Which AI assistant fits your work: ChatGPT, Claude, Gemini, Perplexity, or Microsoft Copilot? Start with the task and the apps you already use. These services overlap in writing, research, and analysis; a useful comparison looks at workflow, access, and the amount of checking you need to do.
Checked: 8 September 2026. The product overview compares published capabilities. A separate same-task trial added September 14, 2026 records three actual responses and their limitations. Suggested uses below are editorial starting points. Availability can vary by plan, country, account, and app. Here, Microsoft Copilot means Microsoft 365 Copilot for Office work; it is not GitHub Copilot or a claim about every feature in the standalone Copilot chat.
Need a video, a song, an image, or a finished presentation? Use our AI tools by task guide for creative and office-work options, practical briefs, and free-plan conditions.
ChatGPT: A Broad Starting Point
Features: OpenAI’s official use cases cover drafting, learning, working with datasets, and creating presentations, alongside coding workflows. This makes ChatGPT a candidate when your tasks change from one day to the next.
Try it for: Turn your own notes into an outline, revise the draft, and ask for a checklist of facts to verify. For a spreadsheet task, ask it to explain the inputs and calculations behind its conclusions.
Access and limits: Free and paid ChatGPT plans exist. Check the exact feature you need in your account rather than assuming every workflow is included. OpenAI’s Work and Codex pricing guide explains plan-dependent allowances for those features; those allowances should not be treated as a universal limit for every ChatGPT feature. Verify calculations and generated files before using them.
Claude: Writing, Documents, and Code
Features: Claude supports writing and editing, code generation, data visualization, and web search. Its official plan comparison also distinguishes chat features from tools such as Claude Code.
Try it for: Revise a report in several passes: first structure, then clarity, then consistency. Supply the source material and ask it to identify claims that need support. Developers can separately evaluate Claude Code for repository work.
Access and limits: Free offers core chat capabilities. Pro adds more usage and access to tools including Claude Code; Max offers higher usage options. Paid access still has limits. Review which tools your plan includes, and check whether a rewrite changed dates, qualifications, or commitments. This guide does not establish that Claude writes better than the other assistants.
Gemini: Google-Connected Work
Features: Gemini can work with connected Google services. Google’s connected-app overview describes tasks such as finding information in Gmail and using other Google apps.
Try it for: Find relevant email details and prepare a short meeting brief, when the necessary connection is available and enabled. Check that the assistant used the intended messages and dates.
Access and limits: Google offers free Gemini access and paid Google AI plans with higher limits and additional features. The subscription comparison distinguishes these levels. Connected apps and Gemini built into Gmail or Docs are not interchangeable entitlements: confirm the specific feature, account type, and region. An existing Google account does not automatically provide every integration.
Update: Gemini has a Windows desktop app
Checked September 11, 2026. Google launched the Gemini app for Windows 10 and 11 on September 10. Its Alt + Space shortcut opens Gemini over your current work. The desktop app provides another way to use Gemini, while feature access still depends on your account. The launch announcement marks Gemini Spark and Gemini Omni with Google AI subscription, availability, and age-18-plus conditions. See the official launch details and our Gemini Windows app setup guide.
Perplexity: Research With Sources to Review
Features: Perplexity describes a search-led approach that gathers web information to answer questions. Its product explanation makes it a useful candidate for researching a topic and following sources.
Try it for: Ask for recent official guidance on a topic, with publication dates and links. Open those pages and check whether they support the summary, including any important exceptions.
Access and limits: Pro expands research access, uploads, and model choices compared with standard access. Some advanced-model access can be limited during heavy usage. Perplexity is a service that can offer several models, not simply one model to rank against all others. Citations make verification easier, but do not guarantee correctness.
Microsoft 365 Copilot: Work Inside Office Apps
Features: Microsoft’s Copilot overview covers help with documents, spreadsheets, presentations, and other Microsoft 365 work.
Try it for: Revise text in Word, explore a workbook in Excel, or build a presentation draft in PowerPoint. It belongs on your shortlist when editing the existing Office file is an important part of the task.
Access and limits: Check your Microsoft 365 subscription and account. Microsoft documents Copilot in Personal, Family, and Premium, including credit-based usage. Work accounts can have different licensing and administrator settings. Access to standalone Copilot chat does not by itself establish access to every Office feature. Review formulas, slide claims, and edits in the actual file.
Actual same-task trial: September 14, 2026
We submitted the identical English prompt below once to each of three web assistants, with all three tasks in one message and no retries. It used fictional notes and tables and requested no browsing, file creation, or external actions. ChatGPT showed a Pro account but no exact model identifier was recorded; Claude showed Free / Sonnet 5 / Medium; Gemini showed Flash, with the exact subscription unconfirmed. These are unequal account conditions, so this small observation cannot establish an overall winner. Perplexity and Microsoft 365 Copilot were not tested.
| Check | ChatGPT | Claude | Gemini |
|---|---|---|---|
| Preserve the explicitly unapproved budget | Preserved: “remains unapproved.” | Error: changed it to “approval status is unknown.” | Preserved: “awaits approval.” |
| Count tickets, include zero, exclude missing time | Correct counts and means | Correct counts and means | Correct counts and means |
| Avoid claiming equal team performance | Correctly declined that conclusion | Correctly declined that conclusion | Correctly declined that conclusion |
| Fixed SUMIFS ranges and expected totals | Correct: 140 / 200 / 0 | Correct: 140 / 200 / 0 | Correct: 140 / 200 / 0 |
| New row outside the original range | Explained that 140 stays unchanged instead of 150 | Explained that 140 stays unchanged instead of 150 | Explained that 140 stays unchanged and omits 10 |
Editorial corrections still needed: all three turned the security-review prerequisite into a before-pilot timing statement. ChatGPT and Claude also marked the exact deadline unknown; Gemini did not. The notes give a condition, not an agreed review deadline, so label this as an inferred dependency. Claude and Gemini also introduced the unstated example order ID O-107 for row 8. Do not copy these additions as source facts. All three summaries stayed under 70 words and retained Maya’s checklist date and the unassigned volunteer recruitment.
Answer key: the proposed budget is USD 2,000 before tax and is unapproved. Team A has 3 tickets, 2 recorded times and 1 missing time; Team B has 2, 2 and 0; overall counts are 5, 4 and 1. Each recorded-time mean is 15 minutes. Equal means do not prove equal performance. A suitable formula is =SUMIFS($D$2:$D$7,$B$2:$B$7,F2,$C$2:$C$7,"Paid"). Appending row 8 leaves its original result at 140; including the extra 10 would produce 150.
Exact prompt: reproduce this small trial
This is a small editorial verification exercise using only fictional data. Reply in English. Use only the information below; do not browse, create files, or take actions. Complete all three tasks and show enough reasoning to check the result. Do not invent missing facts. TASK 1 — Summary. Source notes: A pilot is proposed for October 14, 2026, only if security review is approved. Maya agreed to send a readiness checklist by October 7, 2026. Twenty volunteers are wanted, but nobody has been assigned to recruit them. The proposed budget is equipment USD 1,200 plus training USD 800, before tax. The budget has not been approved. Return a summary of at most 70 words and an action list with owner and deadline; mark unknown fields explicitly. TASK 2 — Table. Each row is one support ticket. Blank means time not recorded; zero is a valid recorded time. Ticket,Team,Handling_minutes T1,A,10 T2,A,20 T3,A,blank T4,B,0 T5,B,30 Return each team and overall ticket count, recorded-time count, missing-time count, and mean among recorded values. State whether the data establishes equal team performance. TASK 3 — Excel. Headers A1:D1 are Order,Region,Status,Amount. Rows 2–7 are: O-101,East,Paid,120 O-102,East,Unpaid,80 O-103,West,Paid,200 O-104,East,Paid,0 O-105,West,Refunded,50 O-106,East,Paid,20 F2=East, F3=West, F4=North. Write an English Excel formula with comma separators for G2 that adds Amount only where Region matches F2 AND Status is Paid, and can be copied down to G4. Keep the source ranges fixed and return 0 for no matches. Show expected G2:G4. Explain what happens if a new East/Paid/10 row is appended at row 8 without changing the original formula. Finish with any limitations of this exercise.
Limits: one response per service, three short tasks, no repeated sampling, no timing comparison, and no spreadsheet opened or executed. We checked the returned text and formula logic. Results can change with the model, plan, prompt, and date. This trial illustrates why source fidelity needs review even when arithmetic is correct; it does not test every capability described in this guide.
Compare the Same Task Before Paying
Pick two candidates and give them the same small, non-sensitive task. Use the same source material and requested format. Record the plan and mode used, whether browsing or file access was enabled, factual mistakes, and how much editing the result needs. A test of one task supports a decision about that task, not a permanent overall ranking.
Using only the notes below, draft a 150-word update for a colleague. Preserve dates and numbers, list open questions separately, and mark missing information as “not stated.” Return the draft followed by a short checklist of facts to verify.
Notes:
[Paste the same sample into each assistant.]
If you mainly use Google apps, start by checking Gemini’s relevant integrations. For Office files, check Microsoft 365 Copilot. For source-led web research, include Perplexity. For a mixed workload or iterative writing, compare ChatGPT and Claude on your own examples. Other assistants may handle the same tasks well; app fit and review effort are useful deciding factors.
Use our guides to writing clearer prompts, checking document summaries, rewriting emails, and studying with AI as sample tasks. Confirm current local pricing and feature limits on the linked official pages before buying a plan.
Choosing for your everyday job? Our AI guide for office workers starts with five working styles and gives each one a practical trial assignment, access checks, and reasons to consider another option.