AI Skill of the Day Digest April: Your Most Requested Skills

Every day, The Neuron's newsletter teaches 700,000+ readers one new AI skill they can use immediately. This collection is all the ones you requested.

We publish these digests so you never have to dig through your inbox to find that one tip you half-remember from Thursday (been there, bro). This batch is different from the usual weekly roundup: we sat down with every skill request readers sent us over the last six weeks, grouped the repeats, checked them against what we'd already covered, and built this from the five topics the most people wrote in about.

Have a specific skill you want to learn? Request it here.

First, a starting-line note: if you're brand new to AI and just need the basics, start with our two foundational guides before reading this one:

How to Actually Use AI in 2026: The Complete Proficiency Guide walks through the 5-Level Stack from Projects to Agents.
AI for Total Beginners: The Full Livestream Guide is the two-and-a-half-hour companion that demos every level live.

Past the basics? This digest is what comes next. Every skill below is something a real reader wrote in and asked us to cover. The theme across all five: readers are past "what is AI," and they want it applied to their actual reality. Their life (not just their job). Their locked-down Copilot-only work environment. Their Excel files. Their diagram tool. Their personal inbox.

How to use this digest:

Skimming? Each skill opens with a bold hook that tells you exactly what you'll learn. If it's not relevant, skip to the next.
Implementing? Every entry includes a copy-pastable prompt in a code block. Grab it, tweak it for your context, and try it today.
Catching up? Start with Skill 1 (the retiree edition) and work forward. They're ordered by request volume, not difficulty.

Let's get into it.

🎓 1. AI for Personal Operations: Apply the 5-Level Stack to Your Life (Not Just Your Job)
🎓 2. The Microsoft Copilot Power User Guide (When Copilot Is All You Can Use)
🎓 3. Generate Flowcharts, Swim Lanes & Process Diagrams from Plain English
🎓 4. Build a Personal Executive Assistant in Claude Cowork (Including Web Scraping, No Zapier Required)
🎓 5. The Spreadsheet Whisperer: Prompts That Actually Understand Your Data
Keep Learning

🎓 1. AI for Personal Operations: Apply the 5-Level Stack to Your Life (Not Just Your Job)

The single most common reader request we got this month came from a recently-retired reader who told us something that stuck:

"I read your 5-Level Proficiency article and realized I want to apply it to my life, not my work. I don't have a corporate environment to point AI at anymore. But I still have a life to run."

Several other readers asked variations of the same question: how do you use AI in retirement, for personal scalability, for time management, for daily activities that aren't work?

Here's the reframe: everything we teach about AI at work applies just as well to the project of running your life. Your inbox, your calendar, your travel plans, your medical records, your home maintenance schedule, your family's contact info. That's your organization. You're the CEO. Let's hire an assistant.

Level 1 (Projects): Build your "Life Operations" workspace.

Create a single Project in Claude or ChatGPT called "Life HQ." Upload: your travel preferences (aisle or window, diet, airlines with miles, hotel loyalty programs), family contacts with birthdays and medical notes, a house inventory (paint colors, filter sizes, appliance model numbers), and a one-page "about me" with your health basics, allergies, and go-to pharmacy. Write custom instructions that cover: how formal you want responses, what you want flagged vs. just handled, and what's off-limits (financial trades, medical prescriptions, legal decisions).

Level 2 (Prompting): Front-load context once, then ask in plain English.

Because your Project knows your context, you can ask things like "plan a 10-day trip to Portugal for my 40th anniversary" and get back something actually useful. Dan Shipper at Every describes this as the difference between telling a stranger what to do and delegating to someone who already knows you.

Level 3 (Skills): Package repeated tasks.

When you've gone through a task twice, package it. Real Journey Travels documented how Claude Cowork can turn "plan my trip" into a reusable skill that pulls confirmations, organizes booking files into trip-specific folders, and even renames attachments from download_558392.pdf to Lisbon_Hotel_Confirmation.pdf. Good candidates: doctor-appointment prep ("pull my test results, summarize what's changed, and draft questions for the visit"), annual insurance review, holiday gift list by person.

Level 4 (Automations): Schedule your morning briefing.

Inspired by Dan Shipper's Plus One setup, schedule a daily task in Claude Cowork or ChatGPT Codex that runs every morning and compiles: weather for today, any flight changes, bills due this week, family birthdays in the next 7 days, and one personalized recommendation based on your interests.

Level 5 (Agents): Let it run toward a goal.

TechPP tested Claude Cowork for exactly this kind of work and found genuine wins in subscription auditing (it finds services you forgot you were paying for), coupon hunting during checkout, and financial tracking that spots hidden expenses.

One warning from their testing: don't give Cowork access to folders with sensitive financial documents. Upload specific statements when you need them analyzed, not the whole folder.

A critical caveat for financial and medical use: MIT Sloan professor Andrew Lo has been studying AI financial advice since 2022 and his conclusion is clear: AI is strong at trade-off analyses, scenario exploration, and behavioral coaching. It's weak at precise tax optimization, regulatory nuance, and arithmetic. And crucially, it bears no legal responsibility.

In his words: "You've got to take their advice with a pound of salt." Use AI to prep questions, run scenarios, and understand your options. Then take those to an actual CPA, estate attorney, or doctor.

Here's a "Life CTO" custom instructions template to drop into your Life HQ project:

You are my personal chief-of-staff. You help me run my life with the same rigor I used to bring to my work.

About me:
- [Name, age, location, any relevant health context you're comfortable sharing]
- [Key relationships: partner, kids, close friends, routine medical team]
- [Current major projects in my life: travel plans, home renovations, health goals, etc.]

How I want you to work:
1. For big decisions (money, health, legal): give me the framework, options, and tradeoffs. Never give me the answer. Always end with "questions to ask a [CPA/lawyer/doctor]."
2. For logistics (travel, appointments, shopping): just handle it. Draft the email, pull the options, make the spreadsheet. Show me the draft; I'll send it.
3. For research: cite your sources. I want links I can verify, not summaries I have to trust.
4. Always ask "what did I miss?" at the end of any analysis.
5. Never make assumptions about my family's health, finances, or relationships that I haven't told you.

Off-limits without explicit permission in the same message:
- Sending any communication on my behalf
- Making any financial transaction
- Contacting my doctors, lawyer, or accountant directly

Favorite insight: The retiree reader who prompted this skill ended his email by saying he used Gemini to help wordsmith it. When Gemini asked if he should mention that fact, it said yes. That's Level 3 thinking already: the tool that helped him ask the question recommended he be transparent about using it. He didn't know he was at Level 3.

P.S: Claire Vo and Jesse Genet are probably the two best creators to follow who have solid personal life agent systems (both using OpenClaw). Check out Jesse' interview with A16z, The Cognitive Revolution, and Claire Vo on How I AI, then check out Claire's OpenClaw setup guide (podcast version) and her walkthrough of her own setup with Lenny on Lenny's podcast.

🎓 2. The Microsoft Copilot Power User Guide (When Copilot Is All You Can Use)

Multiple readers wrote in with a version of this: "The only AI we can use at work is Copilot. What can we do?" If that's you, stop feeling left out. Wave 3 of Microsoft 365 Copilot (announced March 9, 2026) just transformed Copilot from a chat assistant into a full agentic platform. And the best part for Claude/ChatGPT fans: Copilot is now powered by both OpenAI and Anthropic models. You can literally pick Claude inside Copilot Chat.

Here's what you need to know and what to do about it.

What actually shipped in Wave 3 (per Microsoft's announcement):

Copilot Cowork is the long-running, multi-step agent, built in collaboration with Anthropic. It uses the same tech that powers Claude Cowork. Available in the Frontier program now.
Multi-model Chat means you can now select Claude directly in Copilot Chat alongside the latest OpenAI models. Microsoft's Jeremy Chapman demos it here.
Researcher with Critique and Council: one model generates a draft, a second model critiques it, then you can view Council (a side-by-side comparison of Claude vs. GPT outputs on the same prompt). Researcher scored 13.8% higher on the DRACO benchmark after the upgrade.
Agent Mode in Word, Excel, PowerPoint: Copilot can now iteratively build complex deliverables with minimal prompting between steps.
Work IQ is Copilot's secret weapon that ChatGPT and Claude can't match: it already knows your emails, meetings, files, and org chart via Microsoft Graph, so you never have to upload context.

How to unlock all of this:

Get your tenant into the Frontier program. This is the single highest-leverage move for any Copilot user. It's not invite-only; it's opt-in. Your M365 admin goes to Admin Center → Copilot → Copilot Frontier and enables it. You can scope it to specific users if you want to pilot with a small group first.
Enable Anthropic as a subprocessor. Same place: Admin Center → Copilot → AI Providers. This turns on Claude inside Copilot Chat and Researcher. Microsoft's support doc has the step-by-step. Note: EU, UK, and government cloud tenants have this turned off by default due to data sovereignty rules.
Use the right Copilot surface for the task. Microsoft's own Inside Track blog breaks this down. For quick questions, Copilot Chat. For deep analysis, Researcher. For multi-step execution across apps, Copilot Cowork. For data analysis inside Excel, Agent Mode. Most people use only Chat and miss the rest.
Build custom declarative agents. The Microsoft 365 Agent Store lets anyone with a Copilot license build custom agents without code, combining custom instructions, knowledge files, and actions. Your admin approves them for distribution.

The prompt that actually uses Work IQ (the thing Copilot does better than ChatGPT or Claude):

Prep me for my 1:1 with [manager's name] tomorrow.

Pull:
- Any decisions or action items from our last three 1:1s (check my Teams chat history and meeting notes)
- Emails between us from the past two weeks that mention open projects
- Status of the deliverables I committed to last time (check Planner/Loop/Tasks)
- One topic I should raise proactively based on changes in our team's recent work

Output a one-page briefing doc in Word with:
1. "What's outstanding from last time" (table)
2. "Updates on my commitments" (bullets with file links)
3. "Topics I should raise" (3 items with 2-sentence context each)
4. "Questions I should ask" (5 items)

Save to my "1-1 Prep" folder in OneDrive and surface a Teams reminder at 8am tomorrow.

The pricing reality (per eWeek's Copilot cheat sheet): Microsoft 365 Copilot is $30/user/month as an add-on to existing M365 licenses. Cowork is included at no extra cost during the preview. The new Microsoft 365 E7 plan bundles E5 + Copilot + Agent 365 at $99/user/month if you want the full enterprise package. Individual Copilot (the consumer version) is free but dramatically less capable.

When you should still reach for ChatGPT or Claude on the side: creative writing and long-form narrative (Claude is still stronger), one-off research outside your company's data (Perplexity or ChatGPT), and skill creation (Claude's Skills system is more mature). But for anything that needs your work context, Copilot + Work IQ wins, because it doesn't require uploading anything.

Favorite insight: One writer at Envision IT documented their rollout and noted something we'd missed: "Users do not need separate Anthropic accounts. Everything runs within Microsoft's existing security, identity, and compliance boundaries." That's the Copilot user's actual superpower. You get Claude without any of the governance headaches that would stop IT from approving a separate tool.

Now, for what you REALLY want when it comes to Copilot: Tutorials! Here are three we recommend (from friend of our YouTube channel Kevin Stratvert; we were hanging out with this dude last year at Microsoft Ignite the night his channel broke 4M subscribers! He's legit):

Check those out, then write in with any follow-ups you need help with or didn't understand.

🎓 3. Generate Flowcharts, Swim Lanes & Process Diagrams from Plain English

Three readers wrote in asking the same thing: what's the best AI for producing process flows, swim lane diagrams, and matrices? One specifically asked for "something that replaces Visio or Lucidchart, where I can describe something verbally and get a diagram back."

Good news: dragging shapes in Visio is no longer the only way. The right tool depends on whether you want a technical diagram, a business-facing flowchart, or something you can embed in docs and never update manually again.

Here are the four paths, ranked by how most readers should approach this:

Path 1: Claude Artifacts + Mermaid (free, fastest for most people)

If you're already using Claude or ChatGPT, you can generate diagrams right inside the chat, no external tool needed. Claude's Artifacts feature renders Mermaid diagrams natively in the Artifact panel. Mermaid is a text-based diagramming syntax (think: "write English, get a flowchart"). It supports 23 diagram types including flowcharts, sequence diagrams, ER diagrams, Gantt charts, and swim lanes. Copy the output, paste it into any Mermaid-compatible tool (GitHub, Notion, Obsidian, VS Code) and the diagram renders automatically. For swim lanes specifically, EdrawMax has a full syntax guide with color-coding examples.

Note: you can also install the Excalidraw MCP server (Connector) on Claude and use that for diagrams; it's available in the Connector Directory (hit the + sign, then "browse connectors"), so just search for it there.

Path 2: Whimsical AI (best for business flowcharts and team-facing diagrams)

Whimsical is what most non-technical people actually want. Clean output, minimal learning curve, collaborative editing, and AI generation from a single prompt. Jotform's 2026 review gave it high marks for prompt adherence (you can give it a basic description and get something useful). Starts at $10/editor/month; free tier gets you 100 total AI actions. Biggest weakness: SVG export is weak, so it's not great for anything production-facing.

Path 3: Eraser DiagramGPT (best for technical and architecture diagrams)

If your diagrams involve cloud architectures, databases, microservices, or anything engineering-adjacent, Eraser.io is purpose-built. Their DiagramGPT understands prompts like "microservices architecture with Kafka, Postgres, and a read replica" and outputs clean diagrams with correct AWS/GCP/Azure icons. Diagrams are stored as code (diagram-as-code) so you can version them alongside your repos. Starts at $10/month; free tier gets 3 files.

Path 4: Lucidchart AI (best for enterprise, regulated industries, deep integrations)

If you need SSO, SCIM, audit logs, BPMN shapes, and deep Atlassian/Salesforce/M365 integrations, Lucidchart is the incumbent. Their 2024 AI additions let you generate flowcharts from prompts, import org structures from Azure AD, and auto-diagram AWS accounts from account data. Starts at $9/month.

The power-user trick: Mermaid + draw.io combo for editable outputs

Blogger Stephen Turner documented a clever setup where Claude generates Mermaid code, runs a small Python script to compress and encode the diagram, and outputs an HTML artifact with a clickable button that opens the diagram in draw.io (fully editable). Best of both worlds: AI-generated structure plus drag-and-drop refinement.

The prompt for a business swim lane (works in Claude, ChatGPT, or any Mermaid-compatible tool):

Create a swim lane diagram in Mermaid syntax showing our customer onboarding process.

Lanes (columns): Sales, Customer Success, Finance, Product
Swim lane flow:
1. Sales closes deal → generates welcome email
2. Customer Success schedules kickoff call → assigns CSM
3. Finance creates invoice → sends to customer
4. Product provisions account → sends credentials
5. Customer Success runs kickoff call (all hands)
6. Product activates usage tracking

Requirements:
- Use subgraph to separate each department
- Apply distinct colors per department (blue for Sales, green for CS, yellow for Finance, purple for Product)
- Show decision points as diamonds (e.g., "Did they sign the MSA?")
- Add conditional arrows with labels
- Include a timeline estimate on each step

Output: Mermaid code I can paste into Mermaid Live Editor, plus a plain-English summary.

The best reason to use Mermaid over a shape-based tool is that your diagrams become version-controllable. As GitHub engineer Cory Miller noted, diagram-as-code means you can check flowcharts into the same repo as your docs, review them in pull requests, and update them with a find-and-replace. The diagram is no longer a static artifact someone made once and abandoned; it's a living part of your documentation.

🎓 4. Build a Personal Executive Assistant in Claude Cowork (Including Web Scraping, No Zapier Required)

Four readers asked variations of this exact skill: how do I set up Claude Cowork as my personal executive assistant, including web scraping "without having to connect to Zapier or other tools?" Great news: as of the Q1 2026 updates, you can do all of it natively.

Here's the full buildout, based on Sid Bharath's Cowork deep dive, Technically Curious's personal agent guide, and Claude Blattman's Executive Assistant toolkit.

The stack:

Claude Desktop app with Cowork (requires Pro $20/mo or higher)
Gmail + Calendar connectors (native, no Zapier)
Firecrawl plugin for web scraping (JavaScript-heavy sites included)
Context files for policies and preferences
Scheduled tasks for daily briefings
Skills for packaging repeated workflows

The 7-step setup:

Step 1: Download Claude Desktop. Available on macOS (Apple Silicon) and Windows (since January 2026). Grab it from claude.ai/download.
Step 2: Create a dedicated "Assistant" email and project. This is the single most important step and most people skip it. As Katie Parrott's team at Every noted, model your setup after a real executive assistant: create a dedicated email like yourname.assistant@gmail.com, give it delegated access to your calendar, and let it handle AI-drafted messages. This keeps AI-generated correspondence isolated from your main identity and gives you a natural audit trail.
Step 3: Connect Gmail and Calendar natively. Go to Customize → Connectors. Both are official Anthropic connectors, no third-party middleware needed. Authenticate with your dedicated assistant email.
Step 4: Install the Firecrawl plugin for web scraping. In Claude Desktop, run /plugin and search for Firecrawl. This gives Claude the ability to scrape any public website (including JavaScript-heavy pages that regular web_fetch can't handle), crawl entire sites, and extract structured data. The free tier gives you 500 credits per month, which is plenty for personal use. Setup takes 2 minutes.
Step 5: Write hard policies, not guidelines. Claude Blattman's toolkit calls this out specifically: without explicit policies, every session starts with ambiguity. Can Claude mark emails as read? Send messages? Move calendar events? Write these as policy files Claude reads at the start of every session.
Drop this in your project knowledge as assistant-policies.md:
- # Email Rules
  - NEVER send emails directly. Always draft and surface for my approval.
  - ALWAYS add "Drafted with Claude, sent by [my name]" to the bottom of any external email you draft.
  - VIP senders (always flag): [list]
  - Auto-archive: newsletters, marketing, "no-reply" addresses
  
  # Calendar Rules
  - NEVER delete calendar events. Propose a change; I approve.
  - NEVER accept meeting invites on my behalf. Flag and wait.
  - Protected time: [list days/times that are always "busy" for you]
  - Lunch hour protection: 12pm-1pm is never schedulable unless I explicitly allow.
  
  # Research Rules
  - When using Firecrawl or web search, always cite sources inline.
  - If a page requires login, tell me; don't try to work around it.
  - For news or "current" topics, always add a date to the query.
  
  # Scope Boundaries
  - OFF-LIMITS without explicit permission: financial transactions,
  medical decisions, legal communication, HR conversations,
  anything involving my family members' information.
Step 6: Set up a morning briefing as a scheduled task. In Cowork, type "schedule" followed by your task description. Per the Cowork docs, your Mac or PC must stay awake with Claude Desktop open for scheduled tasks to run.
- Use this prompt:
- - Every weekday at 7am, run my morning briefing:
    
    1. Pull all emails received since 6pm yesterday. Categorize into:
    URGENT (needs reply today), IMPORTANT (this week), FYI (archive later).
    For URGENT, draft a reply I can send in under 2 minutes each.
    
    2. Pull my calendar for today. For each meeting:
    - Name the attendees
    - Pull the last email thread with each attendee
    - Flag any I haven't prepared for
    - Draft 3 talking points per meeting
    
    3. Use Firecrawl to check for news mentions of:
    - My company: [company name]
    - Our top 5 competitors: [list]
    - My 3 priority accounts: [list]
    Only surface items from the last 24 hours.
    
    4. Output as a single email draft to my assistant address with subject
    "Morning briefing [date]". Do not send.
    
    Checkpoint-in on any task that takes longer than 5 minutes.
Step 7: Save your best workflows as Skills. After you've iterated a workflow into something reliable, ask Claude: "Use your skill creator to turn this conversation into a reusable skill I can call anytime." Save it to your skills library. Now it's a one-click command instead of a re-prompt.

For web scraping specifically (the reader's direct question): Firecrawl replaces most one-off Zapier use cases. Technically Curious ran a real example with this prompt: "Use firecrawl to pull the list of companies sponsoring the HR Transform conference from [url]" and it returned the full sponsor list, including from dynamically-loaded page content that regular web_fetch couldn't handle.

Favorite insight: Obie Fernandez's personal CTO system gets at the real unlock: "I've had executive assistants in the past. Good ones. This system is better. It never forgets anything, never needs to be brought up to speed, and operates at the speed of conversation." If you set it up right, that's not really hyperbole; the policies and dedicated email are what separate a toy from a system.

🎓 5. The Spreadsheet Whisperer: Prompts That Actually Understand Your Data

Three readers wrote in asking for better prompts to "decipher spreadsheets and data sets at work." A common follow-up: how do you combine spreadsheet analysis with building slides or email marketing automation?

Here's the answer: as of 2026, you have two radically different paths depending on what kind of spreadsheet work you do.

Path 1: Claude for Excel (best for finance, modeling, multi-tab workbook analysis)

Anthropic quietly released the single best tool for spreadsheet work: Claude for Excel, a sidebar add-in that reads your actual Excel file, including formulas, cross-sheet dependencies, and pivot tables. It upgraded to Claude Opus 4.6 on February 5, 2026, and on February 17 it added MCP connector support for S&P Global, LSEG, Daloopa, PitchBook, Moody's, and FactSet. It's included free with any paid Claude plan (Pro $20/mo and up).

Why it's different from uploading your Excel file to ChatGPT: When you upload an .xlsx to ChatGPT, it captures static values and loses all formula logic. Claude for Excel reads live formulas, provides clickable cell references, makes direct edits while preserving dependencies, and navigates across dozens of tabs. DataCamp's review called out the core difference: "Although Claude supports analysis and suggestions, you remain in control. You review every proposed change and choose what gets applied."

Setup (2 minutes):

Open Excel → Home → Add-ins (or Insert → Get Add-ins)
Search for "Claude by Anthropic"
Click install
Memorize: Ctrl + Alt + C (Windows) or Ctrl + Option + C (Mac) toggles the Claude sidebar
Sign in with your paid Claude account

The 4-step analysis framework (from MindWired's guide, adapted from the classic Explore → Diagnose → Predict → Prescribe pipeline):

Explore: "Tell me what's in this workbook. Which tabs relate to each other? Flag anything unusual."
Diagnose: "Why did [metric] change? Show me the formula chain that drives it."
Predict: "If [variable] changes by X%, show me the downstream effect across all tabs."
Prescribe: "Build a summary sheet with the three recommendations a finance director would make from this data."

The dashboard prompt that forces real analysis (from Artificial Corner's tested review):

I have [describe your dataset: rows, columns, what each tab contains].

Step 1: Create a dashboard in a new sheet called "Executive Summary" with key trends.
- Use Python scripts for any calculation Excel formulas can't handle cleanly
- Include 3-5 KPI cards at the top (use conditional formatting for up/down arrows)
- Add 2-3 charts: one time series, one comparison, one distribution
- Add a "What changed vs. last period" callout box

Step 2: Format static input cells in BLUE and formula-linked cells in BLACK.
This lets me audit your work visually.

Step 3: Add a "Questions I should ask" section at the bottom with 3 analytical
questions this data surfaces but doesn't answer.

Before making any change to existing cells, show me the before/after formula.
Don't overwrite anything without approval.

The audit trick (also from Artificial Corner): if a financial assumption doesn't update the right downstream cell, highlight it and ask "Why isn't this cell linked to [x]?" The value of AI in Excel isn't in the build; it's in the audit.

Path 2: ChatGPT for Excel (best for financial modeling and pulling live market data)

OpenAI launched ChatGPT for Excel on March 5, 2026, and it's a legitimate competitor to Claude's add-in. Powered by GPT-5.4 Thinking, it embeds a ChatGPT sidebar directly inside your workbook. Like Claude's version, it reads live formulas, traces cross-sheet dependencies, and asks permission before editing cells.

Where it pulls ahead: financial data integrations. OpenAI partnered with FactSet, S&P Global, LSEG, Moody's, Dow Jones Factiva, MSCI, Third Bridge, Daloopa, and MT Newswire at launch, so analysts can pull institutional-grade market data directly into their spreadsheet workflows without switching apps. Claude for Excel has similar MCP connectors for some of these providers, but OpenAI's list is broader at launch.

The benchmark numbers are worth knowing: on OpenAI's internal investment banking benchmark (building a three-statement model with proper formatting and citations), performance jumped from 43.7% with GPT-5 to 87.3% with GPT-5.4 Thinking. That's the kind of structured, formula-heavy work where GPT-5.4 genuinely excels.

Setup (same 2-minute process as Claude's): Open Excel, go to Add-ins, search "ChatGPT," install, sign in with your OpenAI account. Available to Plus ($20/mo), Pro, Business, Enterprise, and Edu users in the US, Canada, and Australia. Google Sheets integration is listed as "coming soon".

The key difference between the two:

Claude for Excel is stronger at reasoning across complex multi-tab workbooks and explaining why a formula chain behaves the way it does.
ChatGPT for Excel is stronger at pulling live financial data from external providers and building models from scratch using natural language.
If your work is "help me understand this inherited 30-tab monster," Claude wins.
If your work is "build me a DCF model using FactSet data," ChatGPT wins.
Both ask permission before editing. Both link answers to specific cells.
Use whichever matches your workflow, or install both (they're separate add-ins and don't conflict).

A word of caution on both: Anthropic and OpenAI both warn that prompt injection attacks can hide malicious instructions inside spreadsheet cells, formulas, or comments.

Only use either add-in with trusted spreadsheets, not downloaded vendor templates or random files from the internet. And always review changes before finalizing client-facing work.

Favorite insight: The TechTide review of Claude in Excel nailed the real productivity shift: "Think of Claude as an analyst sitting next to you who can read your entire spreadsheet and answer questions on demand." The same also applies to ChatGPT's version now.

The question habit is the whole game: before scanning your spreadsheet manually, ask the AI first. "Is there any row where the cost exceeds the budget?" takes 10 seconds to ask and could save you 10 minutes of scanning. But ya know, don't forget to actually verify that with the source data to confirm...

Keep Learning

This was Part 1 of our reader-requested skill digest for April 2026. Part 2 (coming soon) will cover the next five most-requested skills: AI for investment research (with proper risk disclaimers), guardrails for custom GPTs and Gems that stay on-task, privacy-first AI for regulated and government work, and how to build an internal AI training curriculum for your team.

Want a new skill delivered to your inbox every weekday? Subscribe to The Neuron. It's free.

Brand new to AI? Start with our 5-Level Proficiency Guide and the AI for Total Beginners Livestream Guide before diving into this digest.

Have a specific skill you want to learn? Request it here.

Previous digests:

AI Skill of the Day Digests:

Prompt Tip of the Day Digests:

Foundational guides: