You can open ChatGPT, Claude, Gemini, Copilot and Perplexity right now, in five browser tabs, and not pay a penny for any of them. That’s an extraordinary amount of capability for zero pounds. It’s also where a lot of people quietly waste an afternoon, because the free tier of a chatbot and the paid tier are not the same product running at different prices. They’re different products that happen to share a logo.
Knowing which one you’re actually using changes how you use it.
Three different things get called “free”
When a chatbot says it’s free, it’s usually one of these:
- A free tier of a paid product. ChatGPT, Claude and Gemini all work this way. You get a real model, a real interface, and a cap on how much of the good stuff you can have before you’re throttled or downgraded.
- A subsidised product. The company is buying market share or training data. Free access is the marketing budget. That’s not a scam, but it does mean the terms can shift under you.
- Open weights you run yourself. Tools like Ollama let you run a capable model on your own laptop. Genuinely free, genuinely private, and limited by whatever hardware you own.
Most people end up in the first bucket, and that’s fine — as long as you understand the shape of the cap. The mistake is assuming every free chatbot is interchangeable. It isn’t, and picking the right one for the job matters more than picking the “best” one overall.
If you want a longer argument on that point, there’s a good case for why the best AI chatbot usually isn’t the one with the biggest brand.
What the main free tiers actually give you
ChatGPT
The free plan gets you a solid general-purpose model with a message cap, then quietly drops you to a smaller, faster model once you hit it. You get web search, file uploads in limited numbers, and image generation with restrictions. For everyday writing, explaining and brainstorming, it’s more than enough.
Claude
Claude’s free tier is strong on long documents and careful, structured writing. The cap bites faster if you’re feeding it a 40-page PDF, because long inputs chew through your allowance. When it’s available, it’s often the better choice for editing your own work rather than generating from scratch.
Gemini
Typically the most generous free allowance of the big three, and it plugs into Gmail, Docs and Search if you already live in Google’s ecosystem. The trade-off is a tendency to be agreeable — it’ll agree with a wrong premise unless you push back.
Perplexity
Different animal. It’s a search engine with a chat interface, and every answer comes with citations you can check. A handful of deeper “pro” searches a day on the free plan is usually plenty for research tasks where being wrong is expensive.
Copilot
Baked into Windows and Edge, which makes it the lowest-friction option if you just want a quick answer without opening anything. Also the one most likely to be running a smaller model than you’d choose deliberately.
The limits nobody puts on the pricing page
Rate limits are the obvious one. The less obvious ones cause more trouble.
Silent downgrades. Hit your cap and the chatbot doesn’t stop — it swaps to a weaker model mid-conversation. You’ll notice the answers getting vaguer and more generic and assume the tool has got worse. It hasn’t; you’ve been moved.
Context windows. Free tiers often handle shorter conversations than paid ones. Paste in a long transcript and the start of it may fall out of memory by the end. That produces confident answers that ignore the brief you gave it four paragraphs ago.
Data use. On several free tiers, your conversations may be used for training by default. Paid plans usually let you opt out. If you’re pasting client contracts or unpublished work into a free chatbot, check that setting first — it takes thirty seconds and it’s the single most common mistake people make.
How to get noticeably better results from a free plan
Most of the gap between a free and paid chatbot is capacity, not intelligence. You can close a lot of it with technique.
- Front-load the context. Say who you are, who it’s for, and what good looks like in the first two sentences. Vague prompts get vague answers regardless of which model you’re on.
- Give it the source material. Paste the actual email thread, the actual brief, the actual competitor page. Asking a chatbot to guess at context burns tokens and produces mush.
- Ask for three options, not one. The first answer is usually the most obvious answer. Options two and three are where the useful thinking lives.
- Break the job into steps. One long mega-prompt eats your allowance and gives the model more chances to drift. Five short exchanges are cheaper and better.
- Start a fresh chat when a thread goes sideways. Correcting a confused conversation costs more messages than beginning again.
- Verify anything you’d publish. Free tiers hallucinate at roughly the same rate as paid ones on factual questions. Citations and a quick search are not optional.
For short, self-contained jobs — reformatting a table, drafting a reply, summarising a page — you can skip the account entirely. There’s a decent rundown of what you can realistically finish in ten minutes with AI tools that need no signup, and the honest answer is: quite a lot, as long as the task has a clear end point.
Where free chatbots earn their keep
Content work is the obvious win. Outlining, rewriting a stubborn paragraph, generating twenty headline variants so you can pick one, turning a rambling voice note into a structured draft. None of that needs a frontier model. A free tier handles it comfortably, which is why most writers I know run a mostly-free stack for content work and only pay for the one or two tools they touch every single day.
Admin is the underrated one. Drafting a polite chase email, turning bullet points into a project update, rewriting a message so it doesn’t sound passive-aggressive. Small jobs, high frequency, and exactly what a free chatbot is good at.
Learning is the third. “Explain this like I’m smart but new.” “What am I missing here?” “Quiz me on this chapter.” The tolerance for occasional inaccuracy is higher because you’re building understanding, not shipping a document.
When free starts costing you more than it saves
There’s a threshold, and it’s usually about frequency rather than difficulty. If you’re opening a chatbot more than a dozen times a day, hitting caps by mid-afternoon and waiting or switching tools, the twenty quid a month is cheaper than the friction. The same applies if you’re regularly working with long documents, need reliable file handling, or want the guarantee that your inputs aren’t training data.
Business use tends to cross that line fastest, mostly because the cost of a wrong answer is higher. There’s a useful breakdown of which AI tools are worth paying for in a small business that makes a sensible distinction between tools that replace hours and tools that just feel modern.
What I’d push back on is the assumption that paying automatically gets you better output. It gets you more capacity, longer context and fewer interruptions. The quality of what comes back still depends almost entirely on the quality of what you put in. People who write careful prompts on a free plan consistently beat people who write lazy prompts on a $200/month tier.
Pick one, learn it properly, stop shopping
The most common mistake with free AI chatbots isn’t choosing badly. It’s never choosing at all — spending weeks flitting between five of them, half-learning each, and concluding that none of them are much good. They’re all much good. They’re just not magic, and none of them will do your thinking for you.
So pick one general-purpose chatbot and keep it as your default. Add one search-first tool like Perplexity for anything factual. Use the free tier hard for a fortnight and see where it actually breaks for you — the specific failure will tell you far more about whether you need to pay than any comparison article, including this one. If the answer is “it doesn’t break”, congratulations: you’ve got a capable assistant for nothing, and you can spend the subscription money on something that does.