ChatGPT at a glance
| What it is | A general-purpose assistant for writing, analysis, coding, images and voice |
| Free tier | Yes, permanent — with rate limits and access to fewer capable models |
| Paid tiers | Individual, team and enterprise; higher limits and earlier feature access |
| Custom GPTs | Configured assistants with instructions and uploaded knowledge, shareable via a directory |
| Memory | Retains details across conversations; can be viewed, edited and switched off |
| Data handling | Consumer chats may be used for training unless you opt out; business tiers are excluded by default |
| Ecosystem | The largest — apps, extensions, tutorials and integrations exceed every rival |
| Poor fit for | Confidential material on the consumer tier, work needing guaranteed citations |
What paying actually gets you
The thing that matters
Rate limits. The free tier throttles how much you can do with the better models, and paid tiers raise that ceiling. If you have never hit a limit, you have no reason to pay; if you hit one weekly, that is the entire argument.
The things that sound more important than they are
Custom GPTs get heavy marketing and light use. A configured assistant with a fixed instruction set is genuinely useful for a task you repeat weekly — a specific report format, a recurring code review style. Most people build one, use it twice, and never open the directory again.
Feature access arrives earlier on paid tiers, which matters if you want to be first and not otherwise.
Memory, and the awkward part
Memory makes the assistant less repetitive — it stops re-establishing who you are and what you work on. It also accumulates inferences you did not deliberately provide, and those persist into unrelated conversations.
Review it periodically. The settings expose exactly what has been retained, and reading that list is occasionally surprising. Turning memory off is a legitimate choice, particularly on a shared or work machine.
Getting more out of it than most people do
| 1. Say what the output is for | “A paragraph for a board paper read by non-technical directors” produces something usable. “Explain this” produces a lecture. Purpose beats phrasing. |
| 2. Give it the source material | Paste the actual document, data or code. Answers grounded in what you supplied are dramatically more reliable than answers from memory. |
| 3. Ask it to argue against itself | “What is wrong with that answer?” surfaces weaknesses more reliably than any prompt trick, because it changes the task from generating to critiquing. |
| 4. Check every fact, name and number | It states wrong things in exactly the tone it states right things. This does not improve with a better prompt. |
| 5. Start a new chat when it drifts | Long conversations accumulate wrong assumptions. Restarting with a clean summary is faster than correcting them. |
| 6. Turn off training on consumer tiers | One setting, once. Do this before pasting anything from work into it. |
ChatGPT against the two it is shortlisted with
| What decides it | ChatGPT | Claude |
|---|---|---|
| Ecosystem | Largest by a distance | Smaller |
| Long documents | Competent | Stronger |
| Prose quality | Good | Often better |
| Images and voice | Built in | Limited |
| Integrations | Everywhere | Fewer |
| Choose when | You want one tool for everything | Writing and long context dominate |
Gemini is the third, and the deciding factor there is rarely capability — it is whether your documents and email already live in Google Workspace. If they do, the integration outweighs most feature differences.
For questions where the source matters more than the answer, Perplexity is a different category of tool and usually the better choice.
Where it is genuinely weak
- Confident wrongness. The failure mode is not saying “I don’t know” — it is a fluent, plausible, incorrect answer. This is inherent, not a bug awaiting a fix.
- Citations. Unless it has searched, references are unreliable and sometimes fabricated.
- Recency. Training has a cutoff, and without search it will answer confidently about a world that has moved on.
- Arithmetic and counting. Better than it was, still not a calculator.
- Consumer data handling. Fine for personal use, a policy question for anything confidential.
Who should pay for it
Anyone hitting free-tier limits weekly — that is the whole test, and it beats any feature comparison. People who use it daily for work, where the subscription is trivial against the time. Anyone who wants images, voice and analysis in one place rather than assembled from three tools.
Not worth paying for if the free tier has never stopped you, if your main need is sourced research, or if your employer prohibits pasting work material into consumer AI services — in which case the answer is a business tier, not a personal one.
Other general assistants
- Claude — stronger on long documents and prose.
- Google Gemini — if your work already lives in Workspace.
- Perplexity AI — when you need sources, not just answers.
- Jan — if the requirement is that nothing leaves your machine.
Compiled from OpenAI’s documentation and public sources. We have not hands-on tested this tool. Last reviewed 17 August 2026.
