I Used GPT-5, Claude 4, and Gemini Ultra 2 Every Day for a Month. Here’s What Nobody Tells You.

Comparison posts are everywhere right now. Most of them read like spec sheets. Context window this. Parameter count that. Benchmark scores on tests that don’t reflect anything you’d actually ask an AI to do.

So I did something different.

For the past month, I forced myself to use all three — GPT-5, Claude 4, and Gemini Ultra 2 — as my daily drivers. Rotating between them for real work. Writing. Research. Brainstorming. Summarizing long documents. Answering emails. The stuff you and I actually do on a Tuesday afternoon.

No synthetic benchmarks. No paid partnerships. Just honest notes after thirty days of genuine use.

Here’s what I found.

What I Tested (And What I Didn’t)

Before anyone asks in the comments — yes, I used paid versions of all three. ChatGPT Plus with GPT-5 access. Claude Pro with Claude 4. Google One AI Premium with Gemini Ultra 2.

I tested across these tasks:

• Writing blog posts and newsletter drafts
• Summarizing long PDFs and research papers
• Brainstorming article ideas and headlines
• Responding to emails (tone adjustment, clarity)
• Coding help (basic Python scripts, nothing advanced)
• General research questions with web browsing
• Long, meandering conversations to test memory

What I didn’t test: heavy code generation, mathematical proofs, or API performance. This is a consumer-focused comparison for people who use these tools to get work done.

Round 1: Writing Quality

If you write anything for a living, this is probably the section you care about most.

GPT-5 has gotten noticeably better at sounding human. The stiff, adjective-heavy prose that plagued early GPT models is mostly gone. Give it a good prompt with tone instructions, and the output reads like something a competent human wrote. It still defaults to a slightly cheerful, customer-service voice if you’re vague, but you can steer it easily. For blog drafts and newsletters, it’s my go-to when I need volume fast.

Claude 4 is the better writer, but with a catch. Its prose feels more natural out of the box — fewer clunky transitions, better paragraph rhythm, more varied sentence structure. It handles nuance better than the other two. The catch? It can be too careful. Ask for a strong opinion or a persuasive argument, and Claude sometimes hedges so much the writing loses its edge. For thoughtful, balanced content, it’s excellent. For spicy takes? You’ll need to push it.

Gemini Ultra 2 surprised me. Gemini’s writing was genuinely bad in 2024 — stiff, repetitive, allergic to contractions. The 2026 version is a different animal. It’s still slightly more formal than GPT-5 or Claude, but the gap has shrunk dramatically. Where Gemini shines is factual accuracy during writing tasks. Fewer hallucinations, fewer made-up statistics, better citations when you ask for sources. If you write content where getting facts wrong is embarrassing, Gemini might be your safest bet.

Winner for writing: Claude 4 for quality. GPT-5 for speed and volume. Gemini Ultra 2 for accuracy.

Round 2: Memory and Long Conversations

Your news post touched on this, so let me go deeper with real examples.

ChatGPT (GPT-5) has the most aggressive memory implementation. After about a week of use, it started remembering my preferences without being asked. My writing style. My industry. The fact that I prefer bullet points over paragraphs in summaries. It also surfaces these memories proactively — sometimes helpfully, sometimes annoyingly. Once it said “Based on your preference for direct feedback…” before giving me a critique, and I honestly forgot I’d told it that three weeks earlier. Slightly unsettling. Mostly useful.

Claude 4 takes a different approach. Its memory works within projects — containers where you group related conversations. Inside a project, Claude remembers everything across sessions seamlessly. Outside projects, it resets. I actually prefer this. It gives me control over what gets remembered and when. For long research projects where context matters across days, Claude with projects is unmatched.

Gemini Ultra 2 ties memory to your Google account across services. This sounds cool in theory. In practice, it’s inconsistent. It remembered a search I did last week and referenced it in an email draft, which was impressive. But it also completely forgot a preference I’d stated explicitly three messages earlier in the same conversation. Google’s integration is ambitious but still buggy. When it works, it feels like the future. When it doesn’t, it’s frustrating.

Winner for memory: Claude 4 for control and reliability. GPT-5 for the “it just knows” factor. Gemini Ultra 2 for potential — once the bugs are ironed out.

Round 3: Handling Long Documents

I threw a 45-page research paper and a 30-page contract at all three. Asked for summaries, key points, and follow-up questions.

Claude 4 ate this for breakfast. Long document handling has been Claude’s strength for a while, and version 4 widened the lead. Summaries were accurate, well-structured, and hit the nuances that the other two missed. When I asked specific follow-up questions about sections buried deep in the document, Claude found the relevant passages quickly and didn’t confuse details. If you work with long PDFs regularly, this alone might justify the subscription.

GPT-5 is solid but not spectacular here. Summaries are accurate enough for casual use, but it sometimes misses subtle points or oversimplifies complex arguments. The longer the document, the more obvious the quality drop becomes. Fine for a ten-page report. Less reliable for a fifty-page one.

Gemini Ultra 2 handles long documents well when they’re Google Docs — the integration is seamless. But upload a PDF, and the experience is clunkier than the other two. Summaries are accurate but dry. Good for extracting facts. Less good for capturing tone, argument structure, or nuance.

Winner for long documents: Claude 4, by a comfortable margin.

Round 4: Web Browsing and Research

All three can browse the web now. The experience is not equal.

Gemini Ultra 2 has the advantage of Google Search integration, and it shows. Research queries return more relevant, up-to-date results. Citations are cleaner. The “double-check with Google” feature highlights which claims are verified and which aren’t — genuinely useful for research tasks where accuracy matters.

ChatGPT (GPT-5) browsing is faster than it used to be, but still occasionally pulls from outdated sources or misunderstands which information is current. It works fine for casual research. I wouldn’t trust it for anything where being wrong has consequences.

Claude 4 browsing is the weakest of the three. It works, but it’s slower, and the sources it pulls sometimes lack depth. Claude seems aware of this limitation — it often suggests I verify information independently, which I appreciate as a disclosure but also signals the feature isn’t fully baked.

Winner for research: Gemini Ultra 2. Google’s search advantage is real.

Round 5: The Little Annoyances

Every tool has quirks. After a month, here’s what bugged me about each.

GPT-5 still overuses certain phrases. “It’s worth noting that…” shows up constantly unless I explicitly tell it not to. The cheerful tone can feel forced. And despite the memory improvements, it occasionally forgets instructions I gave earlier in the same conversation and needs reminders.

Claude 4 can be frustratingly cautious. Ask it for an opinion on which tool is better for a task, and you’ll get a balanced list of pros and cons with no clear recommendation. Sometimes that’s appropriate. Sometimes I just want a straight answer, and Claude dances around it. Also, its refusal rate on borderline-safe topics is slightly higher than the others.

Gemini Ultra 2 suffers from integration awkwardness. Sometimes it drafts in Google Docs. Sometimes in the Gemini interface. Sometimes it loses connection to your Drive for no obvious reason. The potential is huge — a single AI across email, docs, search, and calendar — but the seams are still visible. You feel like you’re using three products duct-taped together, not one cohesive experience.

Round 6: Free Tier Face-Off

Not everyone wants to pay. Fair enough. Here’s how the free versions stack up in May 2026.

ChatGPT free gives you GPT-5 with usage caps, web browsing, and basic image generation. The caps are reasonable — most casual users won’t hit them daily.

Claude free offers Claude 4 with limits that reset every few hours. The free tier here is generous enough that I know several people who’ve never felt the need to upgrade.

Gemini free is solid but nudges you toward paid more aggressively than the others. Features like deeper Google integration and longer context windows sit behind the paywall. Usable, but you feel the limitations faster.

Winner for free tier: Claude 4 edges it out, with ChatGPT close behind.

So Which One Should You Actually Use?

Depends entirely on what you do.

Pick GPT-5 if you write a lot, want speed, and don’t mind occasionally reminding the AI about your preferences. It’s the best all-rounder for most people.

Pick Claude 4 if you handle long documents, run research projects across multiple days, or want the most natural-sounding writing without much tweaking. Also the best choice if privacy and controlled memory matter to you.

Pick Gemini Ultra 2 if you live in Google’s ecosystem already — Gmail, Docs, Drive, Calendar — and want an AI woven through all of it. Also the best for research-heavy tasks where source accuracy is critical.

What I Do (For What It’s Worth)

After a month, I settled into a two-tool setup.

Claude 4 handles my serious writing, long document work, and deep research projects. GPT-5 handles quick drafts, brainstorming, and anything where I need fast volume. Gemini Ultra 2 I keep around for research tasks and because the Google integration keeps improving — I expect it to catch up within the year.

Paying for two subscriptions isn’t ideal. But for now, the combination covers my workflow better than any single tool.

Over to You

What’s your setup in 2026? Still team ChatGPT? Made the switch to Claude? Using something entirely different I should test?

Drop a comment. I read every single one, and the best suggestions become future posts.

*If this comparison helped you, bookmark it or share it with someone stuck in decision paralysis.