How to use ChatGPT for free (and get more out of it than most paying users)

Last verified: August 2026

That last point is the one everything else in this article depends on, so it goes first.

As of August 2026, none of the three major vendors publish numeric caps on their free tiers. Not OpenAI, not Google. The old world of "40 messages every three hours" is gone. Limits are now dynamic — they flex with prompt complexity, which features you use, how long your conversation is, and total demand on the service at that moment.

So if you find an article confidently telling you exactly how many free messages you get per day, it is either out of date or it was never sourced. Google's own documentation states plainly that limits "factor in the complexity of your prompt, the features you use, and the length of your chat," refreshing on a rolling basis up to a weekly ceiling. That is as specific as the vendor itself will go.

Everything below is dated August 2026 and should be re-checked against the vendors' own pricing pages before you rely on it.

How to start using ChatGPT free, step by step

1. Go to chatgpt.com. Not an app store, not a search result — the number of convincing paid clones charging for access to a free product is genuinely a problem.
2. Create an account. Email, or Google/Microsoft/Apple sign-in. No card details.
3. Verify your email. Complete the short onboarding.
4. Start typing. You are on the free tier by default with unlimited standard text chat. Nothing to configure.
5. Note the model selector. Free accounts get the standard model, with limited access to a reasoning variant.
6. Turn on memory or turn it off deliberately. Under Settings → Personalisation. On means better continuity across chats; off means less accumulated context you did not intend.
7. Check your data controls before pasting anything sensitive. Settings → Data Controls. Consumer tiers may use conversations for training unless you opt out. This is the single most important setting for business use.

That last step is where most UK businesses get themselves into trouble. Pasting client data, employee records or commercially sensitive material into a consumer AI tool is a UK GDPR question, not a preference. If the data identifies a living person, you need a lawful basis and you need to know where it goes. Free consumer tiers are the wrong place for it. Paid business tiers with data-processing agreements exist for exactly this reason.

What you actually get free, in August 2026

Two caveats stated plainly. Microsoft's free-tier documentation could not be verified directly at the time of writing, so treat that column as the least reliable. And model names on free tiers change without announcement — the specific names above will date faster than anything else here.

Google Gemini vs ChatGPT: which is better free?

Different tools, and the honest answer depends on one question: where does your work already live?

Choose ChatGPT for general reasoning, writing and code. It remains the strongest all-round free text model, and the unlimited standard-tier chat is more generous than it sounds — most people never touch the throttled features.

Choose Gemini if you are in Google Workspace, or if you need to feed in something very long. Its context window is its real advantage: whole contracts, long transcripts, large documents in one go, where other tools force you to chunk.

Choose Copilot if you are a Windows and Office business. It is already there, it is licensed, and the friction of adoption is near zero — which in practice beats a marginally better model your staff never open.

The realistic answer for most people is two of them. Use ChatGPT as the default thinking tool and whichever of Gemini or Copilot matches your document stack. There is no cost to running both, and the failure modes differ enough that a second opinion is genuinely useful on anything that matters.

What none of the free tiers will do is anything that needs to happen reliably, repeatedly, without a person present. A chatbot in a browser tab is a thinking tool. It is not a system. The moment you need something to run every time a form is submitted or a call is missed, you have left the chatbot's territory entirely — that is the gap our recruitment agency and trades clients usually come to us to close.

How to write AI prompts: the CTIR framework

Most people get mediocre output because they ask a mediocre question. The fix is not clever phrasing tricks. It is supplying the four things the model cannot infer.

C — Context. Who you are, what the situation is, what has already been tried. Models have no access to your circumstances unless you provide them.

T — Task. The specific output you want. Not "help me with my pricing" but "rewrite this pricing page so the three tiers are directly comparable." If pricing structure is the actual problem, our pricing page is a useful reference point for how tiers can be made directly comparable.

I — Instructions. The constraints. Length, format, tone, what to exclude, British English, what to do when it lacks information.

R — Role. Who the model should answer as. Genuinely changes which parts of its training get weighted.

Weak prompt

Write me a follow-up email to a lead.

CTIR prompt

Role: You are an experienced UK B2B salesperson who sells technical services to sceptical small-business owners.
Context: A dental practice owner in Hampshire booked a 30-minute automation audit two weeks ago, attended, received a written roadmap, and has not replied since. The roadmap recommended an AI receptionist at around £3,000, on the basis they were losing roughly six calls a day.
Task: Write a follow-up email of no more than 120 words.
Instructions: British English. No exclamation marks. Do not use "just checking in", "circling back", "touching base" or "I wanted to reach out". Reference the specific number from their audit. One clear question at the end. Assume they are busy, not uninterested. If you need information I have not given you, list it at the end rather than inventing it.

The second prompt takes ninety seconds to write and removes an hour of editing. That trade is the entire skill. Our dental practices and accountants pages have more on how that kind of follow-up gets automated once you know what "good" looks like.

Four techniques that reliably improve output

• Give it examples. Two or three samples of the output you want, then ask for a fourth. This outperforms any amount of description.
• Ask it to plan first. "Before writing, outline your approach in five bullets and wait." You catch a wrong direction before it produces two thousand words in it.
• Ban your own tells. Keep a list of phrases you never want to see and paste it into every prompt. The generic AI register is the fastest way to lose a reader.
• Make "I don't know" allowed. Explicitly permit the model to flag missing information. Unprompted, it fills gaps with plausible invention, because that is what next-token prediction does.

Two things that do not work

Politeness does not improve output. Neither does insisting it will lose its job. What looks like a personality is a probability distribution, and pressure is not an input to it.

And do not trust arithmetic, dates, prices or citations without checking. Language models generate the most probable text, and a plausible-looking reference is exactly the kind of text that is highly probable and completely false.

When a chatbot stops being the answer

There is a clear line, and it is worth naming because crossing it unknowingly wastes a lot of time.

A chatbot works when a person is present, the task varies each time, and someone reviews the output before it matters. Drafting, research, thinking out loud, first passes.

A chatbot fails when the task must happen every time without prompting, must follow the same path reliably, must read or write to your actual systems, or must run at three in the morning when nobody is watching.

That second list is not a prompt-engineering problem. It is a systems problem, and no amount of skill in a chat window solves it — because there is nobody in the window at three in the morning. See our Zapier vs n8n comparison for what building that kind of system actually involves.

The practical pattern we see working: use free chatbots to establish what a good output looks like, by hand, repeatedly, until the shape of it is clear. Then build the system that produces that output automatically. The manual phase is not wasted time — it is the specification, and skipping it is why so many automation projects produce something nobody wants.

Start with the audit, not the software

Free AI tools are genuinely good now, and for a lot of businesses the honest advice is to spend a month using them properly before spending anything on automation. You will learn more about where your time actually goes than any vendor can tell you.

When repetitive work is still there after that month, it is worth looking at properly.

Thirty minutes on a call, no pitch. We map how work moves through your business and tell you what is worth automating — and what is not.

How to use ChatGPT for free: FAQ

Sources and links

• OpenAI ChatGPT pricing — openai.com/chatgpt/pricing (verify current free-tier features and model names)
• Google Gemini subscriptions and usage limits — gemini.google.com/faq
• ICO guidance on AI and data protection — ico.org.uk — AI and data protection
• Internal: Zapier vs n8n guide · AI automation cost for UK small business · Free AI automation audit