, not reasons to postpone preparation. Organizations can inventory current Gems, identify critical dependencies, and assign owners before automated conversion.
Individual users can begin with a smaller audit. List the Gems used repeatedly, copy their instructions, preserve their source files, and record a few representative outputs.
Next, recreate one important workflow as a skill. Test explicit activation with /, then test automatic selection and combinations with another skill.
That process helps reveal whether the modular model improves daily work. It also exposes conflicts before Google removes the familiar Gem interface.
Users should avoid converting every Gem into one oversized skill. Smaller components usually make errors easier to isolate and instructions easier to maintain.
They should also separate stable methods from changing facts. A skill can define the research process, while current documents provide the evidence for each task.
For teams, the migration is an opportunity to review undocumented prompt practices. Reusable instructions become more valuable when someone owns their accuracy and approves significant revisions.
The same review should cover sensitive information. A skill package is portable, so users must understand which files and instructions can leave a managed environment.
Competitor activity will provide another useful benchmark. OpenAI and Anthropic are also building around reusable skills, which makes interoperability a practical test rather than an abstract promise.
A workflow that transfers with only minor adjustments would validate the open standard. A package that requires a complete rewrite for every product would expose platform dependence beneath the common filename.
Google does not need perfect portability to make the strategy useful. It does need predictable behavior, clear permissions, and enough feature coverage to justify replacing Gems.
The company has already made the largest distribution decision by extending Google Gemini skills to eligible free users. Now it must prove that access, migration quality, and capability can advance together.
For anyone with important Gems, the best next step is direct testing before the retirement window closes. Compare migrated skills against real work, document failures, and keep recoverable copies of instructions and source files.
The transition is not just a new menu item. It changes customization from a collection of separate assistants into reusable workflow infrastructure. Whether that improves Gemini will depend on the details users can verify over the next several months.



