Gemini vs ChatGPT: We Gave Both Free Versions the Same 4 Tests

Search "Gemini vs ChatGPT" and you'll get benchmark tables, parameter counts, and pricing pages — but almost nobody feeds the same four everyday tasks to both free versions and shows you every answer side by side. We did exactly that on October 27, 2026: a classic logic puzzle, a writing brief with one strict rule baked in, a breaking-news question that most of the internet still gets wrong, and a messy list that needed to become a clean table. Every answer below has a screenshot behind it, including the one where the test we designed to embarrass one of them backfired.

Short answer: ChatGPT free won by a nose — 2 wins, 1 loss, 1 tie. On the logic puzzle both were correct, but Gemini's 6-step answer tracked the water level after every move, which is easier to follow. On the writing brief (100 words, exactly one playful metaphor), ChatGPT delivered 97 words with exactly one metaphor; Gemini delivered 99 words but sneaked in two figurative images. On our fresh-news question — Google's recent rename of NotebookLM — both answered correctly (Gemini Notebook, notebook.google.com, old address redirects), beating most human-written tutorials that still use the old name. On the table-cleanup test, both produced a perfect table. Neither bot hit a usage wall during our 8-answer session.

How we ran the test

On October 27, 2026, we opened both chatbots in separate browser tabs on the same day, same location, same network:

• Gemini free at gemini.google.com, signed in with a Google account, left on the default Auto model tier — the model picker also showed 3.1 Pro and a Thinking level setting, but we didn't touch them, because Auto is what a regular free user gets by default.
• ChatGPT free at chatgpt.com, signed in with a free account, no model changes — the composer showed a "Think" toggle which we left off.

Each question was asked in a fresh conversation on both sides, using identical wording. We asked four kinds of tasks, each probing a different skill people actually use AI for:

1. Reasoning: the classic 3-liter / 5-liter jug puzzle — measure exactly 4 liters, shortest set of steps.
2. Writing with a constraint: a 100-word product description for a handmade ceramic mug, warm and friendly tone, exactly one playful metaphor. The rule is the test.
3. Fresh knowledge: "Google recently renamed NotebookLM. What is its new name, what is its website address now, and what happened to the old address?" We knew the verified answer because we tested the rename ourselves for our NotebookLM free account guide — most tutorials online still use the old name and URL.
4. Formatting: turn a deliberately messy one-line price list into a clean table with three columns.

We graded only what we could see: correctness, instruction-following, and clarity. We didn't count tokens or measure response speed — those numbers wouldn't mean much across two different sessions anyway.

Test 1: The water-jug logic puzzle

The prompt: "I have a 3-liter jug and a 5-liter jug, and unlimited water. How do I measure exactly 4 liters? Give me the shortest set of steps."

Both got the right answer. The classic solution — fill the 5-liter jug, pour 3 liters off into the small jug, leaving 2; empty the small jug; move the 2 liters across; refill the big jug and top off the small one (taking exactly 1 liter) — leaves exactly 4 liters in the 5-liter jug.

ChatGPT compressed the whole thing into four numbered steps and bolded the result ("Exactly 4L remain in the 5L jug"). Correct and tight — but a beginner has to mentally track the water levels between steps.

Gemini spread the same solution over six steps, and under every single step it printed the current contents of both jugs ("Current contents: 5L jug = 4L, 3L jug = 3L"). Same correct destination, but you can verify each move with your eyes instead of your memory.

Verdict: both correct; Gemini's format is friendlier for beginners. If you learn by following along, the running totals matter. Point to Gemini.

Test 2: The 100-word brief with one strict rule

The prompt: "Write a 100-word product description for a handmade ceramic mug. Tone: warm and friendly. Include exactly one playful metaphor."

This is the test that separates "writes nicely" from "does what it was told." We counted the words ourselves.

ChatGPT delivered 97 words — a hair under the target but well inside normal tolerance — and exactly one metaphor: it called the mug "a little hug for your hands." One image, used once, warm and friendly. Brief followed.

Gemini delivered 99 words — actually closer to 100 — but it couldn't resist a second image. It turned every sip into "a warm hug" and then also compared the handle to "a soft woolen mitten on a chilly morning" — which it even put in bold, as if proud of breaking the rule. That's a metaphor plus a simile where we asked for exactly one playful metaphor.

Verdict: ChatGPT wins on instruction-following; Gemini writes arguably prettier prose. If your use case is creative writing with no constraints, Gemini's copy is lovely. If you need AI to obey a spec — word counts, keyword rules, brand voice limits — this one data point favors ChatGPT.

Test 3: The fresh-news question (our trap question)

The prompt: "Google recently renamed NotebookLM. What is its new name, what is its website address now, and what happened to the old address?"

We designed this question expecting to embarrass at least one chatbot, because the internet is full of stale NotebookLM tutorials. We were wrong — in a good way.

Both passed. ChatGPT answered that NotebookLM is now Gemini Notebook at notebook.google.com, that the old address automatically redirects so bookmarks keep working, and it even dated the announcement — July 16, 2026 — with citation chips pointing at Google's blog.

Gemini also said Gemini Notebook at notebook.google.com with permanent 301 redirects from the old address — all correct — though its citation chips pointed at sources labeled "Kortex" and "NotebookLM Tools," which are less authoritative-looking than Google's own blog.

Verdict: both correct — a tie that flatters both. Two notes: first, if two AI chatbots both know about the rename but most blogs don't, that tells you something about how fresh blog content stays. Second, check the sources behind an AI answer, not just the answer. Our full breakdown of the rename, with the quota numbers that changed at the same time, is in our Gemini Notebook free-account test and our NotebookLM pricing walkthrough.

Test 4: Turning a messy list into a table

The prompt: turn "Basic 1.99 100GB, Standard 2.99 200GB, Plus 4.99 400GB, Pro 19.99 5TB, Premium 2TB plus AI bundle" into a clean table with columns Plan, Price per month, and Storage.

Both produced a correct, well-formatted table — five rows, three columns, prices formatted as dollars. The subtle part was Premium: we never gave a price for it, and both honestly wrote "Not specified" instead of inventing one. That's exactly the behavior you want from an AI handling data.

Gemini added a small touch: it asked afterwards whether we wanted any formatting adjusted. ChatGPT just delivered. Verdict: a tie. Both correct, both honest about the missing price.

The scorecard

What this means if you're choosing one

Honest caveat first: four questions is a taste test, not a benchmark. Different prompts on different days can absolutely flip the writing test — AI outputs aren't deterministic. But the pattern we saw matches what the two products are optimized for:

• Choose ChatGPT free if you give AI precise instructions — content with word counts, formats, or rules it must not break. In our one constrained writing test, it was the only one that obeyed.
• Choose Gemini free if you're a beginner or you like answers that show their work — the step-by-step format with running state was the most teachable answer in our whole test, and the default Auto tier requires zero configuration.
• You can also just use both — that's what we do. Both free tiers answered everything in our session without a single usage-wall popup. We can't tell you a daily question limit because we never hit one, and we won't make up a number.

And if you're curious how other head-to-heads came out when we ran them on free accounts: our ChatGPT vs Claude test ended very differently — Claude's free tier never answered a single question because of an ID verification wall — and our DeepSeek vs ChatGPT vs Gemini three-way comparison has the full standings for all three.

Frequently asked questions

Which is better for free users, Gemini or ChatGPT?

On our four identical tests on October 27, 2026, ChatGPT free won by a nose — it was the only one that obeyed the "exactly one playful metaphor" rule, and its fresh-news answer had the most authoritative sources. Gemini free matched it on logic and table formatting, and its step-by-step format was friendlier for beginners. Both answered all 4 questions with no usage-wall popups during our session.

Did both AI chatbots know that NotebookLM was renamed?

Yes. Both Gemini free and ChatGPT free correctly said NotebookLM is now Gemini Notebook at notebook.google.com, with the old address redirecting. ChatGPT even cited Google's announcement date of July 16, 2026. Most human-written tutorials online still use the old name, so both bots were more current than most blogs.

Do Gemini and ChatGPT free versions have daily limits?

During our 4-question session on October 27, 2026, neither chatbot showed a usage-limit popup — all 8 answers generated normally. We can't say how many questions a free account allows per day, because we didn't hit any wall in this test, and we won't invent a number.

Which AI follows instructions more precisely?

In our writing test, ChatGPT free followed the brief more precisely: we asked for a 100-word product description with exactly one playful metaphor, and it delivered 97 words with exactly one. Gemini free delivered 99 words but included two figurative images — a metaphor plus a simile — which broke the "exactly one" rule. If you use AI for content with strict constraints, that difference matters.