Google Slides can build the deck. Don’t skip the source pass.

From a blank presentation, eligible users can choose “Generate presentation,” describe the deck they need and add reference files. Google says Gemini can use Google Docs, Sheets, PDFs and prior presentations. It may also suggest relevant files, emails and chats from Drive context for the user to add.

The process includes a planning step before slide generation. Gemini asks about length, tone, focus and audience, then produces an outline for approval. A previous deck can serve as a style reference so the result follows an existing look rather than inventing a new one.

The output is fully editable. Text, charts, colors and layouts can be revised through ordinary Slides controls or follow-up instructions to Gemini. The feature is launching in English for Google Workspace Business and Enterprise Standard and Plus customers, as well as Google AI Pro and Ultra subscribers.

This is not a tiny design shortcut. It compresses research assembly, outlining, formatting and revision into one familiar screen. For internal briefings, sales updates, class presentations and small-business proposals, that can move a person past the blank page quickly.

Rough work advertises that it is unfinished. A beautiful slide does the opposite. The chart is aligned, the headline sounds confident and the colors match last quarter’s deck. That visual finish can make an unchecked claim feel settled.

Google’s outline step helps with direction, and attaching source files is better than asking a model to invent a deck from a topic alone. Neither step proves that every number, date or conclusion made it onto the slide correctly. A summary can flatten a caveat. A chart can use the right numbers with the wrong denominator. A neat headline can turn a tentative finding into a promise.

There is also a selection problem. If Gemini suggests an email, chat or older deck as context, the material may be relevant without being current or authoritative. A three-year-old sales slide can match the brand perfectly and still carry a target the company abandoned.

So treat the generated deck as a designed first draft, not a finished deliverable. That is not a warning against the feature. It is the condition that makes the time savings believable.

First, run a source pass. Mark every number, date, quote, customer claim and comparison. Open the primary file behind each one. If a claim has no source, remove it, soften it or find evidence before the meeting. For a ten-slide internal deck, record how many factual claims you checked and how many needed a correction. That gives the team a local accuracy baseline instead of a vague feeling that the output was “pretty good.”

Second, run a story pass. Read only the slide headlines in order. They should form an argument, not a list of related facts. Ask what the audience should decide, remember or do differently. Generated presentations tend to reward coverage; real presentations usually need subtraction.

Third, run a room pass. Check names, permissions and private context. Remove customer details that do not belong in the audience. Make sure speaker notes do not expose material hidden from the visible slide. Open the file from the account and device that will present it, then test links, charts, fonts and video without relying on the generation screen.

Time all three passes. If Gemini creates the deck in eight minutes and review takes seventy, the result may still be worthwhile. Just call it a sixty-two-minute gain only if the old process took longer than seventy-eight minutes. Generation time is not time saved.

Sable Quinn sees a taste problem before a formatting problem. A generated deck can be competent and still sound like it could belong to any company. She would keep one slide that contains the uncomfortable decision, firsthand example or sharp comparison the model could not infer from old files. If every slide survives a company-name swap, the deck is polished but anonymous.

Mara Vale is less worried about obvious nonsense than plausible drift. When a slide borrows from an email, spreadsheet or old presentation, the source should remain easy to open from the review copy. “Based on Drive” is not enough. The presenter needs the exact file, tab, date or passage before someone in the room asks where the number came from.

Those are different checks. Sable protects the point of view. Mara protects the trail back to evidence. A useful AI presentation needs both: something worth saying and a quick way to prove the parts that can be checked.

Pick one recurring deck you already make: a weekly project update, monthly sales review, client recap or class lesson. Use the last manual version as the baseline. Record total active minutes, not the time between opening and sending the file.

For the AI-assisted version, count generation and prompt time, source-check minutes, claims corrected, slides deleted, slides substantially rewritten, broken elements found and final approval time. After the presentation, note any question you could not answer from the cited material and any follow-up correction you had to send.

Then compare the whole task. A deck that took less time, needed fewer late corrections and was easier to defend is a win. A deck that appeared in minutes but required a nervous line-by-line audit has only moved the work.

Google Slides has made the first draft dramatically cheaper. Good. Spend some of that gain on the two parts an audience notices when they go missing: a clear judgment and facts the presenter can stand behind. Keep the rest of the time.