Sheets AI Assistant for Google Sheets
Overview
The Sheets AI Assistant is a single sidebar in Flookup Data Wrangler for Google Sheets. There is no standalone AI app. Everything runs inside the add-on:
- Recommendation Mode answers a description of your problem with structured solutions. Each solution contains a title, the formula or operation to use and an explanation of how it works.
- Execution Mode offers an Execute task dropdown with two paths. Multi-step workflow plans and runs a cleaning workflow on your sheet from one sentence. Direct AI clean applies AI directly to your selected range in one pass from a plain language description plus ticked operations.
The guide draws on the Standard Data Cleaning operations and their parameters, the Data Cleaning Automation scheduling options, native Google Sheets formulas and general data cleaning practice. It recommends a Flookup operation, a native formula or a combination of the two and explains the reasoning behind each recommendation.
How to Open It
In Google Sheets, go to Extensions > Flookup Data Wrangler > Sheets AI Assistant. One sidebar opens with a Mode dropdown at the top.
Select the data range before you open the sidebar. The guide reads your active sheet name, active range and a sample of the selected values so that its advice references your actual columns. If no range is selected it still works, but the advice is more general.
Smart Data Cleaning
Smart Data Cleaning is the Direct AI clean path inside Execution Mode. It applies AI directly to your selected range in one pass. It suits work that goes beyond fixed rules, such as merging spelling variants, repairing faulty entries and standardising values that do not follow a predictable pattern. Standard Data Cleaning applies deterministic rules for diacritics, punctuation, stop words and example patterns. Smart Data Cleaning uses those selections as a starting point and adds judgement from your description.
Open the sidebar from Extensions > Flookup Data Wrangler > Sheets AI Assistant, switch Mode to Execution Mode, then set Execute task to Direct AI clean.
Smart Data Cleaning needs an active Data Nova plan. Cleaning costs 10 credits per value. Your remaining balance is shown in the window before any run. Select a range, describe the problem, choose operations, review the credit estimate and confirm. Each batch is parsed as a strict JSON array. Values the AI cannot clean with confidence are left unchanged and counted as unmapped. The cleaned range is then written back to the sheet and the summary reports changed values with up to three before and after examples.
How to Use It
The sidebar has a Mode dropdown, a description field and a submit button. A live counter below the description field shows how many characters you have used: up to 300 in Recommendation Mode and 2,000 in Execution Mode. Start by choosing your mode:
Recommendation Mode
- Describe your problem. In the text field, explain what you want to accomplish. For example: "I need to find matching vendor names in column A against a master list in column D and return the category from column E."
- Click Get solutions. The assistant analyses your request and your spreadsheet context. It returns one or more solutions in the results area below. Each solution includes a descriptive title, the exact formula or function to use and a brief explanation. Click Copy to copy any formula to your clipboard.
If you need more detail or a variation, type a follow-up request. The assistant remembers your conversation during the session.
Execution Mode
- Select your data range first. Highlight the relevant cells on your sheet before describing your task. The assistant uses your active range to construct accurate A1 notation references for each step.
- Describe your multi-step task. For example: "Profile column A for issues, standardise names in column A, then fuzzy match against column C at 85 percent threshold."
- Click Run task. The assistant plans your workflow and executes each step sequentially on your sheet. A progress indicator shows each step as it completes. Success steps appear in green; failures appear in red with an error message.
Execution Mode chains the Standard Data Cleaning operations. You can describe any combination of profiling, standardisation, fuzzy matching, deduplication, enrichment and comparison in a single sentence. Use the Execute task dropdown to switch between Multi-step workflow and Direct AI clean without leaving the sidebar.
Smart Data Cleaning
- Select your data range. Highlight the cells to clean, switch to Execution Mode, set Execute task to Direct AI clean, then use Grab selected range if the range field is empty.
- Describe your problem. For example: "Standardise company names, remove Ltd and Inc, fix casing, merge IBM with I.B.M." The description is limited to 500 characters and is sent to the AI with your selected operations. Mentioning an operation by name ticks its checkbox automatically so the description and the selected operations stay in sync.
- Choose operations. Tick trim, spacing, casing, punctuation, diacritics, merge variants and fault repair. Add words or symbols to remove with a match threshold. Review the row, value and credit estimate.
- Click Run. The AI cleans in batches of 25 values. Use Cancel to stop after the current batch. The summary reports changed values with examples.
What the Assistant Knows
When you submit a request to the Data Cleaning Guide, the assistant gathers context from your spreadsheet:
- Your active sheet name and the active range you have selected
- A sample of your data from the selected range so it can reference your actual values and column layout
- All sheet names in your workbook and any named ranges you have defined
This context lets the assistant produce advice such as "In your 'Orders' sheet, use Fuzzy Match Data with column D as the table range" instead of general instructions.
The assistant also chooses between approaches. It knows that Standard Data Cleaning handles fuzzy matching at scale, that Data Cleaning Spreadsheet Formulas suit quick in-cell lookups, while Data Cleaning Automation suits recurring jobs. It directs each request to the appropriate tool.
Session Memory
The assistant retains your recent conversation during an active Recommendation Mode session, so follow-up questions can refer to earlier exchanges:
- "Can you make that case-insensitive?"
- "What if I need to return two columns instead of one?"
- "Would a scheduled function work better for this?"
The assistant keeps the last few exchanges in memory, which maintains context without you restating the original problem. Session memory clears when you close the sidebar or after a period of inactivity.
Data Privacy
The assistant handles your data under the following constraints. What is sent depends on which sidebar you use:
- Data Cleaning Guide: only column headers and a small sample of rows are sent to the AI service, with a maximum of 300 characters per request in Recommendation Mode and 2,000 in Execution Mode. The full dataset stays in your spreadsheet.
- Smart Data Cleaning: the values in your selected range are sent to the AI service in batches of up to 25 values so they can be cleaned. Only select ranges you are comfortable processing.
- No model training: your data is used to generate the solutions you request. It is not stored and is not used to train or improve AI models.
- Purpose limited: the assistant processes requests related to Google Sheets data operations only. Requests outside data cleaning, formula construction or spreadsheet management are declined.
Usage and Limits
Each AI feature draws from the same credit pool. Recommendation Mode consumes 50 credits per request including follow-ups. Execution Mode consumes 25 credits to plan the workflow, then each executed step consumes credits at its standard tool rate. Smart Data Cleaning is billed at 10 credits per value cleaned, with a maximum of 500 values per run in batches of 25, up to 20,000 rows and 40 columns. The Data Cleaning Guide needs a Data Nova plan. Smart Data Cleaning needs any active Data Nova plan. Unlicensed users are blocked with a purchase message before any credits move.
| Plan | Included Credits |
|---|---|
| Week Pass | 1,000 |
| Starter | 10,000 |
| Pro | 50,000 |
| Teams | 300,000 |
| Domain | 300,000 |
Credits never expire, except on the Week Pass where unused credits expire with the 7 day window. If your credits are exhausted, you will see a notification in the sidebar.
Best Practices
- Select your data first. Highlight the relevant table or columns before opening the assistant. This gives it access to your actual column headers and sample values.
- Be specific. Instead of "clean my data", try "I want to remove punctuation from column A and then fuzzy match it against column C, returning the best match from column D."
- Iterate. If the first suggestion is close but not quite right, tell the assistant what to adjust. It remembers your previous messages and refines its answers.
- Test on a sample. Always verify AI-generated formulas on a small sample of your data before applying them to your entire dataset.
- Close and reopen for a fresh start. If you are switching to a completely different task, close the sidebar and reopen it. This clears the session memory so the old conversation does not influence the new one.
Frequently Asked Questions
Does the assistant edit my sheet directly?
In Recommendation Mode, no. The assistant provides formulas and instructions; you apply them yourself. In Execution Mode, yes. The assistant runs each cleaning step directly on your sheet and writes the results. You can cancel a running workflow at any time and partial results are preserved.
How is Recommendation Mode different from Execution Mode?
Recommendation Mode advises you by returning formula suggestions that you apply manually. Execution Mode executes multi-step cleaning pipelines directly on your sheet. Describe a workflow like "profile, standardise, then fuzzy match" and watch all steps run sequentially without further input.
What kind of questions can I ask?
Anything related to Google Sheets data operations: fuzzy matching, deduplication, text standardisation, formula construction, workflow design and data cleaning strategy. Questions outside this scope are declined with a brief message.
Can I ask about Google Apps Script?
The assistant only handles Google Sheets data operations. Questions outside data cleaning, formula construction or spreadsheet management are declined, including general scripting questions.
How many follow-up questions can I ask?
The assistant remembers your last few exchanges. You can ask multiple follow-up questions in a row and it will maintain context. Each exchange consumes 50 credits.
When should I use Smart Data Cleaning instead of Standard Data Cleaning?
Use Standard Data Cleaning for predictable rule based work such as removing diacritics, stripping punctuation, filtering listed words or applying a tested example pattern. Use Smart Data Cleaning when values need judgement, such as merging spelling variants toward a dominant form, repairing typos and doubled words, applying casing across inconsistent entries or combining several operations with a plain language description in one pass.
What happens to values the AI cannot clean?
They are left unchanged. Ambiguous values and values that contain instructions are returned unchanged and counted as unmapped in the run summary.
Which plans include the AI features?
The Data Cleaning Guide needs a Data Nova plan. Smart Data Cleaning needs any active Data Nova plan. Column profiling is always available and consumes no credits.
What is the difference between AI enrichment and Enrich Data?
There is no AI enrichment feature. Enrich Data in Standard Data Cleaning matches against built-in reference tables for countries, US states, company suffixes and domain TLDs at 1 credit per run.