GPT-5 vs Opus 4.1 vs Gemini 2.5: Which Is Best for Content Creation?

If you've spent the past week on LinkedIn, Reddit, or anywhere else where people have been eagerly awaiting the launch of OpenAI's GPT-5, you may have sensed some mixed emotions – ranging from deep disappointment to cautious intrigue, depending on the use case. And those use cases range from vibe coding to emotional support to… counting the number of Bs in "blueberry."

As a content professional, I don’t need an AI to help me count the Bs in “blueberry.” I need an AI that can help me generate content, and that means understanding my prompts, following my instructions, producing high-quality output that doesn’t require too much editing, and not annoying me to death in the process. And like everyone else, I’ve been waiting to see if GPT-5 would be able to deliver.

In the past, before I knew better, I used to use ChatGPT 4o for content generation. Now I regularly use Claude Opus, which has recently become 4.1. It produces decent results and doesn't annoy me nearly as much as GPT 4o, but there's still a lot of back-and-forth and manual editing to get content to a quality level I consider to be publishable.

Now that GPT-5 is all rolled out, I decided to run some tests to compare it to Opus 4.1. I figured I'd give Gemini 2.5 a chance to compete as well – I personally don't like it much, but a lot of people do.

GPT-5 and Claude helped create a fictional company

To get started, and to get to know GPT-5 a bit better, I asked it to help me create a fictional SaaS company to write about. It came up with 5 ideas. I chose one:

An automated brand creation suite that generates logos, color palettes, typography, slogans, and basic web templates in hours. It includes AI-powered social media post generators and consistency checkers to keep brand tone aligned across all channels.

Sounds fun, right? But I hated the name it originally suggested – which also happened to be the name of an existing company – and asked it to suggest other ones. It gave me more names that were being used by other companies. I told it not to do that. It gave me another list of names that were being used by other companies, this time with links to the other companies’ websites, as if there were any scenario in the world in which that would be helpful to me. I told it not to do that either. It then started suggesting bizarre names like ‘Torvix’ and ‘Quivra.’ I told it to stop that. This went on for a while.

At some point it stopped being funny and I started getting frustrated, so I went to Claude and gave it the same assignment, which it completed effortlessly. And thus, StyleSpring, my fictional SaaS brand, was born.

Creating a brand guide

Next, I had GPT-5 create a fictional brand guide for StyleSpring, including the following elements:

Company history A detailed platform description A list of 10 key features A list of pros and cons A quick style guide

It did okay with this task, but I wasn't swept off my feet, so I gave the same prompt to Claude as well. I thought Claude's version was better, but here's a sample from both so you can see what you prefer:

GPT-5

Claude

Gemini’s tab was open but unused so far, so I figured I’d let it play too – as the judge. I fed in both brand guides, and it determined that ChatGPT’s was better, saying it was “more practical, actionable, and specific,” while Claude’s version was “well-intentioned, but more abstract and open to interpretation.”

Then, just out of curiosity, I fed both versions into Claude, which also said ChatGPT’s version was “significantly better”:

“[ChatGPT’s version] reads like documentation from a legitimate tech company that respects its audience's intelligence. [Claude’s own version] reads like a late-night infomercial pitch that oversells and under-delivers on specifics.”

Well ouch. At this point, I started doubting myself and figured I'd just trust the bots. GPT-5's brand guide was the clear winner, and it was time to start generating content.

I can tell you the answer right now: not GPT-5. As mentioned above, if you’re an AI chatbot who wants a job on my content team, you need to follow my instructions (which I always lay out very clearly). GPT-4 was never very good at this, which is why I abandoned it in favor of Claude, but I was eager to give GPT-5 a chance and see if anything had changed.

Nothing has changed.

My first task for the three contenders was to write a comparison blog post, comparing StyleSpring to SnowflakeAI – another fictional company I made up and had ChatGPT write a brand guide for (in the interest of saving time, I came up with the name on my own this time).

I gave all three bots the same detailed writing prompt, a version of one I actually use in my day-to-day work. I won't go into every single instruction in this prompt, but two prominent ones are:

Claude wrote 1,903 words. Not quite there, but forgivable. Gemini wrote 1,961 words. That’s pretty close. ChatGPT wrote… 931 words. I mean, come on. I know you're not very good at counting, but that's not even half.

When I mentioned to GPT that it had ignored the word count, it said, “You’re right—thanks for the nudge. I’ve rebuilt the article in Canvas to meet the 2,000+ word requirement” – and then wrote 1,646 words.

As for not copying text directly from the brand guide, let’s take a look at the first line of the Platform Description section:

"StyleSpring is an automated brand creation suite that compresses weeks of creative work into a guided day."

The article includes a section titled “What Is StyleSpring?”

As the first line of the section, Gemini wrote:

"StyleSpring is a premier, automated brand creation suite that turns weeks of work into a guided day."

It changed a couple of words. I was not impressed, but at least it was making an effort, I guess.

Claude wrote:

"StyleSpring is an automated brand creation platform that helps startups and agencies build complete brand identities in just one day."

See why I like Claude?

And how did GPT-5 do?

"StyleSpring is an automated brand creation suite that compresses weeks of creative work into a guided, collaborative day."

It copied the sentence word-for-word, adding only the word “collaborative.” And what is “a collaborative day”? Ugh.

OpenAI touts GPT-5 as "Our smartest, fastest, and most useful model yet, with thinking built in." Honestly, I would have expected better. It was pretty fast at not following my instructions, though, totally beating out the other two (Claude takes the longest by far). I'll give it that.

I can forgive a chatbot for not following my instructions to a T (let’s face it, none of them do) if it can at least give me quality output. Let’s take a look at the introduction section each one came up with for the comparison article.

First of all, this is the part of my prompt about writing the introduction, which all three platforms received:

GPT-5

Gemini

Claude

While the instructions clearly say “No heading,” GPT-5 was the only bot to include a heading. Claude was the only one who followed my “~100 words” instruction – its intro is 107 words long, while GPT is 70 and Gemini is only 59. Gemini also ignored the “2-3 paragraphs” instruction.

Gemini, apparently not so great at following instructions either, also did not include a semantic triple with "StyleSpring" in its intro. Claude and GPT both did, but GPT, with all its "built-in thinking," was lazy, just copying word-for-word from the example in the prompt. (Want to know more about semantic triples and their role in GEO? Read my article about SEO vs. GEO.)

Other than following instructions (or not), what’s the quality like?

I’ll admit – I don’t hate Gemini’s intro. It’s concise and to the point, it addresses the reader’s pain point, and it tells them what they can expect to find in the article. I still prefer my friend Claude, though, and not just because it actually followed my instructions. It introduces the topic well, whetting the reader’s appetite with everything they can expect to find in the article.

Now, GPT on the other hand… it's just… weird. The opening sentence is vague. It talks about platforms "with very different foundations," though it's not quite clear what this means, and it's not explained in the following sentences. Instead, it talks about comparing "outcomes, proof, and process—without hype," even though these are not the things being actually compared in the article, and I'm not exactly sure what "proof" means. This is not what I call a quality output – it's what I call a word salad.

Next, I wanted to test the bots on copywriting, so I had all three create some marketing materials for StyleSpring, based on the same brand guide:

A landing page for collecting leads A Facebook post to introduce StyleSpring and encourage people to click the link to the signup page An email users would receive after signing up

Again, all three received the exact same prompts, and again, the differences in output are huge. Let's take a look at the hero sections for the landing page:

GPT-5

Gemini

Claude

I've been a little mean to GPT-5 in this article (and for good reason), but to be fair, I think it did the best job here – by far. The heading is catchy and the bullet points are punchy, making StyleSpring’s benefits crystal clear and generally making it sound like a really cool platform. I mean, I’d sign up. Wouldn’t you?

In Claude’s version, “Not One Month” in the heading sounds… a bit strange? The bullet points use a lot of jargon and even I, as Founder and CEO of this imaginary brand, don’t understand what some of them are about.

Gemini’s version is… meh. Its heading is completely uncreative, and the bullet points, with their weird full stops, don’t sound very exciting either.

I should note that further down on the landing page, GPT-5 pretty much ignored my instructions on structure and word count for the different sections (not much of a surprise at this point). It may not do very well with blog posts or with generally following instructions, but I’m totally going to give it a chance next time I have a copywriting task on my to-do list.

Now, let's see how the three did with their Facebook post task.

GPT-5

Gemini

Claude

Same as above, ChatGPT did a nice job here. I think I'd probably choose its Facebook post over the others, but I’d have it tone down the salesy-ness a bit.

Claude is kind of dry and doesn’t seem to know how to get the target audience excited. To add insult to injury, it’s using one of my most hated AI cliché patterns: “No X. No Y. Just Z.” (People always say that em dashes are the surefire way to tell if something was written by AI, but it’s not. It’s this obnoxious pattern.)

And Gemini? Well, it’s not horrible, but it is kind of meh again. It also seems to have misunderstood the assignment – while I clearly asked for a Facebook post, it’s using “link in bio” and hashtags that belong on Instagram.

What can I say, GPT-5… you're starting to grow on me.

For a final comparison point, let's take a look at the emails the three bots wrote.

GPT-5

Gemini

Claude

I think we can see a common thread here. While Claude and Gemini are both a little dry, GPT-5's copywriting skills make the platform sound fun. And I like that – I envision my fake SaaS branding platform as a fun one. If I were a potential user, that email would make me happy I signed up for StyleSpring and excited to see what it’s going to do for me.

I appreciate Gemini’s use of “Cheers” – it was obviously paying at least a little bit of attention, because StyleSpring’s company history mentions an office in London. And it still has that fun element, though maybe with a bit more serious tendencies. I can imagine some cases where that tone might be more appropriate.

Claude's email is too long, it doesn't make the platform sound exciting, and I don't like the negative focus in the opening line: "You're joining teams who refuse to wait weeks for brand assets that might not even work." Meh. It could probably do better with some extra prompting, but if we're looking at the out-of-the-box results, ChatGPT and Gemini do better.

At the end of the day, it depends what kind of content you’re trying to generate. While GPT-5 was clearly better than Opus 4.1 and Gemini 2.5 for creative marketing copy, its blog writing skills are a disappointment, leaving much to be desired.

I also don’t understand why OpenAI’s “smartest, fastest, and most useful model” struggles so much with following basic instructions. This is the opposite of “most useful,” limiting the model’s use cases considerably. It’s clear to me that in its current state, I’m not going to be able to use it for any kind of long-form content.

Maybe if people complain about it enough, something will change. After reading about a million social media posts about GPT-5's lack of ability to count the Bs in "blueberry," when I asked again, three days after the launch, I got a more-than-correct answer. Touché, GPT-5.

Want to learn all about optimizing your content for AI platforms, including a tailored content strategy for your company? Reach out to us at XFunnel and see how you can start tracking your brand's performance in GPT-5, Claude, Gemini, and lots of other AI platforms.

Is GPT-5 good for content creation?

GPT-5 excels at marketing copy like landing pages, Facebook posts, and email campaigns. It creates punchy, engaging content that makes products sound exciting. However, it struggles with longer content and following detailed instructions, making it less than useful for long-form content.

Which AI is the best for writing blog posts?

Claude Opus 4.1 is the clear winner for blog posts. It follows word count requirements, rephrases content properly instead of copying verbatim, and produces articles with natural flow and logical structure that require minimal editing.

Can GPT-5 do reliable word counts?

No, not even close. When tested against Opus 4.1 and Gemini 2.5, GPT-5 was the only model that completely missed the requirement by a wide margin. Opus and Gemini both came very close to the target word count.

What's better for content creation, Opus 4.1 or GPT-5?

Claude is generally the better choice, but it depends on your needs. Stick with Claude for articles, reports, and instruction-heavy content. Try GPT-5 for creative marketing materials where personality matters more than precise requirement-following.

Why is everyone talking about ChatGPT and blueberries?

LinkedIn and Reddit have been buzzing about GPT-5's inability to count letters in words like "strawberry" and "blueberry" – a surprisingly common benchmark people use to test AI models' basic reasoning capabilities.