The Butler
Back to briefings
Google / Everyday AI

Your Gemini Gems are changing tomorrow. Save the useful parts, not just the name

· The Butler · 4 minute read

The metallic robot Butler works at a desk, wearing formal attire.
Butler editorial illustration, not a product screenshot.

A well-used AI helper becomes a small piece of personal infrastructure. It knows the shape of your weekly report, the tone of your customer replies or the checklist you follow before publishing. When that helper changes format, the important question is not whether its name survived. It is whether your next real task still works.

Google's current transition guide gives adult personal-account users two dates: November 17, 2026, when Gems become read-only as they transition to skills, and December 17, when those Gems are removed. Work and school accounts follow later schedules, so this is not a universal tomorrow deadline. Check Google's account-specific timeline.

A different way to reuse instructions

Google introduced skills as reusable instructions that can be combined within a conversation rather than requiring a separate helper for each task. Its announcement describes attaching reference material and bringing multiple skills to the same job. That makes a useful distinction: a writing voice and a report format can become separate ingredients instead of two copies of the same sprawling prompt. Google's skills announcement.

For a small business, imagine an estimate-writing skill and a plain-English editing skill. You might want the first without the second when preparing an internal worksheet, but both when writing to a customer. That is our example, not a claim that a particular combination has been tested here.

Make one small recovery folder

Before reorganizing anything, save the instructions you genuinely use, the original reference files and one representative input with a satisfactory old output. Name the folder after the job, not the AI product. A folder called Weekly supplier summary will still make sense if you change services next year.

Google says supported files migrate automatically, but skills have a 100 MB total file limit; oversized collections can lose their largest files during transfer. Some formats are excluded. Its guide also lists tools that do not yet work with skills, including Canvas, Deep research and video creation. Automatic migration should therefore not be mistaken for feature parity. Migration details and limitations.

Our suggestion is to keep this first pass deliberately boring. Do not rewrite every instruction on moving day. If you change the wording, the reference files and the interface together, it becomes harder to tell why the result changed.

Test the job you actually care about

Take your saved example and run it through the migrated skill. Compare the parts that matter: Did it use the correct reference? Did it keep the required sections? Did it leave out a detail you had explicitly asked it to retain? A more polished paragraph does not compensate for a missing price assumption or an omitted action item.

Then try a second example with an awkward detail, such as an incomplete note or contradictory source material. You are looking for whether the helper asks a sensible question instead of confidently filling the gap. This is a practical comparison, not a formal model benchmark.

If the new setup passes, start separating instructions only where that would save repetition. If it fails, your recovery folder gives you something concrete to repair rather than a vague memory that the old helper was better. Readers starting from scratch can use our AI setup guides before building an elaborate collection.

The useful outcome is not a tidy list of new skills. It is being able to finish Tuesday's work without discovering that Monday's reliable shortcut quietly lost an attachment.