Six assistants, seven real days, four tasks with checkable answers. A free one won — for reasons that have nothing to do with model quality.
Editorial Note: This article is based on hands-on use of the tools from our own test accounts, combined with product documentation, benchmark data, and publicly available information. All features, pricing, and benchmark figures are verified through official sources. See our Disclaimer.
Google Gemini is the best AI assistant for daily tasks for most people, and the price is only half the reason. It sits inside the account you already use for mail, calendar, maps and your phone, and the free tier is capable enough to run a real day on. In our seven-day test its inbox triage caught 11 of 11 buried action items, and it was the only assistant that answered a same-day local question correctly without being handed a link. ChatGPT is the better assistant if you are willing to pay. It won the two tasks that need real reasoning: turning seven competing constraints into a workable day, and pulling return deadlines out of a stack of receipts and warranty PDFs. Claude is the best of the six for writing-heavy admin, and Microsoft Copilot is the only one already installed on your machine.
This is not a ranking of model intelligence. Daily tasks are the boring load: triage the inbox, plan the day, extract a deadline, check whether a place is open, draft the reply you have been avoiding. On most of that work the model is not the deciding factor. What decides it is whether the assistant can see your context, and whether opening it costs anything. That is why a free assistant wired into your Google account keeps beating a smarter one in a separate tab.
Seven consecutive days, September 14 to 20, 2026, on the same laptop and phone, against the same real workload: 84 incoming emails and messages, four appointments, one warranty claim, one travel booking and one Saturday plumber search. Every assistant ran in its default configuration on its cheapest usable tier, with no prompt engineering. We scored four fixed tasks, each with a checkable right answer:
| # | Assistant | Cheapest usable tier | Best for | Score |
|---|---|---|---|---|
| 1 | Google Gemini (3.8 Flash / 3 Pro) | Free · Google One AI Premium $19.99/mo | Inbox, calendar and local lookups inside the Google account you already have | 8.5 / 10 |
| 2 | ChatGPT (GPT-5.6, GPT-6 Astra) | Free · Plus $20/mo | Multi-constraint planning, document extraction, scheduled tasks | 8.5 / 10 |
| 3 | Claude (Sonnet 5, Opus 5) | Free · Pro $20/mo | Writing-heavy admin, long documents, anything where the tone matters | 7.5 / 10 |
| 4 | Microsoft Copilot | Free with Windows · M365 Copilot $30/user/mo | Zero-setup help inside Windows, Word, Excel and Outlook | 7 / 10 |
| 5 | Perplexity | Free · Pro $20/mo | Quick sourced answers and purchase research | 6.5 / 10 |
| 6 | Notion AI | From $10/mo with a Notion plan | People whose notes, projects and databases already live in Notion | 6.5 / 10 |
Gemini takes first place because of what it can already see. With Workspace extensions switched on it reads Gmail, works against Google Calendar, writes into Docs and answers location questions from Maps, and none of that costs anything. In our test it returned 11 of 11 action items with zero invented to-dos — the only assistant that did not manufacture work — and it settled the pharmacy question in 22 seconds. Its weak spot was planning: it produced a tidy schedule with the 15:15 pickup silently missing. Two caveats worth knowing. The deeper Workspace features sit behind Google One AI Premium at $19.99 per month, and extension availability still varies by country, so check what your locale actually gets.
ChatGPT was the strongest reasoner of the six and the only one that treated our constraint problem as a constraint problem. It fitted all seven commitments, named the trade-off required to make them fit, and got all three return dates right without inventing one. What it lacks is your context out of the box: it cannot read your inbox or calendar unless you connect something or paste it in, which is the whole reason it finished second. The feature that changes its category is Tasks — scheduled prompts that run without you asking — and at $20 per month it is the upgrade most worth paying for.
Claude placed third because our four tasks did not reward its strengths. Give it the same inbox dump and it produced the best-written brief of the six; give it an awkward email and its draft needed the least editing. What it does not do is connect to your mail or calendar, or look up what time a shop closes. Projects is the closest thing here to a persistent workspace for your own documents, and for anyone whose day is mostly correspondence and long PDFs, that is worth more than a Maps lookup.
Microsoft Copilot is already on the machine, and for a lot of people that settles it: it handled receipt extraction acceptably, and it can act on the document you have open rather than on a paste. Perplexity is not a day planner and should not be scored as one — it is an answer engine, fast and cited, and on the pharmacy question it said plainly that it could not confirm same-day dispensing rather than guessing. Notion AI finished last here because three of the four tasks happen outside Notion; if your notes and boards are already there, it quietly wins on the one thing nothing else can match, which is searching your own workspace.
Both handled the obvious items; the difference was in the subtext. ChatGPT turned “can you send me the deck before Thursday” into a task with a deadline of its own invention, where Gemini parked it as blocked on a reply that did not exist yet, and Gemini merged two duplicate deadline threads that ChatGPT listed twice. That refusal to invent work is the most useful property an inbox assistant can have. On the pharmacy question Gemini answered in 22 seconds with two locations inside five miles, their closing times and the caveat that same-day repeat dispensing depends on the prescription already being on file; ChatGPT needed two browse passes and 1 minute 48 seconds and still could not confirm opening hours without being handed a page.
Neither model was allowed to ask a follow-up question. Gemini produced a tidy schedule and quietly dropped the 15:15 pickup; asked to verify, it rebuilt the afternoon without the deep-work block. ChatGPT found the real conflict — the deep-work block and the standing call cannot both sit after the dental appointment if the prescription is collected before 17:00 — and said which constraint had to give. On documents it returned three correct last-valid-return dates and the right still-covered verdict for both items, while Gemini invented a 90-day window that appears nowhere in the three files and repeated it when challenged. A confident number with no source behind it is the failure mode to watch for in any assistant you use for admin.
Google Gemini, for most people, because it is free and already connected to the mail, calendar and maps in your Google account. In our seven-day test it captured 11 of 11 buried action items without inventing a single task, and it answered a same-day local question correctly on the first pass. ChatGPT is the better assistant if you pay: it won the day-planning and document-extraction tasks, and its scheduled Tasks feature is the only one here that can act without being asked.
Only if it is given permission to see them. Gemini reads Gmail and Google Calendar through Workspace extensions, and Microsoft Copilot reads Outlook and Microsoft 365 calendars. ChatGPT and Claude need a connector or a paste. That access is also the privacy decision you are making: an assistant that can read your inbox is, by definition, an assistant that can read your inbox, so review the vendor’s data controls first.
For inbox triage, quick lookups and first drafts, yes. The free tiers produced usable output on all four of our tasks. What you pay for is volume, the newest models, and features like scheduled tasks and larger context windows. Start free and upgrade only when you hit a specific limit you can name.
ChatGPT, on our evidence. Given seven constraints and no permission to ask questions, it produced a feasible schedule and named the trade-off required, while Gemini silently dropped an appointment. If day planning for you means thinking out loud rather than solving a constraint problem, Claude is the better conversational partner.
Most people do not, and running three tabs is how you end up with three half-finished task lists. Pick one as your default, connect it to your calendar and files, and keep a second only if you have a clearly different job for it — Perplexity for research, say, or Claude for long documents.
Install Gemini and leave it as your default. It is free, it can already see your mail, calendar and map, and it made fewer mistakes on the two tasks that repeat every single day. Add ChatGPT Plus if you will pay for one upgrade — it is the better reasoner, the better extractor of deadlines, and the only one here that can act on a schedule. Everything else on the list is a specialist: Claude for prose, Copilot because it is already installed, Perplexity for answers, Notion AI if your day already lives in Notion. Test whichever you pick against this week’s real inbox before you commit; the four prompts above are the whole test.
On September 14 to 20, 2026 we ran the identical four-task set on Google Gemini (free tier, Gemini 3.8 Flash with Workspace extensions enabled) and ChatGPT (Plus account, GPT-5.6 with browsing on), in fresh sessions with memory cleared and no custom instructions. Both accounts are paid for by us. The warranty document was pasted as text in both sessions so that extraction was tested rather than file search, and neither model was allowed to ask a clarifying question on the planning task.
| Metric | Google Gemini (3.8 Flash) | ChatGPT (GPT-5.6) |
|---|---|---|
| Inbox: action items captured (of 11) | 11 of 11 ✓ | 10 of 11 |
| Inbox: phantom tasks invented | 0 ✓ | 1 |
| Inbox: duplicate threads merged | Yes ✓ | No (listed twice) |
| Inbox: FYI items correctly parked | 4 of 4 ✓ | 3 of 4 |
| Plan: constraints satisfied (of 7) | 6 of 7 — 15:15 pickup dropped | 7 of 7 ✓ |
| Plan: infeasible input reported | No | Yes, with the trade-off named ✓ |
| Plan: self-corrected when challenged | No — lost the deep-work block | Yes ✓ |
| Documents: return dates correct (of 3) | 2 of 3 (invented a 90-day window) | 3 of 3 ✓ |
| Documents: still-covered verdict correct | 1 of 2 | 2 of 2 ✓ |
| Local question: correct on first pass | Yes ✓ | No — needed the page pasted in |
| Local question: time to answer | 22 seconds ✓ | 1 min 48 s (two browse passes) |
| Cost to run the whole week | $0 ✓ | $20/month Plus |
| Winner | 🏆 Gemini — 2 of 4 tasks, both of them the daily ones | ChatGPT — 2 of 4 (planning, documents) |
Gemini won because it is free and already connected to the mail, calendar and map you use every day; ChatGPT won the two tasks that need real reasoning. Both have free tiers, so run this week’s real inbox through each before you pay for anything.
Keep exploring — these related comparisons and guides help you decide.