How to Use ChatGPT with Files, Web Search and Projects: A Practical Guide

Quick answer

• Use files when the answer depends on existing material.
• Use web search when the information may have changed.
• Use projects when several chats depend on the same sources and instructions.
• Use Chat for questions, ideas, and quick revisions.
• Use Work for larger outputs and reviewable files, when available.
• Use a separate conversation for each independent deliverable.

How to work with files

ChatGPT can work with documents, PDFs, presentations, spreadsheets, images, and other supported formats. A file does not explain the intended task by itself. When attaching one, state:

1. which file is the main source;
2. what should be done with it;
3. the required output format;
4. what must not be changed;
5. how the result should be checked.

For example:

Compare these two articles. Show factual changes, removed sections, and new claims. Do not rewrite the text. At the end, identify which changes require a source.

Specific instructions reduce the risk of receiving an attractive result that solves the wrong problem.

Summarise documents without losing essential context

A summary can remove details that matter. Define the reader and decision:

Summarise this report for someone deciding whether to buy the product. Preserve prices, limitations, and dates. Connect each conclusion to the relevant page and identify any unreadable content.

For important documents:

• confirm the page count;
• inspect tables, charts, footnotes, and appendices;
• request page references;
• compare the summary with the original;
• do not assume that every image or scanned page was interpreted correctly.

A short summary is not automatically a faithful one. Tell the assistant which information must survive compression.

Compare files and versions

When reviewing two versions, define what counts as a meaningful difference:

• editorial content;
• figures and dates;
• code and behaviour;
• structure and navigation;
• images and links;
• SEO and metadata.

Example:

Compare the current version with the previous one. Ignore spaces and formatting changes without practical effect. Show only differences that change meaning, functionality, or SEO.

Preserve an approved copy before replacing a file. Version history is especially important when AI is editing several sections at once.

Create files that are ready for review

When requesting a document, presentation, spreadsheet, or other artifact, provide:

• the source material;
• required file type;
• intended structure;
• target audience;
• visual identity;
• quality and verification criteria.

Open the result before approving it. Look for clipped text, empty pages, missing images, broken formulas, unreadable tables, incorrect page order, and inconsistent typography. A technically created file is not necessarily a finished file.

When to use web search

Search is appropriate when the answer depends on information that can change, including:

• news and recent events;
• prices and availability;
• event dates and schedules;
• software versions and product specifications;
• laws, regulations, and public rules;
• current executives or company information;
• recommendations affected by the present market.

A useful request is:

Search only current, authoritative sources. State the date of each important fact, separate facts from inferences, and place citations next to the claims they support.

If the question can be answered from a document you already possess, use that primary file before searching for secondary summaries.

How to verify research

Treat web content as untrusted until evaluated. Check:

• who published the source;
• when the page and underlying event were published;
• whether the source directly supports the claim;
• whether an official or primary source exists;
• whether reliable sources disagree;
• whether later information changes the conclusion.

For medical, legal, or financial decisions, request official evidence and obtain qualified professional advice where appropriate. AI can organise information, but it cannot absorb the responsibility of a regulated professional.

Quick search and deep research

A normal search is suitable for a narrow question with a small number of sources. Deep research is more appropriate when the task requires many sources, comparison, a structured report, and explicit analysis of uncertainty.

Before beginning a larger investigation, define:

• the central question;
• relevant time period;
• country or market;
• preferred primary sources;
• excluded sources or topics;
• final format;
• conditions for reaching a conclusion.

This prevents a large quantity of material from being mistaken for a focused answer.

How to use projects

A project brings together related chats, files, instructions, and sources. Create one when the work:

• continues over time;
• produces several separate deliverables;
• depends on recurring rules;
• needs a consistent identity or editorial voice;
• repeatedly uses the same source files.

Examples include a website, training programme, long-term investigation, product launch, or editorial series.

A project does not mean that every task belongs in one enormous conversation. Separate chats make individual outcomes easier to understand and review.

Organise conversations inside a project

A practical structure could be:

• Topic research
• Article outline and draft
• Factual review and sources
• Code and publication
• Metrics and later updates

Use clear chat titles and archive completed work when appropriate. Keep permanent decisions in project instructions or a dedicated editorial-reference file rather than relying only on conversational memory.

Chat, Work, or Codex?

Availability can vary by account and product version. Choose based on the result you need. You can keep a natural conversation in each mode, but the tools and preferred workflow differ.

A complete workflow for creating an article

1. Open the editorial project.
2. Create a chat to define the article angle.
3. Research current primary sources.
4. Save or attach the central evidence.
5. Use a separate chat to produce the draft.
6. Add personal experience or original editorial judgement.
7. Perform factual and language reviews.
8. Create the final version as a file.
9. Open and inspect the file.
10. Publish only after approval.

This division preserves continuity without mixing research, drafting, technical implementation, and publication into one unclear request.

Give precise revision instructions

Avoid saying only “improve this.” Explain what is wrong and what must remain intact. Useful instructions include:

• “Preserve the structure and reduce repetition.”
• “Do not alter figures or quotations.”
• “Present the diagnosis before changing anything.”
• “Show the differences before replacing the file.”
• “Mark unsupported information.”
• “Keep everything that has already been approved.”

When a change could remove content or overwrite a working version, request a draft or comparison first.

Privacy and permissions

Before attaching a file or connecting an external source:

• remove passwords, private keys, and access tokens;
• hide unnecessary personal information;
• confirm who can access the source and result;
• review the permissions of connected applications;
• do not upload another person's private information without authority;
• request a draft rather than automatic external delivery;
• require confirmation before consequential external changes.

The fact that a tool can access or modify something does not mean every request should authorise it to do so.

Common mistakes

• Uploading files without describing the task.
• Searching the web when the original source is already available.
• Trusting citations without opening the pages.
• Mixing independent projects in one workflow.
• Allowing two processes to write to the same file simultaneously.
• Replacing a version without preserving a copy or history.
• Requesting execution when only analysis was intended.
• Failing to inspect the final artifact.

Ten useful requests

1. “Summarise this file and identify the pages used.”
2. “Compare these versions and ignore cosmetic differences.”
3. “Analyse this spreadsheet and identify duplicates.”
4. “Search official sources published or updated during the past 12 months.”
5. “Separate confirmation, third-party statements, and inference.”
6. “Create an outline before writing the draft.”
7. “Preserve approved decisions in the project instructions.”
8. “Do not modify the file; present a diagnosis first.”
9. “Create the result in an editable format.”
10. “Open and verify the file before delivering it.”

Final checklist

• Was the correct file used?
• Were the task and output format clear?
• Were current sources opened and checked?
• Are facts and inferences separated?
• Does the chat belong to the correct project?
• Is an earlier version preserved?
• Are access permissions appropriate?
• Was the final result visually inspected?

Maintain continuity without creating confusion

A long project benefits from a short reference document containing its objectives, approved decisions, tone, central sources, and next actions. Update that document whenever a decision changes instead of relying exclusively on old messages.

At the beginning of a task, state the desired result and point to the relevant files. At the end, record what was completed, what remains pending, and which versions were created. This makes it easier to resume the work days later, move it to another conversation, or understand why a decision was made.

Continuity does not mean preserving every message inside one chat. It means preserving the right information in a clear and reusable place.

Conclusion

Files provide evidence, web research provides freshness, and projects provide continuity. When all three are used with clear objectives, ChatGPT becomes more than a question-and-answer interface: it supports an organised working process.

The central principle remains simple: provide the right context, limit the action, verify the sources, and inspect the result.

NTS View

Productivity with AI does not depend only on the model. It depends on the architecture of the work: reliable sources, focused conversations, preserved versions, appropriate permissions, and clear completion criteria.

NTS Verification

This guide reflects official OpenAI documentation available during preparation. Features vary by application, plan, workspace, and product version; verify current availability before publication.