As of 13 August 2026, AI can clean your business CSV files.
This still needs a person who signs their name to it.
Can you do it?
15 minutesto a draft.
1 hourto something you’d act on.
Cost, all in£0
Skill neededchat-fluent
Who has to check ityou
What the alternative costsA purpose-built alternative is Formula Bot, which turns plain English into Excel and Sheets formulas and analyses data.
If this goes wrong: valid records can be changed or removed, and the error can spread into reports, invoices or customer records before you notice.
What to actually do
Hand it to a person
The route this page recommends
A person who owns the outcome does this end to end, worth it when the failure is dear.
Use a tool built for this
Second choiceDo it yourself
The distant thirdA chat interface, chat-fluent skill, and roughly 1 hour until you can act on the result.
How to actually do it
- Open the CSV in a spreadsheet or data tool and make a separate backup copy of the original file.
- Write down the required column names, date format, number format, duplicate rule and any permitted replacements for blanks or inconsistent values.
- Upload the backup copy to an AI tool and paste the prompt, replacing the bracketed sections with your documented business rules.
- Read the proposed transformations and answer the tool's questions about ambiguous values before allowing it to create the cleaned file.
- Save the cleaned CSV, exceptions CSV, change log, validation report and reproducible code in the same project folder.
- Compare the original and cleaned row counts, column names, totals and a sample of changed rows, then inspect every row in the exceptions CSV.
- Open the cleaned CSV in the system that will use it and check that dates, numbers, identifiers and required fields are interpreted correctly before replacing the original.
Prompt
I have uploaded a business CSV file. Create a cleaned copy and leave the original unchanged. Cleaning rules: - Keep every valid record unless a rule below says otherwise. - Detect and report blank columns, duplicate rows, duplicate business records, inconsistent column names, inconsistent date formats, inconsistent capitalisation, leading or trailing spaces, invalid email formats, and numbers stored as text. - Apply only these business rules: [PASTE YOUR EXPLICIT RULES HERE]. - Use these expected columns and formats: [LIST THE REQUIRED COLUMNS AND FORMATS]. - Do not guess what an ambiguous value means. Put each ambiguous row in a separate exceptions file and explain why it was not changed. - Do not silently delete, merge or overwrite data. Return: 1. The cleaned CSV. 2. An exceptions CSV containing every row that needs a human decision. 3. A change log showing each transformation, the number of affected rows and examples before and after. 4. A validation report showing original and cleaned row counts, column names, duplicate counts, missing values and any values that still fail the stated rules. 5. The code or reproducible steps used, so I can run the process again. Before producing the cleaned file, list the proposed transformations and ask me to confirm any rule that is not explicit.
Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.
What it gets wrong
- AI cannot know whether two similar customer or product records are genuinely duplicates without a reliable business rule.
- AI cannot decide whether an unusual value is an error, a legitimate exception or a change in trading practice.
- A plausible cleaned file can still contain silent changes, especially where dates, identifiers and numbers have mixed formats.
- AI cannot take responsibility for data loss or for reports and decisions made from an incorrect cleaned file.
- You still need to test the output in the system that consumes the CSV, because a file that looks correct in a preview can fail on import.
Even on a YES, the friction has a name: judgement under ambiguity, verification cost and stakes of error.
How we scored this
Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.
| Axis | Score (0–2) |
|---|---|
| Output | 2 |
| Inputs | 2 |
| Verification | 1 |
| Liability | 1 |
| Effort delta | 2 |
| Total | 8 / 10 |
The methodology and its thresholds are published in full.
FAQ
- Can ChatGPT clean a CSV file?
- Yes. It can profile the file, suggest transformations, write code, create a cleaned copy and produce a change log. Give it explicit rules and keep the original unchanged.
- Can AI remove duplicates from my business data?
- It can identify likely duplicates and remove exact duplicates when you define the rule. It cannot reliably decide whether similar customer, supplier or product records should be merged without your business criteria.
- Is it safe to upload a business CSV to AI?
- Only use a tool and account approved for your organisation, and remove unnecessary personal or confidential data before uploading. Check where the file is stored and processed, and do not upload it if your business policy does not permit it.
- How do I check an AI-cleaned CSV?
- Compare row counts, columns, totals, duplicate counts and missing values with the original, then inspect every changed and exception row. Finally, test the cleaned file in the system that will import it before using it for reporting or operations.
Nearby answers
Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.
The newsletter
AI news, new answers and product picks, straight to your inbox.