For an Australian professional firm building a small internal tool without a developer, the short answer is: Claude is the easiest to work with for building a working mini-app you can see and use, ChatGPT is the most forgiving all-rounder, and Gemini is the strongest pick if you live in spreadsheets and Google Workspace. If your firm already runs Microsoft 365, Copilot is the one that stays inside your own tenant — which matters more than raw capability for anything touching client data.
This article is for the partner, practice manager, bookkeeper or office admin who wants a calculator, a formula or a checklist — not a career change. No code gets walked through line by line. We compare how each assistant follows plain-English instructions, shows you the result, and recovers when the first attempt is wrong.
The three builds we compared
To make the comparison fair, we used the same three jobs a real firm might actually want. None of them needs a developer, and all three are achievable in an afternoon on a free or entry paid tier.
- A calculator — a simple fee or settlement-date estimator with a few inputs and one clear answer.
- A spreadsheet formula — a single formula or short script to clean up or summarise a column of data you already have.
- A checklist generator — a one-page tool that spits out a tailored document-request or onboarding checklist based on a few dropdowns.
These are the sweet spot for vibe coding. If you are new to the idea, our plain-English guide to vibe coding explains the approach before you start.
Claude: best for seeing the tool as you build it
Claude's Artifacts feature is the headline here. When you ask it to build a calculator or checklist generator, it opens a live preview beside the chat — a working thing you can click, type into and test immediately. For a non-coder that preview is the whole game, because you judge the result by using it, not by reading code.
The build flow is natural. You describe what you want in a sentence, Claude produces a working version, and you refine it by talking. "Make the date field a dropdown." "Add GST to the total." "The heading is too big." It handles that conversation well and rarely loses track of what it already built.
Where to start: claude.ai, free tier works for a first build, the paid Pro tier gives you more usage and Projects — a place to keep a tool and its instructions together so you can return to it. Our walkthrough on building a client intake app with Lovable or Claude shows this in more depth.
ChatGPT: the most forgiving all-rounder
ChatGPT's equivalent to Artifacts is Canvas, a side panel where it writes and edits a document or small app you can revise in place. It is a little less immediate than Claude's live preview for interactive tools, but it is excellent at the spreadsheet-formula and checklist jobs, and it is the most patient when you describe things imprecisely.
If your request is vague — "build me something to work out conveyancing fees" — ChatGPT will ask sensible clarifying questions rather than guessing badly. That matters for beginners who do not yet know how to specify what they want.
ChatGPT also has custom GPTs, which let you save a tool with fixed instructions so staff can reuse it without re-explaining. We cover that pattern in building a custom GPT or Claude Project for your firm. Start at chatgpt.com; the free tier covers a first build, with the paid Plus tier giving fuller access to Canvas and custom GPTs.
Gemini: strongest if you live in Google Workspace and spreadsheets
Gemini has its own Canvas mode for building and previewing small apps, and it has caught up quickly. Its real advantage for a professional firm is proximity: if your practice runs Google Workspace, Gemini sits inside Gmail, Docs and Sheets, so a formula it writes can go straight where you need it.
For heavier spreadsheet work and Google Apps Script, Google AI Studio (aistudio.google.com, free to start) is a more capable environment that still talks to you in plain English. We used exactly this approach in our guide to AI-generated Google Apps Script for real estate agents.
Which AI is best at spreadsheets?
All three write competent Excel and Google Sheets formulas from a plain-English description such as "give me a formula that flags any invoice older than 30 days". The differences show up in the awkward cases.
- Gemini has the edge if you are in Google Sheets, because it understands the Sheets function set and Apps Script closely and can drop results straight in.
- ChatGPT is the most reliable for explaining why a formula works and for untangling a broken one you paste in — useful when you inherit a spreadsheet nobody understands.
- Claude is strong at turning a messy spreadsheet into a small working tool rather than just a formula.
Whichever you use, always test the formula on a copy of your data first. If you are trying to retire an old macro, our guide to replacing Excel macros with AI-written scripts goes further.
Claude Artifacts vs ChatGPT Canvas vs Gemini Canvas, at a glance
| What matters to a beginner | Claude (Artifacts) | ChatGPT (Canvas) | Gemini (Canvas / AI Studio) |
|---|---|---|---|
| Live preview you can click and test | Strongest | Good | Good |
| Handling a vague request | Good | Strongest | Good |
| Fixing its own mistakes on request | Strong | Strong | Good |
| Spreadsheets and Google Workspace | Good | Good | Strongest |
| Saving a reusable tool for staff | Projects | Custom GPTs | Gems |
| Free tier good enough for a first build | Yes | Yes | Yes |
These are qualitative impressions, not benchmarks. The products change monthly, so treat the table as a starting point and re-test for yourself — there is a quick method for that below.
The real skill: reacting when the first result is wrong
Every one of these tools will get something wrong on the first go. The skill is not prompting perfectly — it is correcting clearly. Beginners give up here; people who get results keep talking.
When a build is off, say exactly what you see and what you expected, in one specific instruction at a time:
- "The total isn't adding GST — it should be 10% on top of the fee."
- "When I leave the date blank it shows an error. Make it ignore blank dates."
- "Good, now change the two text boxes to dropdowns with these options..."
If three or four rounds of correction still leave it broken, stop and start a fresh chat describing the whole tool from scratch. That almost always works better than wrestling a confused conversation back on track.
What about Microsoft Copilot for firms on 365?
If your firm runs Microsoft 365, Copilot deserves a serious look — not because it builds tools better, but because it runs inside your own Microsoft tenant, under your existing licensing and data-handling arrangements. For anything that will ever touch client information, that containment matters more than whether the preview is slightly nicer in Claude.
Copilot is strongest for Excel formulas, summarising documents and drafting inside Word and Outlook. It is less of a free-form app builder than Claude or ChatGPT. We compare the two directly in Microsoft Copilot vs ChatGPT for accounting firms.
The privacy line you must not cross
None of these tools changes the rule: do not paste real client personal information into a public chatbot while you are building. The OAIC's general guidance on commercially available AI products is clear that the Privacy Act applies the moment personal information goes into an AI system, and that publicly available chatbots should not be fed personal information.
In practice that means build and test with fake data, use your firm's already-licensed AI first (Copilot in your Microsoft tenant, or Gemini in Google Workspace) for anything sensitive, and keep a short record of where data flows. This is general guidance, not legal advice — our article on AI and client data under the Australian Privacy Act sets out the full position.
A tool you vibe-coded in an afternoon should never go client-facing without review, should never handle logins or payments, and should never be the thing that has to be right every time. When it reaches that point, hand it to a developer. The vibe coding risks for professional services firms article covers where that line sits and how to keep these tools out of shadow IT.
How to re-test for yourself in 20 minutes
Because these products move fast, pick the build closest to your own need and run it through two of them side by side. Use the same plain-English request in each, make the same correction when they get it wrong, and judge on three things: did the preview let you actually use it, did it follow your instruction, and did one correction fix the problem? The winner on your specific job is the one to keep.
If you would rather see the tools in the context of real workflows a firm can automate first, our free guide, 10 AI Workflows to Save 10+ Hours a Week, lays out where this kind of build pays off — and where you are better off automating an existing process than building anything new at all.
Common questions
What is the best AI for non-coders to build apps?
For a complete beginner, Claude is usually the gentlest start because its Artifacts feature shows a working, clickable version of your tool beside the chat, so you judge it by using it. ChatGPT is the most forgiving if you struggle to describe what you want. Try both free tiers on your actual task and keep whichever gets you a usable result faster.
Which AI is best at spreadsheets and formulas?
All three write solid Excel and Google Sheets formulas from a plain-English description. Gemini has the edge inside Google Workspace because it understands Sheets functions and Apps Script closely. ChatGPT is best for explaining why a formula works and for fixing a broken one you paste in. Always test any formula on a copy of your data first.
Can I use these tools with real client data?
Not in public chatbot versions. The OAIC's general guidance is that the Privacy Act applies once personal information enters an AI system, and that publicly available chatbots should not be given personal information. Build and test with fake data, and for anything sensitive use your firm's licensed AI — Microsoft Copilot in your own tenant or Gemini in Google Workspace.
Should a Microsoft 365 firm just use Copilot instead?
For tool building specifically, Claude and ChatGPT are more flexible. But Copilot runs inside your own Microsoft tenant under existing licensing, which matters for anything touching client information. A practical approach is to prototype with fake data in Claude or ChatGPT, then rebuild the final version in Copilot or your Google Workspace Gemini where the data governance is already handled.
What should I never build with these AI tools?
Avoid anything client-facing without review, anything handling logins or payments, anything that must be correct every single time, and anything processing real client personal data. These afternoon builds are for internal calculators, formulas and checklists. When a tool becomes mission-critical or needs secure data handling, stop and hand it to a developer before it causes a problem.
Want these ideas working in your firm? We build controlled AI workflows for Australian professional services firms — starting with a free automation audit.