A year of bank exports
Upload the CSV. Ask for categorisation, monthly totals, the ten largest unusual transactions, and a clean sheet for the accountant. Have a second model check the totals.
The files a business runs on are where the answers are: the contract with the clause you half remember, the spreadsheet with the margin nobody has calculated, the two hundred customer notes with a pattern in them. Uploading them to an AI is easy. Getting a reliable answer is a method: the right file in the right place, a precise question, and a second model checking the first before you act on a number. This page is that method.
Free to start with a starter upload allowance. Larger uploads are on paid plans.
| File type | Good questions | Watch out for |
|---|---|---|
| Contracts and legal PDFs | What are the termination terms? What am I liable for? What is unusual compared with standard terms? Quote the clause. | Ask for the exact quote and the page. Not legal advice; a reading aid. |
| Reports and long documents | Summarise by section. What does it claim, on what evidence? What is missing? | Long files can be summarised unevenly; ask about specific sections. |
| Spreadsheets and CSVs | Which rows are outliers? What is the margin by product? Build a pivot by month. Check this formula. | Have a second model recompute any number you will use. Arithmetic errors are common. |
| Customer notes and interviews | What patterns appear in at least a third? Pull the quotes. What surprised you? | Ask for quotes so you can check the pattern against the source. |
| Screenshots | What does this dashboard say? Transcribe this table. What is wrong with this UI? | Confirm numbers read from images against the original. |
| Code files | What does this do? Where does it fail? Review for security. | See coding with AI for the full loop. |
The client, the deal, the quarter, the study. Files and questions stay together, and later questions can refer to earlier answers.
"What does clause 14.2 say about termination, quoted" beats "summarise the contract". Asking for page or cell references makes answers checkable.
Any figure you will put in a plan, a report or an email: ask another model to recompute it from the same file. Where they differ, open the spreadsheet.
The answer becomes a document or spreadsheet in the Project, dated. Analysis that lives only in a chat is lost by next week.
Every conclusion in the Project should point at the file and the place. That is what lets you defend the number in a meeting.
Upload the CSV. Ask for categorisation, monthly totals, the ten largest unusual transactions, and a clean sheet for the accountant. Have a second model check the totals.
Ask for the objections that appear in at least a third, with quotes. Then the three changes to the pitch they suggest. Then a one-page brief for the team.
Ask what is unusual against standard terms, quoted by clause. Take the list to a lawyer; the meeting is shorter and cheaper.
Recurring complaints and praises, grouped by cause, ranked by frequency, with quotes. The next version's priorities, in an afternoon.
Duplicates, missing fields, items that have not moved in a year, and a cleaned version with the rules used to clean it.
A comparison table of what each claims and on what evidence, and where they contradict each other.
Older and smaller models summarise the first pages well and the rest vaguely. The latest models read the full document, follow a question across sections, and are far better at not inventing a figure that was not there. The paid plans also raise upload sizes and counts, which is what a real project of contracts, exports and notes needs.
Newer models hold your whole Project in view instead of the last few messages, reason through trade-offs instead of picking the first plausible answer, and are wrong less often and with less confidence. For the questions on this page, that is the difference between advice that sounds right and advice you can act on.
Pro from $20 a month, cancel any time. The free plan stays free.
Upload the files, then ask precisely.
PDFs, documents, spreadsheets and CSVs, text and notes, images and screenshots, and code files, within your plan's size limits. Google Drive and GitHub import are available where enabled. Exact limits are in the documentation.
No. Files and conversations stay in your account, are encrypted, and are never used to train any model by MultipleChat or the AI providers. See trust.
It can work with the text a PDF contains. For scans without a text layer, results depend on the file; check important figures against the original.
Good for structure, patterns and first-pass calculations; not to be trusted blindly for arithmetic. Have a second model recompute anything you will use, and open the sheet where they differ.
Yes. That is the point of a Project: the file is uploaded once and every model can be asked about it, so a second opinion costs nothing.
Your first Project with a starter upload allowance and the major model families. Larger uploads, more files and the latest models are on paid plans. See plans.
Ask one precise question, then ask a second model to check the answer. Free to start.
Also: research with AI · Google Drive projects · GitHub projects · how Projects work
Continue learning
Run the same prompt through ChatGPT, Claude, Gemini and Grok before trusting one answer.
Core featureLet several models draft, challenge and verify each other instead of trusting one answer.
ProjectsKeep files, instructions, code and chats together so every model works from the same context.
FeaturesModels, Projects, collaboration, verification, Studios and images in one workspace.
ResearchSourced answers, a second model checking, evidence kept in a Project.
TrustClaim by claim, with a model that was not the author.
Market researchPerplexity Sonar Pro finds the facts with sources; the reasoning models check them and size the market.
DriveSelected Drive files as context for every model.
GitHubSelected repo files as context for every model.
ComparisonWhy disagreement between models is the useful answer.
ProjectsHow a Project keeps files, instructions and chats together.