Skip to main content
Reference Guide beginner

Digitize Your Recipe Book: AI Import from Excel, CSV and Notes

Move years of recipes out of spreadsheets and notebooks in one pass — AI extraction structures them, semantic matching resolves every ingredient, and your whole library gains computed water activity, shelf life, and nutrition.

Yauheni Padniuk 6 min read Updated July 2, 2026
A closed leather recipe notebook and pen beside a copper bowl and ingredients.

The Real Barrier Is Migration, Not Software

Every pastry chef already owns recipe software’s real competitor: a battle-worn Excel costing sheet, a notes app full of half-structured formulas, a binder from three kitchens ago. The recipe manager you evaluated last year lost not because it was worse than the spreadsheet — but because retyping a hundred recipes into a hundred forms was never going to happen between production days.

Formul.io removes that step. Its AI bulk import reads the files you already have — no template, no cleanup pass, no fixed column order — and turns them into structured, calculable recipes. The trade it is designed around: you spend judgment, not data entry.

What the importer accepts

Excel (.xlsx, .xlsm), CSV and TSV exports, plain text, Markdown, and PDF — up to 25 MB per file. If it opens in a spreadsheet or a text editor, it imports. A number-locale switch handles decimal commas, so European-formatted quantities like “1,5 kg” parse correctly.

From Messy Rows to Structured Recipes

One upload starts a planning job that reads the file the way a person would — then does the tedious part at machine speed. Only the extraction stage is generative; matching, enrichment, and plan assembly are retrieval and arithmetic, which is why the result can be inspected line by line and replayed exactly.

1

Upload one file

Send the spreadsheet, export, or pasted notes as-is. The AI identifies recipe boundaries, ingredient lines, and quantities in unstructured content — a costing sheet, a supplier layout, or free text all work.

2

AI extraction

A language model turns every source line into structured data: recipe name, ingredient, quantity, unit. This is the only step that uses AI — everything after it is deterministic.

3

Semantic ingredient matching

Each extracted name is matched against the ingredient catalogue by meaning, not spelling. A supplier's shorthand and a generic name resolve to the same ingredient, and matching works across languages.

4

Nutrition enrichment

Names with no catalogue match are looked up against USDA and CIQUAL reference foods. On commit they become ingredients with full, sourced nutrition data instead of an estimate.

5

A frozen plan

The result is a deterministic import plan you can inspect line by line. Nothing has been written to your library yet.

The plan is the contract. The AI runs once, during planning; the commit step later replays that frozen plan deterministically. What you approve in the preview is exactly what lands — the model cannot quietly change its mind between review and commit.

Matching is deliberately conservative. Exact and high-confidence semantic matches resolve automatically; anything borderline — a house abbreviation, a brand name, a typo — becomes a flagged item for your decision, never a silent guess.

The Preview Is the Quality Guarantee

Bulk import earns trust by refusing to be silent. Before anything is written, the dry-run preview shows a reconciliation report in which every source line is accounted for — matched, flagged, or skipped with a stated reason — plus computed metrics for each recipe in the plan.

Preview elementWhat it tells youWhat you do
Matched lineThe ingredient resolved to a catalogue entry with high confidenceSpot-check; usually nothing
Flagged itemA name is ambiguous — several plausible matches, or noneOne decision: map it to an existing ingredient or create a new one
Reference matchNo catalogue hit, but a USDA/CIQUAL reference food fitsAccept the sourced nutrition or override it
Per-recipe metricsWater activity, shelf life, and nutrition computed from the planSanity-check the numbers before committing

What the dry-run preview shows before anything is committed

The flagged queue is where the afternoon is saved: you review the short list of genuinely ambiguous ingredients, not every row of every recipe.

The per-recipe metrics deserve a second look before you commit. A ganache that previews at a suspiciously high water activity usually means a quantity landed in the wrong unit somewhere in the source file — caught in the preview, that is a one-line fix instead of a corrupted library.

Fix once, it applies everywhere

Resolving a flagged item edits the plan, not one row. Map “Trimoline” to invert sugar once, and every row using that name — across every recipe in the file — updates in the re-run preview. Choose to remember the mapping and your next import resolves it automatically.

Committing is just as deliberate. Nothing lands until you commit the previewed plan, and the commit is idempotent: re-importing the same file does not create duplicates, so an interrupted migration can simply be run again.

Four Migration Scenarios

ScenarioYou bringWhat happens
The Excel costing sheetA workbook of a hundred rows accumulated over yearsOne upload and one pass over the flagged list — a library migration in an afternoon instead of a retyping project
Free-text recipe notesRecipes pasted from a notes app or old messagesThe AI structures names, quantities, and units itself — no reformatting into a template first
The quality-first reviewA file you don't fully trustThe reconciliation report accounts for every line; anything unparseable is flagged with a reason, never dropped silently
The legacy libraryRecipes that never had numbers attachedEvery committed recipe gains computed water activity, shelf life, and nutrition the moment it lands

The speed claim needs no magic. Extraction, matching, and enrichment are machine time; your time goes only where judgment is required — the flagged matches. For a typical costing sheet that is a handful of supplier-specific names, not a hundred forms.

Free-text import deserves its own mention because it clears the guilt pile: the recipes that never made it into any spreadsheet. Paste them as they are — quantities buried in sentences, ingredients in shorthand — and they go through the same extraction, matching, and preview as everything else.

After Import, Your Library Has Physics

A digitized recipe in a generic notes tool is still just text. A committed Formul.io recipe is a live formulation: the calculation engine computes water activity, shelf life, composition, and nutrition from real ingredient data, and the recipe can be analyzed, scored, and simulated like anything built in the app from scratch.

That changes what “archive” means. The ganache your predecessor left in row 47 can now be diagnosed — run the analysis, read its quality score, see why it never lasted the claimed three weeks — and rebalanced in simulation before anyone weighs a gram.

The payoff compounds across the platform: the same digitized library feeds costing, scaling, PDF data sheets, and the AI development workflow. One migration, every downstream feature.

AI Recipe Development: From Idea to Production-Ready Formula

What those computed metrics unlock: AI drafting, quality scores, and what-if simulation on every recipe in your library.

Plans and limits

Bulk import is a Pro feature, available from the app’s import page or through a connected AI assistant. The Free plan still imports recipes one at a time — from pasted text or a URL — under its monthly AI recipe allowance.

Questions Confectioners Ask