Vol. 1 · Edition 035Free · No paywall

Everyone Needs a Samwise

AI news · Synthesized · Opinionated · 🌿

4
min
from confusing renewal to plain-English breakdown
Everyday AI
By Sam Taylor with Samwise

On which confusing documents AI handles well, the prompt that actually works, and the one step you should never skip before you sign anything.

That renewal notice on your counter? Four minutes with AI will tell you what actually changed.

The renewal notice arrived last week. Seven pages. Dense columns of coverage limits with slightly different numbers than last year, a row of deductibles that look similar but aren't, a footnote about "Schedule B modifications" with no Schedule B attached. You put it on the counter. It has been there for ten days.

This is the most common thing in the world. The most common response is also the most common: leave it there until the auto-pay goes through, and assume nothing important changed, because what else are you going to do.

Here is something else you can do.

What kinds of documents this actually works for

AI is not good at every document task. It makes up specific facts it doesn't have. It misreads dense tables. It hedges on anything that sounds like legal advice. But for one specific task — taking two confusing documents and telling you what's different, in plain English — it is genuinely reliable.

The cases that work well:

  • Insurance renewals — old plan vs. new plan, what actually changed between pages
  • Phone plan upgrades — when the carrier "improves" your plan without asking
  • Lease renewal addenda — the page of changes the landlord attaches every year
  • Multiple insurance quotes — comparing what's actually included in Quote A vs. Quote B
  • Employment offer letters — two jobs, two packages, which one is actually better

The cases that work less well: documents that are mostly images or handwritten, anything where specific dollar figures need to be precisely right (more on this below), and anything where the answer depends on laws specific to your city or state.

The prompt that actually works

Most people try this once and get back a wall of text that restates what's already in the document. That's because "explain this document" is the wrong prompt. You need to tell the AI exactly what you want.

Paste this into ChatGPT, Claude, or Gemini before you paste your documents:

"I'm going to paste two versions of the same document. Please:

  1. Make a plain-English table of everything that changed between them, with the old value on the left and the new value on the right.
  2. Flag anything where a change could affect what happens if I need to make a claim or use the coverage.
  3. Translate any terms I wouldn't recognize into plain English, right there in the table.
  4. Note anything that looks unusual or worth asking someone about.

Don't give me a full summary of either document — just the differences and the flags."

Then paste both documents. If they're PDFs, all three tools let you upload them directly now. If not, copy and paste the text.

$0
What the document comparison costs on the free tier of ChatGPT, Claude, or Gemini

→ Source: All three tools offer free plans that include document uploads

What AI gets right

The comparison table. This is where AI earns its keep. It will find the three lines that changed in a seven-page document and put them side by side. This would take a careful reader twenty minutes. AI does it in under a minute.

Plain-English translation. Terms like "aggregate deductible" (the most you pay total before insurance kicks in for the year), "subrogation rights" (the insurance company's right to sue the person who caused your loss), and "co-insurance clause" (a formula that reduces your payout if you were underinsured) have specific, learnable meanings. AI translates these accurately most of the time.

Spotting unusual additions. If your renewal has quietly added a flooding exclusion that wasn't there before, or dropped roadside assistance, AI will usually catch it and flag it.

What AI gets wrong

Specific dollar amounts. This is the important one. AI can misread numbers in densely formatted tables, especially from PDFs where the formatting doesn't translate cleanly. If the AI tells you your deductible is $1,000 and the actual document says $1,500, you won't find out until you need to use it. After the AI gives you the comparison table, go back to the original document and verify every dollar amount yourself. It takes two minutes.

Advice about what to do. "Is this plan better for me?" is a judgment call that depends on your specific situation. AI will give you the tradeoffs. It cannot tell you what to do with them.

Jurisdiction-specific rules. What a coverage term means in California may not be what it means in Texas. AI gives you the general answer, not the state-specific one.

Prompts that work vs. prompts that waste your time
What you askWhat you getWorth it?
'Explain this document'A summary that restates what's already thereNo
'What's different between these two?'Unstructured comparison, often incompleteSometimes
'Table of changes, flag anything affecting my coverage'Structured, scannable, what you actually needYes
'Should I renew this plan?'A hedged non-answer or dangerous overconfidenceNo

Source spread

  • Claude.ai — Free tier supports PDF uploads and document comparison in the main chat interface. No account required for basic use.
  • ChatGPT — file uploads FAQ — Official documentation on how ChatGPT handles PDF uploads and what's available on which plan tier.
  • Gemini — Google's free option handles PDFs; useful if you're already in the Google ecosystem.
  • Insurance Information Institute — Insurance glossary — For verifying a term yourself when you want to be sure what AI told you is right.

What's real

  • The comparison table is genuinely faster than reading manually — often five to ten times faster
  • Plain-English translation of standard insurance and financial terms is accurate enough to be useful
  • The flag-unusual-things prompt catches things most people miss in manual review

What deserves a side-eye

  • Any specific dollar figure the AI gives you — check it against the original before you rely on it
  • Advice about which plan to choose — that is not a task for AI
  • Complex nested exclusions in specialty policies — AI can misread the conditions that trigger them

What to do about it

Short list. None of this requires a login to anything you don't already have.

  • Open Claude.ai or ChatGPT — free tier works fine. Upload the PDF directly or paste the text of both documents.
  • Use the prompt pattern above. Differences table plus flags. Not a summary.
  • Verify every dollar amount against the original document. Don't trust the AI's number when money is involved. Find it yourself. Two minutes.
  • Ask one follow-up question. "Which of these changes is most likely to affect me if something goes wrong?" Gets you the one thing worth focusing on.
  • For anything with a signature deadline, give yourself time to call the company if the AI flags something confusing. The AI is a starting point, not an ending point.

Further reading

🌿

Liked this? Get the weekly digest.

Free. Monday mornings. The week's stories, synthesized. Unsubscribe anytime.

Your take

How'd I do on this one?

What did I miss?

Tell Samwise (and Sam).

Disagree with the take? Spotted a fact I got wrong? Have context I should have included? Drop it here. Anonymous unless you leave an email.