Learn ChatGPT the way a researcher actually uses it — not trivia, but a dependable workflow. You'll go from your first real prompt to uploading papers and data files, getting web answers you can verify, generating and reading images, and organizing recurring work with Projects, custom instructions, and custom GPTs. Eight short lessons end in a capstone that takes one real question from background reading to an analyzed result, with a human verification step built into every stage.
Get started
Work like a pro
Go deep
ChatGPT Small biz AI-native SMEConfigure a custom GPT once — its instructions, tone, and any reference files — so a recurring job (answering FAQs, formatting reports) runs the same way every time without re-explaining.
A saved, shareable assistant your whole team opens instead of re-writing the same prompt.
Create a GPT with instructions + a few example documents, then test it on a fresh case.
ChatGPT Small biz FounderIn ChatGPT settings → Data Controls, turn off "Improve the model for everyone" before any session that involves customer names, financial figures, or contract terms. For persistent protection, use ChatGPT Team or Enterprise, which has no-retention by default.
Confidential business information shared in the session is not used for future model training, reducing data-leakage risk with clients or partners.
Settings (bottom-left avatar) → Data Controls → toggle off "Improve the model for everyone." Verify the toggle persists across browser sessions before sharing sensitive data.
ChatGPT Founder Small biz AI-native SMEGive ChatGPT the bullet points and have it draft a clear email, proposal, or summary — then refine tone in follow-ups.
A solid first draft in seconds, then tightened to your voice.
"Turn these bullets into a friendly but professional client update, ~150 words."
ChatGPT Founder CreatorPaste your messy notes and have ChatGPT structure them into a clear slide-by-slide outline with a title, key points per slide, and a closing.
A presentable skeleton in seconds that you refine, instead of staring at a blank deck.
"Turn these notes into a 10-slide outline: title, 3 bullets per slide, and a one-line takeaway each."
ChatGPT ScientistPaste a broken Python/R snippet into ChatGPT, describe the error, and get a corrected version with an explanation of what went wrong.
A working script and an explanation you can learn from — no install needed.
Paste your code + the error message → "Fix this and explain the bug in one paragraph."
ChatGPT Scientist AI-native SMEUpload a CSV or Excel file and ask ChatGPT to clean it, run the analysis, and plot the result — it writes and runs the code for you and hands back the chart.
A computed answer plus a figure from your own data, no Python set up on your machine.
Upload your file → "Clean this, run a correlation between the numeric columns, and plot the strongest one as a scatter with a trend line."
ChatGPT Scientist AI-native SMEWhen ChatGPT provides a literature citation, paste the exact title and author list into Google Scholar, PubMed, or Semantic Scholar before including it in any document. Ask ChatGPT to provide the DOI and then resolve the DOI yourself.
Fabricated references — a well-documented failure mode of all large language models — are caught before they reach a report, grant application, or manuscript.
Follow-up prompt: "Give me the DOI for each paper you cited." Then open doi.org/<DOI> directly. If the DOI does not resolve, the citation is fabricated.
ChatGPT ScientistOpting out of model training is not the same as zero data retention — OpenAI's API and consumer tiers still transmit data to their servers and retain it for abuse-monitoring purposes. Anonymise or aggregate data before asking for analysis; never paste names, dates of birth, or medical record numbers.
You stay compliant with data-protection regulations (GDPR, HIPAA-equivalent obligations) that apply regardless of the provider's training opt-out.
Before sharing any dataset, run a quick de-identification step: replace names with "Participant A/B/C", dates with relative offsets (Day 0, Day 7), and IDs with sequential integers.
ChatGPT Scientist Creator AI-native SMEOpen every research-focused session with a system-level instruction: "If you are uncertain about any fact, say 'I'm not sure' and tell me what you are basing the answer on. Never produce a confident-sounding answer when your confidence is low."
The model's confident tone — its most dangerous trait for factual work — is partially counteracted, and you are more likely to receive explicit uncertainty signals you can act on.
Paste this instruction as the first message in every new conversation used for literature or factual research tasks.