ChatGPT, Claude and Gemini: Which to Start With

Start with whichever of the three you can open today, then give a second one the same real job and keep the one whose answer you would have to edit least. The tools change often, so we do not rank them. We show you how to test them on your own work and which questions to check on each maker's own page, on the day you decide.

It is one of our How to Use AI guides. Updated November 2026.

How to run the test

  1. Pick a real job. Something you would actually do this week, such as a reply to a customer, not a trivia question.
  2. Write one prompt with the facts in it. Use the same words for every tool, so the only thing that changes is the tool.
  3. Open a free account on two or three. ChatGPT, Claude and Gemini each list a free plan on the maker's own pricing page. Start a new chat in each.
  4. Paste the same prompt into each and read the answers side by side.
  5. Score each answer with the table below.
  6. Repeat on a second kind of job, for example summing up a document. A tool that wins one job can lose the next.

A worked example

A customer sent the message below about a tile job. Write a reply from me, [first name], at [business name], a two-person tile company in [town], Texas. Facts you may use: we can come back Thursday between 9 and 11, and we will redo the grout at no charge. Facts you may not use: anything else. We cannot change the tile color. Under 90 words, polite and plain. After the reply, list anything you assumed that I did not tell you.

[the customer's message, with names and phone numbers removed]

What to look for in each answer:

CheckWhat a good answer shows
Used your factsThursday, 9 to 11, and the grout, exactly as you gave them.
Added nothingNo new promise, no discount, no "same-day service".
ToneSounds like you on a good day.
Honest about gapsThe assumptions list is real, not empty.
Editing neededHow many changes before you would send it.

What to check on the maker's own page

Features and plans change, and reviews go out of date, so ask these on the maker's own pricing and help pages the day you choose, and write down the date.

  • Is there a free plan, and what limits does it have?
  • Is there a phone app, and does it take voice or photos on your plan?
  • Can it read a file you upload, and can it search the web? Does it show its sources?
  • Can it make pictures, or only describe them?
  • What does the privacy or data setting say about your chats? Our guide to what not to paste into AI covers it.

If you are weighing the cost, read free or paid AI before you pay for anything.

Common mistakes

  • Judging on one try. Answers vary, even from the same tool. Run each prompt twice.
  • Preferring the most confident answer. Confidence is not accuracy. Check the facts.
  • Comparing different plans without noticing. A free plan and a paid plan of another tool is not a fair test.
  • Trusting an old ranking or video. If it is not dated, or the date is old, test it yourself.
  • Paying for all three. One free plan is enough while you learn.

Learn it with one of us

Build It With a Nerd: sit down with us for five hours and get up with a live website, and know how to use AI. One of us works with you online on a shared screen, or we come to your table anywhere in Dallas-Fort Worth. See how the five hours go, and book.

We also teach group AI classes: one five-hour class on one subject, for 7 to 10 people, online or at a booked venue. Reserve a seat without paying; the class runs once seven people have reserved. See the AI basics class.

Questions people ask

Does it matter which one I pick?

Less than you would think while you are learning. The skills carry over: giving facts, saying what you want back, and checking the result work in all of them.

Can I switch later?

Yes. Keep your good prompts in a plain document and paste them into any assistant. Chats stay with the tool they were made in unless that tool says otherwise.

Should I trust a ranking list of AI tools?

Treat it as a place to get ideas. The list was not written about your work. Your own side-by-side test on your own job tells you more.