Choosing the Right Tool
There are thousands of AI products and new ones every week. You do not need to know them all — you need to know the categories, what each is good for, and six questions that tell you whether a tool is safe and sensible to use.
A clinic administrator wants help with three jobs: writing patient reminder letters in Bahasa Melayu and English, turning recorded staff meetings into minutes, and finding the latest public-health guidance on dengue. She tries one chat assistant for all three. The letters are fine, the minutes are impossible because the assistant cannot take audio in the way she needs, and the dengue answer quotes guidance that is two years out of date. Three different jobs, three different kinds of tool. Picking well starts with the job, not the brand.
Start with the job, not the tool
Before reaching for any product, describe the task in one line and ask three things: What goes in (text, a document, audio, a spreadsheet, a photo)? What must come out (a draft, an answer with sources, an image, minutes)? How sensitive is the material? Those answers usually point to a category straight away. The products named below are examples to help you recognise each category, not recommendations; features, prices and names change often As of 2026.
The main categories
Tap each category to see what it is for and where it falls short.
CHATGeneral assistantsThe all-rounders
Examples: ChatGPT, Claude, Gemini, Microsoft Copilot, Meta AI; and Malaysia’s ILMU, launched by YTL AI Labs in 2025 with a focus on Bahasa Melayu As of 2026.
Good for: drafting, rewriting, explaining, brainstorming, working through problems, analysing uploaded files. Watch for: facts from memory may be outdated or invented unless the assistant searches the web and shows its sources.
FINDAI search and answer enginesAnswers with links
Examples: Perplexity, Google’s AI Overviews and AI Mode, ChatGPT search, Copilot in Bing As of 2026.
Good for: current events, quick factual questions, finding sources to read. Watch for: a cited link does not prove the claim — the summary can misstate what the page says. See AI for Research and Learning.
EDITWriting toolsPolish inside your documents
Examples: Grammarly, QuillBot, and the rewrite features now built into many phones, email apps and word processors As of 2026.
Good for: grammar, tone adjustments and short rewrites without leaving the app you are writing in. Watch for: heavy rewriting can flatten your voice; browser extensions may read everything you type, so check what they are allowed to see.
IMGImage generatorsPictures from words
Examples: Midjourney, Adobe Firefly, Canva’s AI features, and image generation inside ChatGPT and Gemini As of 2026.
Good for: illustrations, mock-ups, presentation visuals. Watch for: copyright and licensing terms differ between tools, text inside images can still go wrong, and realistic images of real people raise serious ethical issues. See Images, Audio and Video.
RECTranscription and meeting assistantsSpeech to notes
Examples: Otter.ai, Zoom AI Companion, the recap features in Microsoft Teams and Google Meet, and transcription built into many phone recorders As of 2026.
Good for: minutes, action lists, searchable records of talks. Watch for: everyone in the meeting should know it is being recorded; accuracy drops with accents, cross-talk and code-switching between languages — a very Malaysian meeting habit.
365Office-suite assistantsAI inside the tools you already use
Examples: Microsoft 365 Copilot in Word, Excel, PowerPoint, Outlook and Teams (Microsoft has been merging its work and consumer Copilot apps into a single “Microsoft Copilot” app, so the names overlap); and Gemini in Google Workspace (in Gmail, Docs, Sheets, Slides and Meet) As of 2026.
Good for: work that lives in your organisation’s files and email — summarising a long thread, drafting from an existing document, building slides from a report. Because they run under your organisation’s account and settings, they are often the tools an employer approves first. Watch for: they can only be as careful as your file permissions; if a folder is shared too widely, the assistant can surface it too.
SPECSpecialist appsBuilt for one domain
Examples: NotebookLM (answers grounded in documents you add), research tools for academic papers such as Elicit and Consensus, translation services such as Google Translate and DeepL, plus a growing range of legal, medical, accounting and education tools As of 2026.
Good for: tasks where a narrower tool is more reliable or better grounded than a general one. Watch for: small vendors may have weaker privacy commitments; check who they are and where data goes.
Six questions before you adopt a tool
This guide’s framing Work through these in order. The first two are deal-breakers: if a tool fails them for the material you have in mind, nothing further matters.
- Is it approved?For work material, check your organisation’s list of approved tools and the kinds of data each may handle. If there is no policy, ask your manager or IT before using a new tool on anything that is not public.
- Where does the data go?Read the privacy settings and terms: is your content stored, for how long, in which country, reviewed by people, or used for training? Can you switch training off? Business plans usually offer stronger commitments than free consumer plans.
- Is it accurate enough for this job?Test it on three or four tasks where you already know the right answer. Does it show sources? Does it admit uncertainty? Accuracy needed for a birthday poem is not the accuracy needed for a tender document.
- Does it handle your languages?Try it in the languages you actually use — Bahasa Melayu, Chinese, Tamil, English, and the mixtures people really speak. Quality varies by tool and by task.
- Does it fit where you work?A slightly weaker tool inside your email or documents can beat a stronger one that needs copying and pasting — and copying is where sensitive material leaks.
- What does it really cost?Free tiers have usage limits and may have weaker privacy terms. Paid plans are often charged in US dollars per person per month. Count the time spent checking output, too.
Languages: Bahasa Melayu, Chinese and Tamil
The major general assistants handle Bahasa Melayu and Chinese reasonably well for everyday tasks, and Tamil with more variable quality As of 2026. Common weak spots:
- Register and formality. Formal Malay for an official letter differs from everyday speech; models sometimes mix them or slip into Indonesian vocabulary.
- Code-switching. A sentence that moves between Malay, English and Cantonese is normal in Malaysia but harder for models and transcription tools.
- Simplified versus Traditional Chinese — say which you need.
- Cost and length. Non-English text often uses more tokens, so long documents fill the window faster (see Working with Context).
Run the same test in two or three tools and ask a fluent colleague which reads most naturally. That fifteen-minute experiment tells you more than any review site.
Free first — and a common dilemma
For learning and personal tasks, start with free tiers: most major assistants, AI search engines and NotebookLM offer one As of 2026. Pay only when you hit a limit you genuinely need lifted, such as longer files, more usage or a business-grade privacy commitment. At work, the question is usually not which tool is best but which tool is allowed.
Your employer approves only its office-suite assistant. A free chatbot gives you noticeably better drafts. What do you do?
Choose one to hear its case.
The case
The approved tool runs under your organisation’s data agreements, so client and staff information is covered. Better prompts often close much of the quality gap.
The weak point
You may be leaving real productivity on the table, and colleagues may quietly use unapproved tools anyway — “shadow AI” — which is worse than an open discussion.
The case
Public, non-confidential tasks — rewording a published announcement, learning a concept — carry little risk, and many policies allow exactly this.
The weak point
The line between sensitive and not is easy to blur under deadline pressure. One pasted client email undoes the logic. Check that your policy really permits it.
The case
Collect concrete side-by-side examples using non-sensitive material, note the privacy terms of a business plan, and bring them to IT or your manager. This is how good tools get approved properly.
The weak point
It takes time, and the answer may still be no for reasons you cannot see — procurement, security reviews, existing contracts. Meanwhile you keep using the approved tool.
You need the latest official guidance on a public-health topic. Which category fits best as a starting point?
Current facts need search and sources; the final check is the official page.
Which question should come first when choosing a tool for work documents?
Approval and data handling are deal-breakers; features and price come later.
Why can an office-suite assistant surface a document you were not meant to see?
Such assistants search what your account can already reach. Tidy permissions are part of safe adoption.
// Pause and reflect
- List the three AI-assisted jobs you do most. Which category does each belong to, and are you using a tool from that category?
- For the tool you use most, can you answer the “Where does the data go?” question? If not, where in its settings would you look?
Your notes are saved privately in this browser.
Pick one real, non-confidential task and run it through two different free tools from the same category — the language test prompt above works well. Score each from 1 to 5 on accuracy, language quality and ease of use, and write one line on its data settings. Keep the scorecard: it is the start of your own tool shortlist.
Key terms on this page
Sources & further reading
- Microsoft, Data, privacy and security for Microsoft Copilot (living documentation, accessed 2026).
- Google, Generative AI in Google Workspace privacy hub (living documentation, accessed 2026).
- YTL Power International, “YTL AI Labs launches ILMU” (press release, 2025).
- Petrov, A. et al., “Language Model Tokenizers Introduce Unfairness Between Languages” (2023).
- OpenAI, Prompt engineering guide (accessed 2026).