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 the first three rows of the selected values so that its advice references your actual columns.
If no range is selected, the assistant can still respond, 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 1-Month or 1-Year Pass plan or the 7-day AI trial, plus your own Google Gemini or OpenAI key. There are no credits. Select a range, describe the problem, choose operations and confirm.
For each cell, Flookup sends the first 300 characters to the AI service. Any characters after this limit are preserved without cleaning. Each batch is parsed as a strict JSON array. The cleaned range is written to the sheet only after every batch has completed successfully, and the summary reports changed values with up to three before and after examples. An unmapped value is a non-empty input for which the service did not return a usable matching result. It remains unchanged and is counted in the summary.
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 can plan and run these Standard Data Cleaning tools: Profile Column, Standardize, Fuzzy Match, Merge Row, Remove Duplicates, Find Duplicates, Extract Unique, Enrich Data, Fuzzy Conditional Formatting and Compare.
Find Duplicates only reports duplicate groups with a confidence score and never changes your data, while Remove Duplicates writes the deduplicated result back to your sheet. The planner selects the tools and parameters for a multi-step task. 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 and value summary.
- 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 a limited context from your spreadsheet:
- Your active sheet name and the active range you have selected
- The first three rows from the selected range so the assistant can reference sample values and column layout
- Up to 15 sheet names and 25 named ranges from your workbook
If the workbook exceeds these metadata limits, the assistant uses the capped context and reports that partial context was used. It does not send every sheet name or named range. 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.
- Temporary conversation history: Flookup stores the current conversation in the user's Apps Script cache for up to 30 minutes to support the conversation. Submitted data is handled by the AI service under its own policy.
- 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
AI runs on your own Google Gemini or OpenAI key, with the Flookup skills behind it running on our side at no extra cost, and Flookup meters nothing. What a run costs depends on the tokens it uses, and your provider bills you directly. Smart Data Cleaning processes up to 500 values per run in batches of 25, across ranges of up to 20,000 rows and 40 columns. The Data Cleaning Guide needs a 1-Month or 1-Year Pass plan. The Sheets AI Assistant and Smart Data Cleaning need an active 1-Month or 1-Year Pass plan or the 7-day AI trial. Users without access are blocked with a purchase message before anything runs.
| Plan | AI Access With Your Own Key | Trial |
|---|---|---|
| 1-Month or 1-Year Pass | Full AI features | Not needed |
| Free tier | 7-day AI trial, then pass required | 7 days |
There are no credits to run low on. If your provider key hits its own limit, only the AI features pause while cleaning and deduplication keep working.
If the assistant reaches its daily token limit, it cannot process another request until the limit resets. If the AI service is unavailable, rejects a request or returns an unreadable response, the assistant reports the error.
Smart Data Cleaning does not write the selected range when a batch fails.
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.
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. A value is counted as unmapped when the service does not return a usable matching result for that input. The run summary reports the number of unmapped values.
Which plans include the AI features?
The Data Cleaning Guide needs a 1-Month or 1-Year Pass plan. The Sheets AI Assistant and Smart Data Cleaning need an active 1-Month or 1-Year Pass plan or the 7-day AI trial, plus your own key. There are no credits. Column profiling is always available.
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 with no credits involved.