TL;DR
Sign up free at claude.ai, then spend your first 30 minutes on one real task — a long PDF, a doc to draft, a function to refactor — not a toy puzzle. Run it on the default model, then build a Project so you stop re-pasting context. Claude pulls ahead of ChatGPT on long inputs and careful writing; it is not faster for quick lookups or images. As of June 2026, the free tier runs a limited Claude Sonnet 4.6; paid Pro is $20/mo ($17/mo billed annually) and adds Claude Opus 4.7 plus a 1M-token context window.
What this covers
A path from creating an account to having one genuinely useful task running through Claude — not a tour of menus. The trap most people fall into: they open Claude, type a casual question, get an answer that looks identical to ChatGPT, and conclude “it’s the same thing.” This guide aims you at the workflows where Claude measurably wins — long documents, careful prose, structured output — so the first half hour leaves you with something you would actually send.
Three concepts you’ll use:
- Claude: Anthropic’s assistant with file upload, a 1M-token context window on Opus 4.7 and Sonnet 4.6 (as of June 2026), web search, and connectors.
- Projects: Persistent workspaces with custom instructions and shared knowledge files. This is the main reason to pick Claude over a bare chat box.
- Artifacts: A side panel that renders editable output — code, docs, HTML, diagrams — that you iterate on without re-prompting from scratch.
Who this is for
People new to Claude, or comparing it head-to-head with ChatGPT and Gemini. Especially useful if you already pay for ChatGPT Plus and are deciding whether to add Claude or switch. For the full three-way breakdown, see Claude vs ChatGPT vs Gemini.
Pick a plan first
You can evaluate Claude entirely on the free tier, but knowing the tiers up front saves a wasted hour. Prices below are individual/consumer, as of June 2026 (official pricing).
| Plan | Price (USD/mo) | Models | Context | Usage |
|---|---|---|---|---|
| Free | $0 | Limited Sonnet 4.6 | Standard | Rolling ~5-hour cap; small uploads |
| Pro | $20 ($17 annual) | Opus 4.7 + Sonnet 4.6 | 1M tokens | ~5x Free; includes Claude Code + Claude Cowork; unlimited Projects |
| Max 5x | from $100 | Same as Pro | 1M tokens | 5x Pro session capacity, priority access |
| Max 20x | $200 | Same as Pro | 1M tokens | 20x Pro session capacity |
Two things to know before you pay:
- Max is not a smarter model. Max 5x and Max 20x run the same Opus 4.7 and Sonnet 4.6 as Pro — you are buying a larger usage bucket, not better answers. Only upgrade past Pro when you are routinely hitting limits during real work.
- Pro now bundles Claude Code and Claude Cowork. If you write code, the $20 Pro plan covers terminal-based agentic coding too, which used to be a separate purchase.
When to reach for Claude
Reach for Claude when the input is long (papers, transcripts, codebases), when the output needs careful prose (legal, marketing, technical docs), or when you want side-panel Artifacts you can iterate on. For longform drafting specifically, follow a real Claude writing workflow rather than one-off prompts. For quick factual lookups and image generation, ChatGPT is usually faster, so keep both.
Before you start
- Pick your evaluation task. Choose something you genuinely do weekly — summarizing PDFs, drafting a doc, refactoring a function — not a riddle.
- Collect the inputs: one long document, one code snippet, or one dataset. You will attach these in step 2.
- Decide success criteria up front. “Output I’d send without rewriting” is a far stronger bar than “looks good.”
- Note your current tool. If you use ChatGPT today, plan to run the identical task on both and compare side by side.
Step by step
- Sign up at claude.ai with email or Google. The free tier is enough for the first hour; upgrade only after you have hit a limit doing real work.
- Decide the training toggle during sign-up. Since Anthropic’s August 2025 terms change, consumer chats (Free, Pro, Max) are used to improve future models unless you opt out — and opting in extends data retention from 30 days to 5 years. Go to Settings → Privacy and set the model-training toggle the way you want before pasting anything sensitive (Anthropic’s data-use article).
- Run your evaluation task on the default model. When the output is something you want to edit and re-run (HTML, code, a doc), graduate to a deep workflow for Claude Artifacts.
- Try the thinking variant. Claude switches to extended thinking automatically on complex prompts, and you can also turn it on explicitly. Re-run the same task and compare quality against latency.
- Create your first Project. Add custom instructions like “Write in plain English, no marketing fluff, max 200 words unless asked.” Attach 2-3 reference files (Projects accept up to 30MB per file, unlimited files). This is where Claude becomes meaningfully better than a chat box: the instructions and files persist across every conversation in the Project.
- Connect one source — Google Drive, GitHub, or the Chrome extension — for a single task. Connectors remove the “let me paste this in” step. See Claude Connectors Setup.
- If the Download button on an Artifact ever returns nothing, see Claude Artifact download fails / empty file.
First-run exercise: the long-document test
This is the fastest way to feel where Claude differs from a generic chatbot.
- Pick a single 10-20 page PDF you actually need to read — a paper, a contract, a long report.
- Upload it to a fresh chat. Ask: “What are the three claims this document makes, and what evidence does each rest on?” Wait for the answer.
- Follow up: “Which of those claims is weakest? Quote the supporting text.” This forces Claude to pin its answer to the source rather than paraphrase.
- Run the same two prompts in your current tool. Note where Claude wins, where it loses, and which is faster.
A 1M-token window holds roughly a 500-page book, so a 20-page PDF leaves enormous headroom. The win here is not raw capacity — it is whether the model cites the actual text instead of inventing plausible-sounding evidence.
Quality check
- Did Claude pull facts from the document, or fabricate plausible ones? Spot-check three quotes against the source.
- Does the output read like a tired template, or like someone read the input carefully? Templated output usually means a templated prompt.
- For code, run it. For data analysis, recompute one number by hand. For prose, ask: would you send it as-is, or rewrite?
- On very long uploads, watch the context indicator. Even with a 1M-token window, recall on the middle of a huge document drifts unless you pin the question to a section (Claude long-context unstable).
Turn the win into a repeatable workflow
- Convert the winning prompt into a Project custom instruction so you never retype it.
- Keep 2-3 reference files in the Project — a style guide, a glossary, past examples — so output stays consistent run to run.
- Log “Claude wins” and “Claude loses” for two weeks, then decide on data, not first impressions.
- Re-check every few months. Default models, usage caps, and connector behavior shift quickly; this guide was last verified June 2026.
FAQ
- Should I switch from ChatGPT, or add Claude?: Usually add, not switch. ChatGPT is faster on quick lookups and image generation; Claude is stronger on long documents and careful writing. Running both at $20/mo each is common for heavy users.
- What does the free tier limit?: A rolling message cap that resets roughly every 5 hours, smaller uploads, and a limited Claude Sonnet 4.6 rather than Opus 4.7. Enough to evaluate, not enough for daily heavy use.
- Is my data used for training?: As of June 2026, yes by default. Since Anthropic’s August 2025 terms update, consumer chats may be used to train future models unless you turn the model-training toggle off in Settings → Privacy. Opting in also extends retention from 30 days to 5 years.
- How big is Claude’s context window?: 1M tokens on Opus 4.7 and Sonnet 4.6 — about a 500-page book — at standard pricing, no premium flag. That’s the same figure as Gemini 3.1 Pro and far above ChatGPT Plus’s in-app window.
- Why does Claude refuse some tasks ChatGPT does?: Claude’s safety tuning is stricter on certain categories. Rephrase with clear intent and audience, and state why you need the output.
- How do I share a Claude conversation?: Use the share button to generate a read-only public link. Never share chats containing anything sensitive.
Common beginner mistakes
- Treating Claude like a quick-answer chatbot. It shines on long inputs, multi-step reasoning, and careful writing, not three-word prompts.
- Skipping Projects for recurring work. You re-paste the same context every chat and lose the consistency benefit Projects exist to provide.
- Leaving the training toggle on by default without realizing your chats now feed model training for up to 5 years.
- Hammering retry on a “Claude is at capacity” / rate-limit message. That escalates cooldowns instead of clearing them (Claude retry rate limit).
- Loading a huge doc and expecting evenly precise recall. Even at 1M tokens, mid-document recall drifts unless you anchor the question to a section (Claude long-context unstable).
- Reading a tool refusal as the model lying. When Claude says it can’t run web search, code execution, or computer use, the tool is usually just not enabled (Claude tool unavailable).
Related
Tags: #Claude #Tutorial #Getting started