← All use cases

Use case P59 · Human

Car buyer uses Claude Pro and NotebookLM to compare vehicle reports, then chooses after a test drive

Sathish Jayapal ·

The short version

What they did

Sathish Jayapal collected vehicle-history reports and dealer documents while replacing his own car. Claude Pro made a comparison spreadsheet, while NotebookLM organized 73 uploaded documents into questions, comparison material and audio overviews he listened to on runs. He bought a Subaru Ascent despite the spreadsheet ranking other options higher, after valuing its space, ride and dealer experience on a test drive. He reports invented details in NotebookLM, and the article has inconsistent cost comparisons and automotive labels, so neither its rankings nor its tax and reliability claims are validated here.

Result (as claimed)

“I bought a 2022 Subaru Ascent.”

“It ranked lower on the spreadsheet.”

“The audio overview feature let me review vehicle comparisons during my long runs”

“Sometimes hallucinated details (always verify NotebookLM’s claims against source docs)“

Prompt

Their prompt:

Across all the 2022 Honda CR-V reports, what’s the most common maintenance task?

This assumes the source reports have been uploaded. Any reported pattern still needs checking against those reports and qualified vehicle advice.

A fuller version to copy:

I’m comparing used cars and want to understand the maintenance recorded in their reports. I’ve uploaded [vehicle-history reports and dealer documents]. Focus on [year, make and model].

1. Identify each vehicle and group any documents about the same car.
2. List the maintenance tasks actually recorded, then count how many separate vehicles show each task.
3. Give me a table of task, vehicle count, dates and supporting document pages, followed by a short summary of the most common tasks.

Use plain words and keep the summary under 150 words. Don’t count the same service twice because it appears in two documents. Say when records are missing or unreadable. Don’t treat absent records as proof a task wasn’t done, or turn this small sample into a claim about the model’s reliability.

Do it yourself

These steps are my suggested way in, not Sathish’s setup.

What you need

  • NotebookLM, which is free with a Google account.
  • A chat assistant: Claude, ChatGPT or Gemini.
  • Vehicle-history reports, dealer listings and reviews for each car you’re looking at.

Start simple

  1. Make a notebook in NotebookLM and upload the reports and listings.
  2. Ask questions like the one above. Click each citation to see where the answer came from.
  3. Give the chat assistant the key facts per car: price, miles, accidents, owners, service history.
  4. Ask it for a comparison spreadsheet with one row per car.
  5. Pick your top two or three to test drive.

Take it further

  • Make an audio overview in NotebookLM and listen to it on a walk or a drive.
  • Ask for the total cost of each car over five years, using insurance and fuel numbers you look up.
  • Ask for a list of questions to ask each dealer.

Check your work

  • NotebookLM can make up details. Check every important claim against the report it cites.
  • Recheck the spreadsheet’s math yourself.
  • Money: get a pre-purchase inspection from a mechanic you choose, and read the final paperwork line by line before you sign.

Source checked. The link, author, date and quote were confirmed against the source. The result itself has not been reproduced. Their prompt is quoted from the source; any fuller version is written for this library. Business size: Individual. Results are quoted as the source claimed them and never improved. Spot an error or a broken link? Email charlie@cbarmorecpa.com.

More in Money and shopping

New cases go out every Saturday in Inside AI.