Comparing tools for a task, as of a given date
This lesson is still being tested: it improves with learners' feedback. If something is unclear, it is the lesson that needs fixing, not you.
Follow this lesson with Claude Code
With the course companion installed, open Claude Code in your working folder and type:
/formation-ia:suivre m09-l02 enIt reads the up-to-date lesson on this site and guides you step by step, without changing anything without your consent. Install the companion. Claude Code is not included in the free plan: an eligible Claude subscription, or a Console account billed by usage for the terminal.
By the end of this lesson, you will be able to
- compare two tools on a specific task, through a real trial;
- fill in an honest comparison checklist: quality, mistakes, capabilities, time, cost, data;
- date your comparison, and know when to repeat it;
- choose the tool for a role, with a written reason.
Why it matters to you
“What is the best AI?” has no answer. “Which tool does this task best, in my plan, today?” does, and you can find it in twenty minutes. Online rankings age quickly, and they do not test your task. Your own dated trial is worth more to you than any opinion.
Before you start
| You need | Two assistants on free accounts (for example Claude and ChatGPT), the book club exercise notes and the comparison checklist (resources) |
| Cost | €0 |
| Duration | 35 minutes |
| Reminder | For a real document, first look at the privacy settings of each tool (module 1, lesson 3) |
Understand it in 3 minutes
An honest comparison follows four rules:
| Rule | Why |
|---|---|
| The same request, word for word | Otherwise, you are comparing two requests |
| The same files, nothing more | Otherwise, one has more information than the other |
| New conversations, with no memory | Otherwise, the history or the memory on one side changes the result. Claude: an incognito chat (ghost icon), or memory turned off for this chat in the ”+” menu. ChatGPT: a temporary chat set to “Unpersonalized” (no personalization). Either way, choose before you send the first message: the setting cannot be changed afterwards |
| A date | Plans and models change: the comparison is valid as of its date |
What you compare, in the checklist:
- whether each criterion of the request is met;
- mistakes and made-up details;
- what the tool could do in your plan (read your files, search the web, read a format);
- the time spent, corrections included;
- the limits reached during the trial, and where your data went.
What the official pages said on 24/10/2026 (check as of your date):
| Tool | Free plan | First paid plan, price shown in France |
|---|---|---|
| Claude | Yes | Pro: €18 per month excl. VAT |
| ChatGPT | Yes, with limited access to Codex (“Accès limité à Codex” on the French page) | Go: €8 per month (may include ads); Plus: €23 per month |
| Codex | Included in the ChatGPT plans, including Free (for Free and Go, in the desktop app according to the Codex plans page, 25/10/2026) | The one in the ChatGPT plan |
The ChatGPT page does not say whether its prices exclude VAT: check the amount when you pay.
One sign that everything moves fast: the Codex plans page announced the retirement of the GPT-5.5 model from Codex on October 14, 2026, when signed in with a ChatGPT account (not with an API key).
Do it together
1Step 1: Prepare the trial
Open the resource Exercise notes: the book club and the resource Comparison checklist for two tools. Prepare this request (it is the complete brief from lesson 3):
CONTEXT: I am part of a book club. I have to write the invitation to our back-to-school evening, to send to the members.
GOAL: an invitation text ready to paste into an email.
TO READ: the notes below. Nothing else.
ALREADY DECIDED: everything written in the notes as "agreed" or decided.
IT WORKED IF:
1. the date, the time and the place are exactly the ones agreed in the notes;
2. the text is 120 words at most, in a warm tone;
3. nothing is made up: information that is missing or not yet decided is written [to be completed];
4. the text says how to register and what to bring.
DO NOT: do not edit or create any file.
FORMAT OF THE REPLY: 1) the invitation; 2) the list of information taken from the notes; 3) the points to check or to decide before sending.
NOTES:
[paste the club's notes here]
2Step 2: Run the two trials
Open a new conversation, with no memory (see the table above) in each of the two assistants, paste the same request, and send. Do not correct anything during the trial: you are comparing the first answers.
3Step 3: Fill in the checklist
For each tool, check the four criteria one by one, against the notes: date, time, place, [to be completed] where needed, registration, what to bring, 120 words at most. Write down the mistakes and the made-up details. Then fill in the rest of the checklist.
In our trials of 24/10/2026, one Claude session that had written the invitation with the complete brief copied “Saturday, November 22” without noticing that this day is a Sunday in 2026, whereas another session, asked to review that result, did notice it.
Those two trials did not have the same request: for your comparison, keep the same request, word for word. And compare criterion by criterion, not “by feel”: an answer can look good and let a mistake through.
4Step 4: Decide, and date it
At the bottom of the checklist, write: the tool chosen for this task, the reason, and a date to repeat the trial.
Your turn
- Choose one real task from your project, one that comes up often.
- Write the request once, with its criteria (you can wait for lesson 3 to perfect it).
- Run the trial in two tools, in new conversations, with documents containing nothing sensitive.
- Fill in the checklist, choose, date it. File the checklist in your journal.
Resources to download
- Exercise notes: the book club: invented meeting notes, with a date that changed and two pieces of missing information.
- Comparison checklist for two tools: the criteria, the choice, the date to repeat the trial.
Check your work
It worked if:
- both trials received the same request and the same notes, in new conversations;
- each criterion is checked in both columns, against the notes;
- your choice has a written reason, and a date to revisit it.
Common mistakes:
- comparing on a general impression, without checking the criteria one by one;
- correcting one tool during the trial, and not the other;
- relying on an online ranking for a task it did not test;
- keeping a comparison for months without repeating it.
Keep it and pick it up later
- Keep your checklists in your journal: they say why you use a given tool for a given task.
- At the monthly upkeep (module 5, lesson 7), look at the “to be checked again” dates.
Going further
optional
- The Choosing your tool page of this course summarizes the tools it talks about, with their checking dates.
New words
3 words explained
| Word | What it means |
|---|---|
| Comparison as of a date | A real, dated trial, valid for your task, your plan and that day |
| Checklist | The table of criteria, filled in for each tool |
| Criterion | A verifiable condition of the request: the exact date, under 120 words, [to be completed] where needed |
Check your understanding
3 questions, no grade. An explanation appears after each answer.
What's next?
Lesson 3 · The handoff brief: the request that lets another tool, or a new conversation, do the work without asking you everything again.
Module 9:
✏️ Do the exercisesMy space: your progress, your exercises and your last lesson.
The trials of two Claude sessions on the club notes are in the test report (T44 and T45). To be checked: the comparison between Claude and ChatGPT in step 2 has not been tested.
Sources and checks
3 sources
Date, version, sources Written on 24/10/2026. Checks of 24/10/2026:
- Claude plans and Pro price in France: claude.com/pricing;
- ChatGPT plans and prices in France, “Limited access to Codex” in Free: chatgpt.com/pricing;
- Codex included in the ChatGPT plans, including Free, retirement of the GPT-5.5 model on October 14, 2026 (when signed in with a ChatGPT account): learn.chatgpt.com/docs/pricing.
Sources checked: