Phase 4 · Gemini · Level 3 · Power User
Gemini in Chrome and scheduled actions
By the end, you'll be able to…
- Recognise what Gemini in Chrome does and judge when it helps rather than distracts
- Set up a scheduled action for a recurring task and know its privacy conditions
- Spot the over-trust traps in a browser assistant and an unattended daily digest
Why it matters
Two Gemini features move it off the chat page and into your day: an assistant living inside the Chrome browser that can see the page you're on, and scheduled actions that run a prompt for you on a timetable. Both are useful and both carry a specific catch: a browser assistant can misread the page it's summarising, and a scheduled digest you never read critically becomes a confident daily briefing you're trusting blind. This lesson is awareness-level: what they are, when they earn their place, and the traps.
Gemini in Chrome: an assistant that can see the page
Most of your Gemini use so far has happened on a chat page you visit. Gemini in Chrome brings the assistant into the browser itself: a panel you open from within Chrome that can take the page you're currently looking at as context. Instead of copying a long article into a chat, you ask about the page in front of you ("summarise this", "what does this say about cancellation?", "compare this supplier's pricing with the last tab") and it answers using what's on screen.
The value is the removed friction. When Tom is comparing office-move sites for Fernway and lands on a dense commercial-lease page, he can ask the in-browser assistant to pull out the notice period and the break clause without leaving the tab or hunting through legalese. It's the grounding idea again, applied to whatever web page you happen to be on. The assistant answers from the page rather than from general knowledge.
Summarise this page for me: the notice period, any break clause, and the total monthly cost including service charges. Quote the exact wording for each, and if any of these isn't stated on the page, say so rather than assuming.
Why this works: A specific extraction request (named items, plain-English explanation, and an explicit 'say so if it isn't stated') turns a page-aware assistant from a vague summariser into a targeted reader, and the honesty clause guards against it inventing a clause the page doesn't actually contain.
Two honest caveats. Availability: Gemini in Chrome rolled out gradually (first on desktop in some regions, then wider) and depends on your Chrome version, region and account, so it may not be on your machine yet; that's an availability matter, not a fault. Privacy: to answer about a page, the assistant works from that page's content, so treat it exactly like pasting the page into a chat. Don't point it at a page showing confidential customer data, someone's private account, or anything covered by the Phase 0 privacy rules. A page-aware assistant is only as safe as the page you aim it at.
Scheduled actions: a prompt that runs on a timetable
The second feature moves Gemini through time. A scheduled action is a prompt you ask Gemini to run automatically on a schedule (once, or daily, weekly or monthly) and deliver the result when it's done. Instead of typing "summarise my unread email and today's calendar" every morning, you set it once and Gemini produces the digest on its own. You manage these under Gemini's settings, and Google's Schedule actions in Gemini Apps help page has the current detail.
Sensible uses are recurring, low-stakes briefings: a Monday summary of the week's meetings, a daily digest of a topic Tom tracks for Fernway's market, a weekly nudge to review something. There are real conditions, and they're worth knowing before you rely on it. Scheduled actions that touch your Google Workspace need the relevant app connected; the feature requires your activity setting to be on (it won't run with that switched off); and there's a cap on how many you can have active at once, currently around ten. Those specifics move, which is why the durable point is: it's a convenience feature with account and privacy conditions, not a free-standing robot.
Every Monday at 8am, give me a short digest for the week ahead: my meetings for the next five working days as a bulleted list with times, and any all-day events flagged separately. Keep it under 150 words and only cover this week.
Why this works: You set a scheduled action by describing it conversationally, so naming the cadence, the exact scope and the output shape up front produces a consistent, scannable digest each time instead of a rambling one, and bounding it to 'this week only' stops it drifting.
Each weekday morning, summarise notable public news from the last 24 hours about the UK small-business software market, three bullets maximum, each with a source link, and flag any item you couldn't confirm from a reliable source.
Why this works: A tightly scoped, recurring research prompt is a fair use of a scheduled action because the job is repetitive. Stating the topic, the timeframe and 'flag anything you couldn't confirm' keeps the digest focused and honest about its own gaps.
The over-trust trap, doubled
Both features quietly increase how much you trust AI without watching it, and that's the risk to name. A browser assistant summarising a page can misread it (flip a notice period, miss a clause, or confidently describe a term that isn't there) and because you asked because you didn't want to read the page, you're the least likely person to catch it. For anything load-bearing (a contract term, a figure you'll act on), the summary is a pointer to the exact wording, not a substitute for reading it.
A scheduled action is the same trap stretched over time. A daily digest that runs unattended for weeks becomes something you glance at and believe, and if it misreads an email or a hallucination creeps into a research digest, it arrives with the same calm authority as the accurate ones. The habit that keeps both safe: treat the output as a briefing that points you at the source, and check anything you'd act on against the real page, the real email, the real calendar entry. Convenience is the benefit; it must not become blind trust.
Try it now
Common mistakes
- Aiming a page-aware assistant at a private page. Summarising a page means the assistant works from its content. Point it at confidential customer data or a private account and you've shared it. Treat it exactly like pasting the page into a chat.
- Believing the page summary over the page. You asked because you didn't want to read the fine print, so you're the least likely to notice a misread notice period or an invented clause. For load-bearing terms, use the summary to find the wording, then read it.
- Trusting a scheduled digest because it's routine. A digest that's been right for weeks earns a glance, not blind faith. A misread email or a slipped-in fabrication arrives with the same calm confidence as the accurate days. Check anything you'll act on.
- Assuming availability equals a fault when it's missing. Both features depend on your account, region and app version. A missing panel or setting usually means "not enabled for you yet", not "broken".
- Scheduling something that acts, not just briefs. These are convenience features for recurring briefings. The moment a task sends, spends, or decides, keep a human in the loop. Automate the summary, never the judgement.
Keeping current
Both features are moving quickly. Gemini in Chrome's availability by region and Chrome version, and the rules and limits on scheduled actions, all change. Trust your screen over this lesson and check Google's official pages: Gemini in Chrome, Schedule actions in Gemini Apps and the Gemini Apps release notes. Accurate as of 13 July 2026.