The Next Input — Issue #091

Your Browser and Excel Are Now One AI

Author

Aaron Bost
November 25, 2025

In partnership with

⚡ The Briefing — 60 sec

🛠️ The Playbook — Chrome x Excel Fusion: Your Zero-Friction Ops Panel

Mission Turn Chrome and Excel into a single AI-powered cockpit for research, enrichment, and reporting.
Difficulty Intermediate
Build time 2–3 hours
ROI 50–70% reduction in manual copy-paste, cleaner data, and near-instant refreshes of critical sheets.

0) Why This Matters

When the same model can watch your browser, speak fluent spreadsheet, and remember how you like things structured, you stop doing work and start checking work.
Opus 4.5 wiring directly into Chrome and Excel means every page is now an ad-hoc API, and every sheet can be continuously refreshed without intern-level keystrokes. This is how you quietly delete whole job fragments from your week.

1) Architecture

Layer

Tooling

Purpose

Browser Brain

Claude 4.5 Sonnet / Haiku via Chrome extension

Read, interpret, and structure webpage content

Spreadsheet Brain

Claude for Excel (Opus 4.5)

Clean, join, dedupe, enrich, and analyse tables

Memory Layer

Supabase

Store schemas, prompts, and transformation templates

Automation Fabric

Make.com or AgentKit flows

Orchestrate browser → Excel → downstream system syncs

Interface

Chrome sidebar + Excel ribbon

Simple control surface for non-technical operators

2) Workflow

A) Web → Excel Data Capture

  1. Identify a “source of truth” page type you keep revisiting (vendor lists, pricing pages, directories, event listings, etc.).

  2. In Chrome, highlight the relevant content (table, list, cards) and open the Claude 4.5 Sonnet extension.

  3. Ask it to extract a structured table aligned to your target schema (for example: Name, URL, Price, Category, Status, Last_Checked).

  4. Claude returns a clean table; send it directly into your chosen Excel workbook via Claude for Excel.

  5. Excel formulas and conditional formatting fire automatically: duplicates flagged, bad prices highlighted, missing URLs surfaced.

B) Excel → Web Enrichment Loop

  1. In Excel, filter to rows missing key fields (for example: empty Price, Status, or Description).

  2. Select that range and invoke Claude for Excel (Opus 4.5):

    • For each row, open the URL in Chrome.

    • Extract price, availability, and short description.

    • Only write to allowed columns.

  3. The browser runs quietly while rows are updated in place.

  4. Rows that cannot be resolved (paywalls, captchas, 404s) are tagged with a Review_Notes column for human follow-up.

C) Insight and Reporting Layer

  1. Once your sheet is structured, send the range to GPT-5-mini for lightweight analysis:

    • price distribution

    • top categories

    • anomalies

    • one-paragraph summary for stakeholders

  2. Generate “change summaries” (what changed since last refresh) to drop into email, Slack, or a weekly report.

  3. Store your favourite analysis prompts in Supabase so the pattern is reusable across similar sheets.

3) Example Prompts

Chrome Extraction Prompt (Claude 4.5 Sonnet)

You are extracting structured data from the highlighted section of this webpage.

GOAL:
Create a clean table suitable for Excel with the following columns:
- Name
- URL
- Price
- Category
- Status
- Last_Checked (ISO date)

RULES:
- Use ONLY content from the highlighted area.
- Ignore navigation, ads, footers, and unrelated sections.
- Ensure each row has at least Name + URL.
- Normalise prices to numeric format and strip currency symbols.

OUTPUT:
Return a CSV-style table with the exact column order above, ready to paste into Excel.

Excel Cleaning and Enrichment Prompt (Claude 4.5 Haiku)

You are assisting with cleaning and enriching an Excel table.

INPUT:
A range with columns: Name, URL, Price, Category, Status, Last_Checked, Review_Notes.

TASKS:
1. Dedupe rows by URL, keeping the most recently updated record.
2. Normalise Category labels into a small, consistent set.
3. Flag missing or obviously incorrect Price values.
4. For rows with missing Price or Status, open the URL in the browser
   and extract:
   - current price
   - availability (In stock / Out of stock / Unknown)

Update only the Price, Status, Last_Checked, and Review_Notes columns.
Return a short bullet summary of what you changed.

Analysis and Reporting Prompt (GPT-5-mini)

You are analysing a cleaned Excel export of market data.

INPUT:
A table with Name, Category, Price, Status.

TASK:
1. Compute:
   - median and mean price overall
   - price range per Category
   - count of items per Status (In stock / Out of stock / Unknown)
2. Highlight any obvious outliers or suspicious values.
3. Write a 5–7 sentence summary suitable for a weekly stakeholder email:
   - one headline insight
   - 2–3 key supporting stats
   - 1 callout risk or anomaly to watch.

Be concise and non-technical.

4) Guardrails

  • Never submit forms, logins, or checkout flows; read-only extraction only.

  • Respect robots.txt and obvious anti-scraping signals; skip risky pages.

  • Maintain a read-only master workbook and write all AI updates into a controlled _AI_Updated copy.

  • Log all modifications (old value, new value, timestamp, model used) into a dedicated Change_Log sheet.

  • Limit enrichment runs to small batches initially (for example: 50–100 rows) until error rates are understood.

5) Pilot Rollout — 3 hours

  1. Setup (30 minutes)

    • Install Claude for Chrome and Claude for Excel (Opus 4.5).

    • Create a dedicated Chrome_Excel_Pilot.xlsx workbook with your desired schema.

  2. Choose one workflow (15 minutes)

    • Example: competitor pricing tracker, vendor master list, or events calendar.

    • Export your current “messy” data as a baseline tab.

  3. First Web → Excel run (45 minutes)

    • Use the Chrome Extraction Prompt on 10–20 representative pages.

    • Inspect and tweak the output until it drops into Excel without manual fixes.

  4. Enrichment loop (45 minutes)

    • Run the Cleaning and Enrichment prompt on a small filtered subset (for example: 30 rows).

    • Manually spot-check 100 percent of those rows to calculate accuracy and catch edge cases.

  5. Lock in templates (45 minutes)

    • Store final prompts and schemas in Supabase with version tags.

    • Document a one-page internal SOP: when to run, how to validate, what to do when pages break.

6) Metrics

  • Minutes saved per refresh cycle compared to the baseline manual process

  • Percentage of rows fully updated by AI without manual correction

  • Error rate on key fields such as Price and Status based on sampled human review

  • Number of enrichment cycles run per week for each sheet

  • Time required to refresh a full sheet before and after deployment

Pro Tip: Treat prompts like code. Version them, tag them to specific sheets, and retire old ones. Prompt sprawl is how AI workflows quietly fall apart.

🎯 The Arsenal — Tools & Prompts

Copy-paste orchestration prompt:

You are orchestrating a Chrome ↔ Excel data pipeline.

OBJECTIVE:
Keep my Excel workbook in sync with the latest information from the web,
with minimal manual intervention and clear audit trails.

RULES:
- Read-only behaviour in the browser (no form submissions).
- Only update approved columns in Excel.
- Log every change with old value, new value, and timestamp.
- Never guess: if data is ambiguous or blocked, leave a clear note instead.

STEPS:
1. Extract structured tables from web pages that match my target schema.
2. Clean and dedupe rows in Excel.
3. Enrich missing values by revisiting URLs when needed.
4. Generate a short summary of changes and any anomalies.

OUTPUT:
- Updated table
- Change log
- 5–7 bullet summary of key insights.

💡 Free Office Hours

Want help implementing anything? Book a free 15-minute Office Hours slot—no sales pitch, just workflows solved.

Software sprawl? That’s SaaD.

Software was supposed to make work easier. Instead, most teams are buried under it.

That’s SaaD – Software as a Disservice. Dozens of disconnected tools waste time, duplicate work, and inflate costs.

Rippling changes the story. By unifying HR, IT, and Finance on one platform, Rippling eliminates silos and manual busywork.

  • HR? One update applies to payroll, benefits, app access, and device provisioning instantly.

  • Finance? Close the books 7x faster with synced data.

  • IT? Manage hundreds of devices with a single click.

Companies like Cursor, Clay, and Sierra have already left outdated ways of working behind – gaining clarity, speed, and control.

Don’t get SaaD. Get Rippling.

🕹️ Game Over

Wire Chrome and Excel together once, then let the models do the loops while you make the calls.

— Aaron Automating the boring. Amplifying the brilliant.