ChatGPT Review

This post contains affiliate links. If you buy through them, I may earn a commission at no extra cost to you. Full disclosure policy.

Bottom line: ChatGPT is the AI that introduced most of the world to the idea that a chatbot could actually be useful, and for a lot of writing tasks, it still delivers. It’s fast, flexible, and capable enough to accelerate a professional writing workflow in ways that weren’t possible three years ago. It also fabricates information with total confidence, writes in a voice that reads like every other ChatGPT output ever generated, and gets less reliable the harder you push it. The tool is capable. It’s just not honest about its limits.

Scribbright rating: 3 out of 5

→ See all writing tools in the Scribbright reviews section.

How we got here

I went to grade school with a kid who was like ChatGPT. She was amazingly well-read, had an answer for everything in class, and basked in our teacher’s approval. So much so that if she didn’t have the actual answer, she’d make one up.

In November 2022, OpenAI released ChatGPT, and the internet collectively lost its mind. A million users in the first week. Five days, actually. Nothing in tech had scaled that fast. By early 2023, it was being used for everything from customer emails to college essays to legal briefs, with varying degrees of success and varying awareness of the risks…which were (and are) absolutely real.

By now, you’ve almost certainly used it. The question isn’t whether ChatGPT is famous. It’s whether it’s actually good for writers who depend on their output for their professional credibility.

After using it extensively for professional writing work, our short answer is: sometimes, conditionally, and never without a close eye on what it’s telling you.

What it does well

ChatGPT is fast and versatile, and it adapts reasonably well across formats. It brainstorms well. Ask it for 15 angles on a topic and you’ll have 15 angles in 30 seconds, most of them worth nothing, a few worth developing. For roughing out first drafts of structured content, outlines, listicles, and conversion copy, it moves the work along. It’s also solid on SEO-formatted writing and product descriptions, where formula matters more than voice. As a person who generates B2B content for a living, I can say it’s a great tool for research and ideation.

GPT-5, the current default model as of this writing, improved meaningfully over its predecessors in coherence and contextual understanding. The product has gotten better.

It handles research queries reasonably well for general background, and web access is available for current information. The custom instructions feature, which lets you establish context and tone preferences that persist across sessions, is genuinely useful for writers working on recurring projects. I’ve used it to create LLMs that ape brand voice for articles and ad copy, and it works…within limits.

The prose problem

This is where things get complicated for writers.

ChatGPT writes like a very competent, very generic professional. Every time. The prose is clean, organized, and instantly recognizable to anyone who reads a lot of it. In blind writing tests where ChatGPT got compared to Claude and Gemini, readers consistently described its output as “too smooth” or “sounds corporate,” and it lost seven of eight writing tasks to Claude in one head-to-head study.

A University of Tübingen analysis of 14 million academic abstracts found ChatGPT’s vocabulary fingerprint so distinctive that at least 10% of 2024 academic papers showed signs of LLM involvement, characterized by systematic overuse of words like “robust,” “pivotal,” and “facilitate.”

If you write for humans who read a lot, some of them will notice. And I’m a firm believer that we develop recognition filters over time that enable us, even unconsciously, to spot adbabble, falsity and, now, machine-drafted text.

Run the same prompt twice, and you’ll get two different answers. Run it ten times, and you’ll have ten different answers, some of which contradict each other. Yikes, right? ChatGPT is probabilistic by nature, which is a polite way of saying it’s making educated guesses, and the guesses vary. For writers trying to hold a consistent voice across a content program, that’s not a quirk. It’s undesired extra workload.

→ If you’re evaluating AI writing tools for your workflow, the Scribbright reviews section covers the category in depth.

The hallucination problem

This is the part every writer needs to take seriously before trusting ChatGPT with anything that demands accuracy.

ChatGPT makes things up. Not occasionally, as a known edge case you can watch for. Regularly. Confidently. With perfect grammar and zero apparent shame. It invents citations. It puts words in the mouths of people who never said them. It generates statistics, study findings, and historical facts that are completely fabricated and read exactly like the real thing. In one experiment, researchers had ChatGPT generate scientific abstracts and then asked human reviewers to identify the fakes. They couldn’t. The made-up ones looked just like real research.

Here’s why this doesn’t get fixed: it’s not a bug, it’s the very architecture of ChatGPT. When a language model doesn’t know something, it doesn’t say so. It predicts what the answer probably looks like and generates that. One AI company executive told the New York Times that hallucinations “will never go away” because that’s just how these models work. OpenAI has cleaned up the obvious ones, the howlers, the easily Googleable gaffes. What’s left are the subtle ones. The ones that sound authoritative. The ones that pass a casual read. Those are the ones that end up in print…or in your client’s strategy deck for the C-suite. Oops.

It also gets worse the harder you push it. A 2025 analysis found heavy users encounter three times more hallucinations than light users, with the rate climbing as queries get more complex. The tool rewards confidence and punishes scrutiny.

For writers, this isn’t abstract. A false statistic in a published piece is a credibility problem. A fabricated quote attached to a real person is a legal problem. A made-up study cited as supporting evidence is, at minimum, professionally embarrassing. ChatGPT will not tell you when it’s guessing, because it’s like that girl I knew in grade school: it’s so eager to make you happy! It’ll present the guess in the same clean, well-formatted, utterly confident prose it uses for everything else, and you will have no way to tell the difference except by checking. Or (and this is almost fun), asking it. Actual exchange:

Me: Was the statistic you quoted above a real statistic?

ChatGPT: Sorry, I can’t verify that statistic from a reliable source. I should not have presented it as factual. Treat it as a hallucinated or unsupported number unless we can trace it to a source.

It’s a pet peeve of mine that an algorithm refers to itself as “I”, but that’s another issue. It’s no R. Daneel Olivaw, for you Asimov fans.

Verify everything it tells you. Not selectively. Everything. Are you up for that?

The pricing question (and why you shouldn’t get too comfortable)

Right now, ChatGPT Plus and comparable subscriptions from competing platforms cost what amounts to a rounding error in a professional writing budget. That’s unlikely to stay true indefinitely.

AI providers are currently pricing their consumer subscriptions well below the actual cost of the compute behind them. It’s a land-grab, not a sustainable business model, and the industry knows it. It’s the Uber model of providing an addicitive service at loss-leading prices, then jacking them up once you’ve gained traction. The more likely long-term direction is token-based pricing: you pay for what you use, at rates that reflect actual infrastructure costs rather than discounted monthly access. Some providers are already moving this way at the API level. Consumer pricing will probably follow as the market matures and the subsidized-subscription phase winds down.

What this means practically: the $20-a-month all-you-can-use model may not be the permanent state of affairs. Heavy users who treat ChatGPT as a replacement for a significant portion of their writing output could find the economics look quite different in two or three years. Or maybe less.

The hedge against this is the same one that applies to any platform dependency: don’t build your workflow entirely around a single tool. ChatGPT, Claude, Gemini, and others each have genuine strengths, and the switching costs between them are low right now. Writers who are fluent in more than one platform will be in better shape when pricing models shift than writers who’ve committed exclusively to one and built every habit around it. Keep your prompting skills general, not platform-specific. Your preferences about which AI produces better output for your work are worth having. A single-platform dependency is not.

How it compares to the alternatives

For writing specifically, ChatGPT isn’t the strongest option available. Independent comparisons consistently put Claude ahead for voice-driven and long-form writing: it matches a writer’s style more precisely when given examples, produces less templated prose, and handles tonal consistency better over long documents. Gemini has an edge on real-time research tasks through its Google integration. ChatGPT’s relative strengths are structured, conversion-focused writing and everyday utility tasks, where speed and flexibility matter more than voice distinctiveness.

Dedicated AI writing tools, Jasper, Copy.ai, and Sudowrite among them, layer on workflow features (brand voice controls, project organization, SEO integration) that general-purpose chatbots don’t offer natively. For writers with specific production workflows, those tools are worth evaluating alongside ChatGPT.

That said, ChatGPT has the largest user base, the broadest format support, and the deepest third-party integrations of any AI on the market. Familiarity and ecosystem have real value.

Who should use it

Writers who get value from ChatGPT treat it as a drafting accelerator, not a finished-copy machine. It’s useful for ideation, structural outlining, breaking through blank-page paralysis, and roughing out first drafts that get substantially rewritten before publication.

It’s a reasonable tool if your workflow involves heavy editing of AI output rather than light cleanup. If you’re publishing AI-assisted content with light revision, the voice problem and the hallucination risk both become more acute.

Where it falls short: anything requiring factual precision, consistent distinctive voice, or reliable citation. Use it to move fast and think broadly. Verify everything before it leaves your desk.

Pros

• Fast, versatile, adaptable across a wide range of formats and tasks
• GPT-5 improved meaningfully on coherence and context over prior versions
• Strong for brainstorming, outlining, and structured conversion copy
• Custom instructions feature useful for persistent context across sessions
• Web access available for current information
• Largest ecosystem and third-party integrations of any AI platform
• Free tier available for light use

Cons

• Hallucinations are structural, documented, and getting subtler rather than rarer
• Prose voice is generic, templated, and heavily AI-detectable
• Output inconsistency across identical prompts makes voice consistency difficult
• Less reliable, not more, as task complexity increases
• No uncertainty signaling: fabrications presented identically to accurate output
• Trails Claude on voice-driven and long-form writing quality
• Flat-rate subscription pricing may not last; token-based costs could change the economics

The verdict

ChatGPT is the most famous AI writing tool in the world, and it earned that position. For a lot of tasks, it’s still excellent. For writers who depend on accuracy, distinctive voice, and consistent output, though, its limitations are professional liabilities rather than minor inconveniences.

Use it early in the process and stay skeptical throughout. It will speed up your work. It will also, if you let it, put things into your published copy that you’ll regret. And don’t get too comfortable with the monthly subscription rate: the economics of AI are still sorting themselves out, and what you’re paying now is not necessarily what you’ll be paying in three years.

The tool is capable. It just doesn’t know the difference between what it knows and what it’s making up, and it will never tell you which is which. That’s a real problem for real writers. Know it going in.

Michael Semer is a professional writer and editor with 40 years of experience across advertising, B2B content marketing, and digital publishing. He has used ChatGPT extensively and has the fact-checking scars to show for it.

→ More AI writing tools reviewed at Scribbright.

Frequently asked questions

Is ChatGPT good for professional writers? As a drafting and brainstorming accelerator, absolutely! As a source of finished or near-finished copy, it takes careful supervision. The prose voice is generic and detectable, the output is inconsistent across identical prompts, and the hallucination risk is real enough to treat everything it produces as a first draft requiring verification before publication.

What is ChatGPT hallucination and why does it matter for writers? Hallucination is the term for when a language model generates information that is plausible-sounding but factually wrong or entirely fabricated: citations that don’t exist, quotes never said, statistics invented whole cloth. It matters for writers because ChatGPT presents fabrications with the same confident prose it uses for accurate information, with no signal that what it’s telling you is, well, wrong. Heavy users see the problem more frequently as they push the tool on more complex tasks. Verify everything before you release it to a client or into the wild.

How does ChatGPT compare to Claude for writing? For voice-driven, long-form, and stylistically distinctive writing, Claude is the stronger tool. It matches a writer’s style more precisely when given examples and produces less templated output. ChatGPT has an edge on structured conversion copy and everyday utility tasks, and carries a larger ecosystem of integrations. For writers prioritizing voice and accuracy, Claude is the better primary tool.

How much does ChatGPT cost? There’s a free tier with usage limits, and ChatGPT Plus unlocks higher capacity and priority access. Verify current pricing at openai.com before making a decision; it changes. Worth noting: the subscription model across all AI platforms is a land-grab phase, not necessarily a permanent pricing structure. Token-based pricing (pay for what you use) is already standard at the API level and may eventually reach consumer subscriptions. Budget accordingly.

Should I use just ChatGPT or multiple AI tools? Multiple tools. Each platform has genuine strengths, the switching costs between them are low, and the pricing landscape across all of them is likely to shift. Writers who are fluent on more than one platform will have more flexibility when economics change than writers who’ve committed to one and built every habit around it. Keep your prompting skills general rather than platform-specific.

Is ChatGPT’s writing detectable as AI? More so than most competing models. Research analyzing millions of academic documents found ChatGPT’s vocabulary patterns distinctive enough to identify at scale, with systematic overuse of specific words that function as fingerprints. AI detectors were largely trained on ChatGPT output first and remain most accurate against it. Writers concerned about detectability should use it as a drafting tool with substantial rewriting, not as a copy source.

Can ChatGPT browse the internet? Yes. Current versions have web access for retrieving current information. This reduces but does not eliminate hallucination risk; the model can still fabricate or misrepresent information from sources it retrieves. Treat cited sources as leads to verify, not confirmed facts.

What’s the best use of ChatGPT in a writing workflow? Early-stage work: brainstorming, angle generation, structural outlining, and rough first drafts you intend to rewrite. It’s a tool for moving fast through the messy early phases of a project. The further downstream your workflow puts AI output, the more the voice problem and reliability issues affect the final product.