Explore Automobile Data with ChatGPT

Build a reusable ChatGPT workspace to analyze historical automobile data.

Introduction

30 Second Summary

Numbers can look convincing when they appear in a polished table or chart. A hidden method can still turn that polished answer into a shaky conclusion.

In this project, you will build an Automotive Insights Lab inside ChatGPT Projects. You'll refine a vague prompt into a checked historical briefing with visible calculations.

What You'll Build

You'll open Automotive Insights Lab to a ready-to-use historical automotive briefing whose claims trace back to a checked table and chart.

By the end of this project, you'll have:

A reusable Automotive Insights Lab where you can start a fresh chat and use automobile.csv without uploading it again.
A before-and-after prompt comparison that shows a broad claim become a findings table with visible methods, cleaned record counts, and limitations.
A checked historical briefing built around a price-versus-horsepower chart. Its cleaned record count matches the calculation audit.
Secret Mission: Compare automobile body styles in a fresh project chat without uploading the dataset again.

Are there any prerequisites?

No prior coding or data analysis experience is required. You'll need a Mac plus a free ChatGPT account.

ChatGPT Free has file-upload and data-analysis limits.

Before We Start

Before the hands-on work begins, commit to the briefing you are building and the standard each numeric claim must meet. This keeps your analysis grounded in the uploaded data.

Set Up Your Automotive Insights Lab

Your written commitment sets an evidence standard for every numeric claim. That standard depends on keeping the historical dataset available throughout the analysis.

A workspace in ChatGPT Projects keeps the dataset beside your instructions. It also gives future chats access to the same context.

In this step, get ready to:

Sign in to ChatGPT with the left sidebar visible.
Download the UCI Automobile dataset as automobile.csv.
Create the Automotive Insights Lab project with the dataset and analysis instructions.

Access ChatGPT with your account

A signed-in account gives you access to the project workspace used throughout this analysis. Choose the path that matches your current account status.

Already have a ChatGPT account

Visit ChatGPT in your web browser.
Sign in using the same method that originally created your account.

When sign-in finishes, you'll see the ChatGPT left sidebar.

Need a ChatGPT account

Visit ChatGPT in your web browser.
Choose the option to create a free account.
Register with a personally controlled email address.
Complete the on-screen account flow.

When account creation finishes, you'll see the ChatGPT left sidebar.

Having trouble signing in?

An account created through Continue with Google, Continue with Microsoft, or Continue with Apple should use that same method when signing in.

Ask for help: Help me access my ChatGPT account.

Download and check the automobile dataset

The UCI Machine Learning Repository provides the historical Automobile dataset as a direct CSV download. Checking its header row confirms that you saved the correct source before uploading it.

Open the official UCI Automobile CSV in your browser.
Press Command+S to open the save dialog.
Select the Downloads folder.
Enter automobile.csv in the Save As field.
Select Save.
Press Command+Space to open Spotlight.
Enter automobile.csv in the Spotlight search field.
Press Return to open the saved file.

In the first row, you'll see headers such as make, horsepower, city-mpg, highway-mpg, and price.

Cannot find the downloaded file?

Search the Mac Downloads folder for automobile.csv.
Check that the filename ends with .csv.
Download the file again from the official UCI link if the header row is missing.

Ask for help: Help me check my downloaded automobile CSV.

Create and verify the project

The project becomes the reusable home for your file and analysis rules. Its instructions keep later responses beginner-friendly while preserving the historical evidence boundary.

Return to the browser tab containing ChatGPT.
Select New project in the left sidebar.
Enter Automotive Insights Lab as the project name.
Choose an icon for the project.
Choose a color for the project.
Upload automobile.csv from the Mac Downloads folder using the project file control.

You'll see automobile.csv listed as a project source.

Why use a project?

A project keeps related chats and files in one workspace. Future chats can reuse the uploaded dataset without another upload.

Project instructions apply within this workspace. Every analysis starts with the same evidence rules.

Open the project's more options menu by selecting ••• beside its name.
Select Project settings.
Find the project instructions field.
Add the historical-data guardrails by pasting this text into the project instructions field:

You are helping a beginner analyze a historical automobile dataset.
Use plain language and short sections.
Separate calculations from interpretations.
State how missing values are handled and how many records each calculation uses.
Do not describe this historical dataset as evidence about the current automobile market.

What do these instructions change?

The first two sentences keep explanations accessible to a beginner.
The third sentence gives calculations their own visible place.
The fourth sentence exposes missing-value handling and record counts.
The final sentence prevents claims about the current automobile market.

Save the project settings.
Confirm that the project instructions field contains all five sentences.
Return to Automotive Insights Lab from the left sidebar.
Start a project chat.

Before you send the setup check, consider whether the project can identify the dataset dimensions from its uploaded source.

Check the uploaded dataset by sending this prompt:

State the number of data rows and columns in automobile.csv. Then create a table listing every column with missing values and the number missing.

What does this prompt check?

The first sentence checks whether ChatGPT can read the complete dataset dimensions.
The second sentence checks whether missing values are identified by column.
The table makes each missing-value count visible for later verification.

The response should confirm 205 data rows. It should also confirm 26 columns.

You'll also see a table listing every column that contains missing values. Each row reports that column's missing-value count.

Missing the expected setup check?

Confirm that you sent the prompt in a chat inside Automotive Insights Lab.
Confirm that automobile.csv appears in the project sources.
Resume in the same project later if ChatGPT reports a file-upload or data-analysis limit.

Ask for help: Help me fix my Automotive Insights Lab setup check.

You've got the foundation in place. Next, you'll test a vague prompt and see exactly which evidence details it leaves behind.

Compare a Vague Prompt

Your ChatGPT Project can already read automobile.csv and report its structure. You now need a baseline that shows how much direction the model gets from a short request.

A vague prompt can produce a plausible claim while hiding the method behind it. This step lets you measure that weakness before improving the request.

In this step, get ready to:

Generate an unchanged one-sentence baseline with a vague prompt.
Score the baseline with a four-row evidence checklist.
Save the baseline to the project sources for later comparison.

Send the vague baseline prompt

This request deliberately leaves the analysis method open. The response gives you a baseline that you can inspect before adding stronger requirements.

Continue in the Automotive Insights Lab project chat from earlier.

Before you send the request, do you expect a broad prompt to show where its evidence came from?

Send a deliberately broad request by pasting this exact prompt into the ChatGPT message box:

Using this dataset, give me one sentence about the automobile industry.

What does this prompt control?

The phrase Using this dataset points ChatGPT toward the uploaded file.
The phrase one sentence controls the response length.
The prompt provides no requirement for named columns or calculations.
The prompt provides no requirement for missing-value handling or a usable record count.

Keep the response unchanged as your first-pass baseline.

You should see a one-sentence claim about the automobile industry. Its broad wording is the intended obstacle.

Did the response miss the requested length?

Send the exact prompt one more time if the response contains more than one sentence.
Keep the first one-sentence response unchanged when it appears.
Wait until access resets if ChatGPT reports a usage limit.

Ask for help: Help me understand why my baseline response does not follow the one-sentence request.

Score the baseline against evidence

An evidence checklist turns a general impression into four visible tests. Score only what the unchanged response actually contains.

Create the four-row checklist in your next message by pasting this template:

Historical dataset scope is stated: Present or Absent
Exact columns are named: Present or Absent
Missing-value handling and usable record count are stated: Present or Absent
A calculation or table supports the claim: Present or Absent

What does this checklist test?

Historical dataset scope checks whether the response identifies the data as historical.
Exact columns checks whether the response names fields from automobile.csv.
Missing-value handling checks whether the response explains its filtering and usable record count.
Calculation or table checks whether visible evidence supports the claim.

Replace every Present or Absent with one label based only on the unchanged response.
Send the completed checklist in the same project chat.

You should now see four scored rows below the baseline. At least one row should be marked Absent.

Unsure how to score a row?

Mark a requirement Present only when the response states that evidence directly.
Mark a requirement Absent when you would need to infer or assume it.

Ask for help: Help me score this baseline without adding evidence that the response does not state.

Save the baseline for comparison

A saved baseline gives your later comparison a fixed reference. Preserving the weak result keeps the prompt improvement visible.

Open the response menu for the unchanged one-sentence baseline.
Select Save to project or Add to project sources, depending on the label shown.
Enter First-pass baseline when a name is requested.

Before you check the project sources, do you expect the saved copy to preserve the same evidence gaps as the original response?

Check the Automotive Insights Lab project sources for First-pass baseline when naming was available.
Return to the project chat from earlier.

You should still see the setup-check response above the unchanged one-sentence baseline. You should also see the completed four-row checklist with at least one Absent result.

Your weak baseline is now preserved with a visible evidence score. Next, you will turn that broad request into a structured analysis brief.

Create an Analysis Brief

Your saved baseline now exposes the cost of a vague request. Its broad claim gives you no visible method to inspect.

A structured prompt gives ChatGPT the missing context.

Its constraints define success criteria. Its fixed table creates an output contract for every finding.

In this step, get ready to:

Generate three evidence-backed findings from automobile.csv.
Add a missing-values audit beneath the unchanged findings table.
Compare the improved result with the baseline evidence checklist.

Send the Structured Analysis Prompt

The baseline left ChatGPT to decide the scope and evidence standard. This prompt specifies both before the analysis begins.

Send the structured analysis prompt below in the project chat from earlier:

# Context
The uploaded UCI Automobile dataset contains historical vehicle specifications, prices, fuel economy values, and insurance-risk information. It is not current market data.

# Task
Identify three useful relationships among price, horsepower, city-mpg, and highway-mpg.

# Constraints
Use only automobile.csv.
Exclude missing values separately for each calculation.
State the number of records used for each result.
Do not claim that correlation proves causation.
Do not generalize the findings to today's automobile market.

# Output
Create a table with these columns:
Finding | Dataset columns | Calculation method | Numeric result | Records used | Limitation
Use plain language suitable for someone new to data analysis.

What Does This Prompt Control?

The # Context section bounds the findings to a historical dataset.
The # Task section names the four columns that the analysis must connect.
The # Constraints section requires separate cleaning for every calculation.
The correlation guardrail keeps an association from becoming an unsupported causal claim.
The # Output section gives every claim the same evidence fields.

Count the body rows in the findings table.
Inspect every row for named columns, a calculation method, a numeric result, a record count, and a limitation.

You should see exactly three findings with a separate cleaned record count for each calculation. That is the key upgrade: every claim now carries a visible evidence trail.

Do Not See Three Rows?

Resend the complete prompt if the response uses prose instead of the requested table.
Check that automobile.csv still appears in the project sources if the response cannot access the data.

Ask for help: My structured prompt did not produce the required three-row findings table. Can you help me compare the response with the prompt requirements?

Audit Missing Values

A missing-values audit checks the data quality behind the record counts. Keeping the original table unchanged preserves a stable comparison point.

Send this follow-up prompt beneath the findings table:

Add a short audit below the table. List the missing-value count for price and horsepower, then state whether city-mpg and highway-mpg contain missing values. Keep the original findings table unchanged.

What Does This Audit Check?

The audit exposes the missing-value counts behind the calculations that use price or horsepower.
The separate fuel-economy check confirms whether city-mpg or highway-mpg requires cleaning.
The instruction to preserve the table stops the audit from silently changing the findings you already reviewed.

Read the audit beneath the unchanged findings table.
Compare its missing-value statements with the record counts used in the relevant findings.

The audit should report numeric missing-value counts for price and horsepower. It should state that city-mpg and highway-mpg contain no missing values.

You have now pinned the findings table to a visible data-quality check.

Did the Table Change?

Resend only the audit prompt if ChatGPT replaces the original table.
Ask ChatGPT to append the audit beneath the previous table if the audit appears on its own.

Ask for help: ChatGPT changed my findings table while adding the missing-values audit. Can you help me preserve the original table and append only the audit?

Compare the Improved Result

A before-and-after comparison shows which prompt details created visible evidence. The same four requirements keep the comparison fair.

Predict how many of the four evidence requirements the improved result now satisfies.
Scroll back to the baseline evidence checklist from the previous step.
Copy its four checklist items.
Paste the checklist into the message composer beneath the improved result.
Change each status to Present only when the improved result visibly supplies that evidence.
Add one sentence naming the structured-prompt detail that produced the largest improvement.
Send the completed checklist and sentence.

Your chat should now show the updated four-item checklist beneath the findings table and audit. A final sentence should identify the prompt detail that made the clearest difference.

Unsure How to Score an Item?

Mark an item Present only when you can point to the supporting text or table field.
Leave an item Absent when the response implies the evidence without displaying it.

Ask for help: Can you help me score my improved automobile analysis against the four evidence requirements without assuming details that are not visible?

Your analysis brief now connects each finding to a method, record count, and limitation. Next, you will turn that evidence into a chart and audit the calculation behind it.

Build and Audit Your Chart

Your structured findings table gives every result a visible method. It also records the cleaned row count for each calculation.

A chart can look convincing even when its filtering and calculation are unclear.

In this step, you'll use ChatGPT to create a horsepower-versus-price scatter plot. You'll audit the chart against the uploaded file.

In this step, get ready to:

Generate a labelled price-versus-horsepower scatter plot.
Audit the chart calculation against automobile.csv.
Record the evidence boundary for the result.

Generate the specific chart

A scatter plot places each vehicle as one point based on two numeric values. Pearson correlation summarises the direction and strength of the linear relationship.

Rows with missing values cannot support this calculation. The prompt below excludes them before reporting the statistic.

Create the chart in the project chat from earlier by pasting this prompt into the message box:

Using automobile.csv, create a scatter plot with horsepower on the x-axis and price on the y-axis.
Exclude rows where horsepower or price is missing.
Title the chart "Horsepower vs Price in the UCI Automobile Dataset" and label both axes.
Below the chart, report the cleaned record count and the Pearson correlation.
Do not claim that horsepower causes price.

What does this prompt control?

The axis instructions place horsepower horizontally. They place price vertically.
The filtering rule removes any row missing either required value.
The reporting rule exposes the cleaned record count plus the calculated relationship.
The final guardrail keeps the interpretation within what observational data can support.

Send the prompt.

You'll see a scatter plot titled Horsepower vs Price in the UCI Automobile Dataset with labelled axes. The response also reports its cleaned record count and Pearson correlation.

You now have a chart you can inspect instead of a claim you must take on trust.

Chart missing or incomplete?

Confirm that automobile.csv still appears in the sources for Automotive Insights Lab.
Resend the full prompt above if the response gives text without a scatter plot.
Leave the project unchanged if ChatGPT reports a data-analysis limit. Resume in the same chat when data analysis becomes available again.

Get help with the chart: Help me work out why ChatGPT did not create the requested scatter plot from automobile.csv.

Audit the calculation

A calculation audit independently checks the cleaned rows plus the statistic behind a chart.

Before you send this, do you expect the recalculated values to match the chart exactly?

Recalculate the chart evidence directly from the project file by pasting this prompt into the message box:

Recalculate the cleaned record count and Pearson correlation directly from the uploaded file.
Return a three-row audit table with:
1. Rows excluded
2. Missing fields responsible
3. Whether the evidence supports association, causation, or neither

What does this audit check?

The opening instruction recalculates the chart statistics from the uploaded file.
The Rows excluded row exposes the effect of filtering.
The Missing fields responsible row identifies why those records were removed.
The final row classifies the evidence without turning association into a causal claim.

Send the audit prompt.

You'll see a three-row audit table. It also reports the recalculated cleaned record count and Pearson correlation.

Audit details missing?

Check that the response contains exactly three audit rows.
Resend the full prompt if the response omits the recalculated count or correlation.

Get help with the audit: Help me check why my chart audit is missing the requested calculation details.

Record the evidence boundary

Correlation supports a measured association between two variables. An evidence boundary prevents that association from becoming an unsupported causal or current-market claim.

Your first sentence describes the observed relationship shown by the chart and correlation. Your second sentence states that this historical observational dataset cannot establish a current-market or causal conclusion.

Add your two-sentence evidence-boundary note below the audit in the same project chat.

You'll now see the chart, calculation audit, and evidence-boundary note together in the project chat.

Before you compare the two outputs, do you expect the chart and audit to agree on the cleaned record count?

Compare the cleaned record count below the chart with the recalculated count in the audit.
Compare the chart's Pearson correlation with the recalculated correlation in the audit.
Confirm that the note contains one observed-relationship sentence plus one limitation sentence.

You should see the same cleaned record count in the chart and audit. You should also see a matching Pearson correlation without any claim that horsepower causes price.

Chart and audit do not match?

Compare the missing fields named in the audit with the chart's filtering rule.
Check that both calculations exclude every row missing either horsepower or price.

Get help comparing the results: Help me identify why my chart and audit report different cleaned counts or correlations.

That's the chart checked against its underlying calculation. Next, you'll turn your verified findings into a reusable historical automotive briefing.

Save a Reusable Briefing

Your checked scatter plot now connects the visual result to a calculation audit. Separate analyses remain difficult to reuse as one decision-ready deliverable.

A fresh chat inside ChatGPT Projects tests whether automobile.csv remains available through project context. You will use that context to create a concise historical briefing that stays within the evidence.

In this step, get ready to:

Confirm that automobile.csv remains available in a fresh project chat.
Create a historical automotive briefing with five required sections.
Save a tightened final briefing to the project sources.

Confirm project source reuse

A reusable project carries its files into future chats. This check proves that your analysis can continue without another upload.

Start a new chat from inside Automotive Insights Lab.
Do not upload automobile.csv again.

Before you send the check, predict whether the fresh chat can name the source file without another upload.

Test source reuse by sending this prompt:

Confirm that automobile.csv is available from this project's sources. State only the filename and its row count.

How Project Context Is Tested

The prompt asks for only the filename and row count. A correct response shows that the fresh chat can access the project source.

The narrow output also makes the result easy to verify. Extra analysis would hide the simple context check.

You should see only automobile.csv with a row count of 205.

Project Source Missing

Check that the fresh chat belongs to Automotive Insights Lab.
Return to Automotive Insights Lab before starting another fresh chat.
Wait until access resumes if ChatGPT reports a file or data-analysis limit.

Ask for help: Why can my fresh ChatGPT project chat not access automobile.csv from the project sources?

Generate the five-section briefing

A reusable briefing combines your checked findings into one bounded summary. Its structure keeps calculations visible while separating them from interpretation.

Create the first version of the briefing by sending this prompt:

Create a concise historical automotive data briefing using automobile.csv and the analysis already completed in this project.

Include these sections:
1. Scope and dataset age
2. Executive summary with three bullets
3. Three evidence-backed findings
4. Chart interpretation
5. Limitations and one sensible next analysis

For every numeric claim, name the columns used and the cleaned record count.
Use plain language and keep the briefing under 500 words.
Do not present the data as a description of today's automobile market.

How the Briefing Prompt Works

The five-section output contract gives each part of the briefing a clear purpose. The word limit keeps the result concise.

The evidence rules require every numeric claim to name its columns and cleaned record count. The scope rule keeps the historical data separate from claims about today's market.

You should see all five requested sections. The executive summary should contain three bullets.

Missing Briefing Sections

Check that you sent the complete prompt inside the fresh Automotive Insights Lab chat.
Send the complete prompt again if the response omits a required section.
Ask ChatGPT to restore the named columns or cleaned record count if either detail is missing from a numeric claim.

Ask for help: Help me identify which requirements are missing from my five-section automotive briefing.

Tighten and save the final briefing

Words such as high or strong can hide the size of a result. A final editing pass replaces that language with measured values or neutral descriptions.

Tighten the briefing by sending this follow-up prompt:

Replace vague words such as high, low, strong, and significant with measured values or neutral wording. Keep the five-section structure and stay under 500 words.

What the Editing Pass Changes

The follow-up preserves the briefing structure while improving the wording. Measured values make each conclusion easier to inspect.

Neutral wording protects claims that need careful interpretation. The finished result stays concise without losing its evidence boundary.

Before you inspect the revision, predict whether every numeric claim now carries both its dataset columns and cleaned record count.

Count the section headings in the revised briefing.
Count the bullets in the executive summary.
Inspect every numeric claim for named dataset columns plus a cleaned record count.
Check the briefing against the 500-word limit.
Scan the briefing for vague words that need measured or neutral replacements.
Read the limitations section for a clear historical-data boundary.

Final Briefing Check

You should find five section headings. The executive summary should contain three bullets.

Every numeric claim should name its dataset columns plus its cleaned record count. The complete briefing should remain under 500 words.

The findings should use measured or neutral wording. The limitations should avoid presenting the historical dataset as today's automobile market.

Open the final response menu.
Select Save to project or Add to project sources according to the label shown.
Confirm that the finished briefing appears in the Automotive Insights Lab project sources.

You should see the final historical automotive briefing saved with the project's other source material.

Briefing Not Saved

Check the response menu on the final revised response.
Select the available project-saving option before leaving the chat.
Repeat the save action if the briefing does not appear in the project sources.

Ask for help: Help me save my final ChatGPT response to the Automotive Insights Lab project sources.

You did it. Automotive Insights Lab now contains a reusable historical briefing with checked calculations and clear evidence boundaries.

Secret mission

Compare Body Styles in a Fresh Chat

Reuse automobile.csv in a fresh Automotive Insights Lab chat without uploading it again. Build a body-style table with counts and medians. Turn its clearest measured trade-off into a historical recommendation with a current-market limitation.

Clean Up Your Resources

Choose whether to keep the lab ready for more analysis, close the browser for now, or remove every copy you created. This project costs $0 on ChatGPT Free, although file-upload and data-analysis limits still apply.

Resources you used:

ChatGPT Project named Automotive Insights Lab.
Project source automobile.csv plus the project instructions.
Saved analyses inside the project, including the baseline, findings table, audits, chart, final briefing, and body-style comparison.
Local dataset file at Downloads/automobile.csv on your Mac.

Keep everything running

Keep this option if you want to reuse the lab for more historical automotive questions. Your project adds no separate project fee on ChatGPT Free.

Leave Automotive Insights Lab in the ChatGPT sidebar.
Keep Downloads/automobile.csv on your Mac as a local copy.
Reuse the saved analyses as evidence for future historical comparisons.

Pause - I'll come back to this later

Pausing means stepping away from the browser. Your signed-in project remains available with its files, chats, instructions, chart, and saved sources.

Close the browser tab containing ChatGPT.
Keep Automotive Insights Lab unchanged in the ChatGPT sidebar.
Return later by signing in with the same ChatGPT account.

Delete - I don't want to use this again

Deleting your lab is permanent. The steps below clear each stored copy in a controlled order.

Locate Automotive Insights Lab in the ChatGPT sidebar.
Open the more options menu marked •••.
Select Delete project.
Confirm the deletion when ChatGPT asks.

That is the online workspace cleared. You should no longer see Automotive Insights Lab in the sidebar.

A file separately saved in Library survives project deletion. The next check catches that extra copy.

Open Library in ChatGPT.
Look for automobile.csv.
Delete that Library copy separately if it appears.

After this check, Library should contain no separate copy of automobile.csv.

Project deletion leaves the local CSV in your Mac Downloads folder. Remove that copy if you no longer want the dataset.

Click the Finder icon in your Mac Dock.
Select Downloads in the Finder sidebar.
Locate automobile.csv.
Move automobile.csv to the Trash.
Empty the Trash to remove the local file permanently.
Return to Downloads.

You should no longer see automobile.csv in your Downloads folder.

Nice Work!

Great work! Your Automotive Insights Lab now holds a checked historical briefing inside ChatGPT Projects.

You learned how to:

Built a reusable analysis workspace around automobile.csv that preserves the complete project context.
Used a structured analysis prompt to turn a vague baseline into a three-row findings table with visible missing-value handling. Every result names its usable record count.
Created a price-versus-horsepower chart with a matching calculation audit that separates correlation from causation.
Secret Mission: Extended the project in a fresh chat with a body-style comparison covering vehicle counts plus price and fuel-economy medians. The saved recommendation stays within the historical evidence boundary.

Ready to quiz yourself?