Choosing and comparing tools safely

A comparison can include output quality, sources, language support, device needs, cost, and privacy. Write down the criteria before a flashy demonstration influences your choice.

Use a harmless sample before uploading real school, family, or work information. A trial can reveal errors, hidden limits, or a poor fit without exposing sensitive data.

Running a modern AI service is genuinely expensive — the computation behind each chatbot answer costs real money in specialised processors and electricity — so when a tool charges you nothing, the cost is being covered somehow, and knowing the main business models tells you what to expect from any tool before you read a single review. The first model is advertising: the service is free because your attention and your behavioural data are the product sold to advertisers, which means the tool is designed to learn as much about you as possible and to keep you engaged as long as possible. The second is freemium: a limited free tier exists to recruit paying subscribers, so expect quality caps, usage limits, and steady nudges to upgrade — but comparatively less pressure to harvest your data, since subscriptions are the revenue. The third is data collection for improvement: many AI services state in their terms that your conversations may be used to train future models, meaning your prompts become part of the product; this is exactly why pasting private information into free tools is risky. The fourth is investor subsidy: some services run at a loss to win market share, which is pleasant while it lasts but explains why beloved free tools suddenly add fees, tighten limits, or shut down — the subsidy ended. Work through an example: a teacher comparing two lesson-planning tools finds one free with adverts and one charging five dollars monthly. Applying the lens, the free tool likely profiles its users and may train on whatever they type, so pupil information must never enter it; the paid tool's incentives point toward keeping subscribers satisfied and its terms likely allow opting out of training. Neither answer is automatically right — the free tool may be perfectly fine for generating generic quiz questions — but the choice is now informed rather than accidental. The misconception to correct is that free means generous and paid means exploitative. Often the opposite alignment holds: a paid tool answers to its customers, while a free tool answers to whoever actually pays, usually advertisers or a data pipeline. The single habit this yields is powerful: before adopting any tool, spend one minute answering the question of who pays for this and how, because that answer predicts the tool's behaviour toward you better than its marketing ever will.

Privacy policies are long because they are written to protect the company, not to inform you — but you do not need to read all of it, because five specific questions extract most of what matters, and each can usually be answered with the document's own search function in a few minutes. First: what do they collect? Look beyond the obvious account details for the reach of collection — contacts, location, microphone, files, browsing activity. The test is proportionality: a note-taking app plausibly needs your notes; it does not need your contact list, and a request without an obvious purpose is a signal. Second: is your content used for training? Search the policy for the words train, improve, and model. Tools differ genuinely here — some train on all free-tier conversations, some allow opting out in settings, some paid tiers exclude training by default — and this single answer decides whether the tool can ever touch sensitive material. Third: who else gets the data? The phrase third parties covers everything from payment processors, which are unavoidable plumbing, to marketing partners, which means your information is merchandise. Fourth: how long is it kept, and can you delete it? A real deletion process with a stated retention limit differs meaningfully from vague language about retaining data as necessary. Fifth: where does it go? Data stored abroad is governed by foreign law as well as Zimbabwe's Data Protection Act, which matters most for organisations handling other people's information. Now the worked example: a school administrator evaluates an AI marking assistant. Ten minutes of searching the policy reveals that uploaded work may be used for service improvement, that data is shared with unnamed partners for analytics, and that there is no deletion guarantee. Without those ten minutes, hundreds of pupils' essays — the personal data of children, the most protected category — would have flowed into an unknown pipeline. With them, the school either chooses a different tool or negotiates a contract with proper terms. The misconception to correct is that everyone consents to everything anyway, so reading is pointless. For your own casual use that cynicism is at least your own risk to take; the moment you handle other people's information — pupils, patients, customers, family — checking becomes a duty, and the five questions turn an unreadable document into a fifteen-minute safeguard.

A learner finds a free AI note-taking app that asks to access contacts and all files. A responsible response is: Review whether those permissions are necessary and choose a safer option if they are not. Correct. Permissions should match the task, not exceed it without reason. Use this case to separate what the technology contributes from what people contribute. The team should compare the intended outcome with a baseline where applicable, record important assumptions, and keep a clear route to correct or stop the process.

Ask whether the examples, data, and assumptions fit your school, company, or community. Generic demos often miss local names, laws, connectivity, and languages.

Before installing a free AI app, check what data it wants. Prefer tools that match the task without unnecessary access.