How to score and fix a Google Sheet™ for AI readiness

A step-by-step run through the add-on, from install to a fixed sheet. Scanning, the score, every explanation and Undo are free and unlimited. Only the fixes, whole-spreadsheet scans and the exportable report are paid, and each is marked Pro where it appears.

Prefer to watch? The same walkthrough as a video, recorded on the live add-on, with a chapter for each step. Open it on YouTube.

What the problem is

Gemini™ in Sheets, ChatGPT, Copilot and every other AI assistant read a spreadsheet by inferring its shape from the cells — there is no schema to consult. Merged cells, a title sitting above the headers, a blank row splitting one table into two, numbers typed as text: none of these confuse a person looking at the sheet, and all of them confuse a model trying to parse it into rows and columns.

This add-on reads the sheet the way an AI assistant would, scores how readable it is, and explains every finding in plain English — what is wrong, and why AI cares.

Before you start

Google Sheets™ in a desktop browser. The add-on does not run in the Google Sheets™ mobile apps.
The add-on installed from the Google Workspace™ Marketplace.
A spreadsheet with at least one sheet of data on it.
Nothing leaves Google. The rules run in the sidebar, in your browser, against the Sheets API. The add-on's only network traffic is to the licence server, and only ever a licence key and an anonymous install id. The privacy notice says so in full.

Step 1 — Install

Install it from the Google Workspace™ Marketplace, or in any spreadsheet choose Extensions › Add-ons › Get add-ons and search for “AI Readiness Scanner and Fixer”.

Step 2 — Open the sidebar

In the spreadsheet, open Extensions › AI Readiness Scanner and Fixer › Open scanner. A sidebar opens on the right.

Step 3 — Scan the sheet

Press Scan this sheet. It reads the active sheet and scores it 0–100. On Pro, Scan whole spreadsheet runs the same checks across every sheet in the file.

A sheet with nothing wrong scores 100 and says so, rather than showing an empty list.

Step 4 — Read the score

100 minus 18 per blocker, 12 per degrader, 3 per advisory. It is blunt on purpose — the number exists to be compared before and after a cleanup, not read as a grade.

Blocker — changes what an AI assistant reads: it will parse the sheet wrong.
Degrader — makes answers worse without breaking the parse outright.
Advisory — worth knowing, small effect on the score.

Step 5 — Read a finding

Every finding names its rule code, what it found, and a short “why AI cares” line.

`STR-01`	Merged cells. A merged cell has one value and several empty ones underneath it. AI reads the empty cells as blank, not as repeats of the header above.
`STR-02`	A title or banner row above the headers. AI treats row 1 as the header row. If row 1 is a title, every column name it reads is wrong.
`STR-03`	Blank rows or columns splitting the data. Most tools stop reading at the first blank row, so anything below it is silently dropped.
`STR-04`	Several tables on one sheet. Without a gap or a sheet boundary, an AI assistant reads two tables as one, with columns that do not belong together.
`STR-05`	Total or subtotal rows in the data. A totals row looks like one more record. Sums and averages then double-count it.
`STR-06`	Not converted to a Sheets table (manual check). Format › Convert to table gives a sheet a name, typed columns and a fixed range that AI tools can rely on. The add-on cannot see whether a range is already a Sheets Table without broad Drive access, so it asks you to check by hand.
`TYP-01`	Numbers stored as text. A number typed with a stray leading space, or pasted from a system that quotes everything, sits left-aligned and fails `SUM`, sort and any AI question that expects arithmetic.
`TYP-02`	Dates stored as text. A date that is really text cannot be filtered, sorted or reasoned about as a date — "last quarter" means nothing to a formula reading a string.
`TYP-03`	Mixed types in a column. A column that is mostly numbers with a few notes typed in among them forces every consumer to guess, cell by cell, what that column actually is.
`HDR-01`	Blank or duplicate headers. Two columns called "Amount" are indistinguishable by name, and a blank header forces a reference by letter, which nobody can read back later — least of all an AI assistant summarising the sheet.
`HDR-02`	Two-row or grouped headers. A header that spans two rows, or groups several columns under one merged label, has no single row that names every column, so tools that expect one header row misread the whole table.
`CLN-01`	Inconsistent text. "Smith", " Smith" and "SMITH" are three different values to anything that groups or matches by text, an AI assistant included.
`CLN-02`	Hidden rows or columns. Hidden data is still read by most AI tools even though a person skimming the sheet does not see it — which can mean an answer that quotes a row nobody meant to include, or a summary that is missing one nobody remembered was there.
`ADV-01`	Volatile formulas, external links, or a very wide sheet. Formulas like `NOW()` or `RAND()` change on every read, a link to another file can go stale, and a very wide sheet is expensive for any tool to load in full.
`ADV-02`	Cryptic sheet names. "Sheet3" or "Copy of Data (2)" tells an AI assistant nothing about what a tab contains when it is choosing where to look.
`SHT-01`	Live imports. `IMPORTRANGE`, `IMPORTDATA`, `GOOGLEFINANCE` and similar functions pull in data from elsewhere that can change or fail independently of the sheet in front of you, which an AI assistant reading a snapshot has no way to know.

Step 6 — Fix it Pro

Six findings have a fix. Each shows a preview of the exact cell changes before anything is applied.

STR-01 — unmerge and fill: splits the merged cell. A merge running down one column has its value copied into the cells below; a title or group label merged across columns keeps its value in the top-left cell only.
STR-02 — delete title rows: removes the rows above the header.
STR-03 — delete blank rows and columns: removes the gap splitting the data.
TYP-01 — convert text to numbers: turns text-formatted numbers back into numbers.
HDR-01 — name headers: fills in blank headers and disambiguates duplicates.
CLN-01 — tidy text: trims stray spaces and normalises case where that will not change meaning.

The remaining findings — STR-04, STR-05, TYP-02, TYP-03, HDR-02, CLN-02, ADV-01, ADV-02, SHT-01 — are explained but not auto-fixed: what a total row should become, or which imports to check, is a judgement call the add-on leaves to you.

Step 7 — Undo, or keep the changes

Every fix first saves a hidden copy of the sheet before it changes anything. Undo in the sidebar restores that copy — free in every tier, always. Keep changes deletes the hidden copies once you are happy with the result.

Step 8 — Whole-spreadsheet scan and report Pro

Scan whole spreadsheet runs every check across every sheet in the file, not just the active one. Export report writes the full finding list to its own tab in the spreadsheet, so it travels with the file and stays readable without the add-on open.

Step 9 — Entering a licence key

Expand Enter a licence key at the bottom of the sidebar, paste the key from the end of checkout, and press Activate. It covers 3 Google accounts, and takes effect immediately.

Limits

Desktop browser only. The add-on does not run in the Google Sheets™ mobile apps.
Sheet size. Sheets up to 2,000,000 cells are supported. Larger sheets are declined with a message rather than scored wrongly.
Hidden rows. The hidden-row check (CLN-02) covers the first 5,000 rows of the sheet.
Undo after a row-deleting fix (such as STR-02 or STR-03) restores row heights and hidden rows for the first 5,000 rows.

FAQ

Does my data leave Google? No. The rules run in the sidebar, in your browser, against the Sheets API. The only network call this add-on makes is to the licence server, for a licence key and an anonymous install id.

Does it work on a .xlsx file opened in Office editing mode? No — convert the file to a Google Sheet™ first (File › Save as Google Sheets™). The add-on reads the Sheets API, which an Excel file kept in Office mode does not expose.

Is there a version for Excel? Yes — the Copilot Readiness Scanner does the same job for Excel and Microsoft Copilot.

If something looks wrong

Email support@appgarden.co.uk, and say which finding looked wrong. A spreadsheet that produced a wrong finding — or missed a real one — is the most useful thing you can send. Only send one you are content to share, and read the privacy notice first.