We tested out four GPT-3 copywriting tools

What exactly is GPT-3, anyway?

GPT-3 (Generative Pre-trained Transformer 3) is an AI model created to mimic human-generated language (aka natural language processing or NLP). It has received a mountain of tech-news buzz, with good reason: From full blog posts to tweets, the AI application produces content that can easily pass as human writing. Some fear GPT-3 will replace professional writers. Still others see the harmful uses for the AI, like spreading disinformation.

Let’s strip away the hype and talk about the basics of GPT-3: what it is, what it isn’t, and what it means for content professionals.

• Research and development lab OpenAI created GPT-3. The organization (whose early investors include tech titans Elon Musk and Reid Hoffman) is on a mission to “ensure that artificial general intelligence benefits all of humanity.”
• GPT-3 has read the World Wide Web, as in all of it. To learn about human writing, the AI consumed 45 terabytes of text. That includes all of Wikipedia, Google Books, and Common Crawl (every published piece of content on the web from 2016 to 2019), “which equals a book about 3,375,000,000 pages long.”
• GPT-3 only needs a few (2-3) examples to deliver on specific writing tasks. The NLP precursors to GPT-3 require a large dataset to complete distinct tasks. GPT-3 is a few-shot learner, meaning once you give it a couple examples of what you need, it can take a request and run with it.
• GPT-3 is a commercially available API for developers. It would take a private company vast sums of money to develop their own version of GPT-3 to power their product. Instead, developers can build their product on the GPT-3 framework for a licensing fee, like how Amazon Web Services (AWS) works for cloud computing.
• OpenAI’s GPT-3 was first to market, but there are more general pretrained transformer engines on the way. GPT-Neo, aka “open-source GPT-3,” is already available thanks to the work of the grassroots AI research scientists at EleutherAI. It has performed shockingly well against GPT-3. Google Brain’s 1.6 trillion-parameter model—“6 times bigger than OpenAI’s GPT-3”—will likely be commercialized as well. There are whispers of additional open-source GPT engines from the academic community under development.

The drawbacks of GPT-3

Despite the impressive nature of the tech, GPT-3 has some significant weaknesses that can impact the quality of your content.

• GPT-3 alone doesn’t fact-check or verify sources. This could be a huge problem for content teams that rely on accuracy of information; for example, newsrooms or teams that publish reports like whitepapers.
• The text GPT-3 generates shows biases. As ethics researcher and Google whistleblower Timnit Gebru revealed, large language models like GPT-3 are exposed to the most harmful content on the internet. That means it can spit out racist tropes and gendered language.
• While GPT-3 is smart, it’s still a mindless piece of technology. Think of the tool as a writing parrot: With just a few examples, it can combine words in a multitude of ways to mimic human writing, but it doesn’t always make sense of what it’s writing. As you’ll see in a moment, the result is often a lot of textual “noise” without a “why” behind what it says.

With these challenges in mind, content teams should assume that anything generated via GPT-3 will need fine-tuning. Build guardrails to guarantee the GPT-3 tool is following journalistic, editorial, and ethical standards alongside brand and writing guidelines.

Top four GPT-3 copywriting tools in 2021: strengths, weaknesses, and the best use cases

Now that you understand what GPT-3 is, let’s look at a few copywriting tools built on GPT-3. We’ve tested out four of the most popular tools to see where they excel, where they need work, and what kinds of content teams might benefit the most from each.

Without further ado, here’s how our contenders performed.

Copy.ai

+ solid user experience

– contents needs lots of help

Copy.ai positions itself as a “creativity tool” that helps people “generate marketing copy in seconds.” Its ease of use impressed us, but our professional content editors say the writing needs work.

Who it’s for

So far, most testimonials are from startup founding teams (founders, CEOs, and marketers) and solo entrepreneurs.

What it writes

Copy.ai supports dozens of writing tasks, from creating LinkedIn Ad copy to generating metadescriptions for blog posts and web pages. The platform lists product descriptions, Instagram captions, blog ideas, blog intros, and value propositions as the most popular use cases.

Sample content

What professional editors have to say

“The synthesis of the first two sentences is (and I’m speculating here because it doesn’t make sense syntactically) ‘AI writing tools are like Photoshop.’ First off, we don’t say how. But the follow-up is that, ‘because AI writing tools are like Photoshop,’ deepfakes and AI have ‘risen in parallel.’ It’s a non-sequitur. No connection is made, and we don’t know what ‘risen in parallel’ means.”

Nathan Wahl

The Pros of Copy.ai

+   From the user onboarding walkthrough to the intuitive user interface, it’s clear that Copy.ai’s design team cares about user experience.

+   We also appreciate that the designers set realistic expectations around the ability of the AI to produce high-quality outputs; they claim that the more context you can provide (like user role, industry, and target audience), the better the outputs the AI can generate.

The Cons of Copy.ai

−   The large collection of copywriting use cases means that Copy.ai’s “writer” doesn’t specialize. As our editorial team pointed out, some results are underwhelming and not particularly helpful.

−   In addition to the previous blog intro sample, we tried out the Copy.ai listicle tool to create a listicle about “common editing mistakes,” which produced generic (and sometimes nonsensical) headings.

In addition to the gobbledygook factor, we discovered another major drawback: Copy.ai’s content doesn’t include citations for quotes or statistics. So, if you use the tool to write blog posts, you’ll need to provide your own research, fact-check thoroughly, and check for plagiarism.

Recommended use cases

While Copy.ai still has a long way to go before it can create professional-quality content writing, there are still a few nifty applications for the tool:

• Use the text generated for blog intros to overcome writer’s block and write your own version.
• Organize your ideas with the blog outline tool.
• Generate multiple ideas for headlines.
• Repurpose existing content by turning blog posts into multiple formats (e.g., Instagram posts, Google headlines, Facebook posts).
• Generate ideas for paid ad testing before having human writers and marketers turn it into a full-blown campaign.

Copysmith

+ Promising output (when it works)

– Fact-checking is a must

Copysmith created the only content our editors were willing to greenlight. But because the tool was glitchy when we ran our test, the overall results are inconclusive.

Who it’s for

Copysmith’s website includes testimonials from agencies and startup marketing teams.

What it writes

Like Copy.ai, Copysmith offers a huge range of content types, from blog writing and ad copy to brand messaging and press releases.

Sample content

Nathan Wahl caught one of the factual errors in the above sample:

“According to the graf, ‘this technique’ is ‘listening to and learning from videos of a person speaking.’ But that’s not what GPT-3 is.”

Nathan Wahl

Milo Davidson thought it was a good start:

“This one is the best; it gives context, drives interest, and gestures at making an argument. But ‘it’s not perfect, so it can be evil’ doesn’t really hold water as a logical argument and needs fleshing out.”

Milo Davidson

That said, the editors reviewed the sample for clarity and writing style, not for fact or plagiarism checks.

The Pros of Copysmith

+   When the platform provided samples, they were well-written.

The Cons Of Copysmith

−   After selecting the type of content we wanted to produce, we experienced various problems getting the platform to work. We reached dead ends waiting for the tool to generate content, and the system crashed several times.

−   Even without glitches, we found it challenging to navigate the tool itself. There’s no product onboarding tour to help, and while Copysmith has a collection of tutorial articles, they aren’t well-organized.

Recommended use cases

Since the platform is so unstable, we recommend against using it for now.

Writesonic (formally Magicflow)

+ lots of output

– needs lots of edits

Writesonic shows promise as a marketing copy assistant tool for small businesses. But while it saves time on coming up with ideas, busy content teams may find it creates more editing work.

Sample Content

Ronald Barba found the writing incoherent and disorienting:

That first sentence is a doozy; I’m not really sure what we’re talking about/referencing and then you bring up something about an AI engine and I’m lost as to how it all connects. I know there’s a thesis there, but I’m just not 100% sure of what that is. The metaphors add to the confusion.”

The Pros of Writesonic

+   Using Writesonic to generate content is straightforward: it walks you through every step.

+   The app generates at least five different versions of text for each copywriting task, so you can choose a winner—or at least something that doesn’t need to be totally rewritten.

The Cons of Writesonic

−   As you can see in the above sample, the blog content generated by Writesonic needs heavy editing. The incoherent sample reviewed by our editors was the best of the bunch that Writesonic delivered. The other four versions were either nonsensical or wildly inaccurate.

Recommended use cases

If you want to give Writesonic a try, you may have better luck saving time on short copywriting tasks like social media ads, email subject lines, or blog titles. Product Hunt reviewers said they were impressed with the tool’s ability to generate landing page copy.

Since the tool delivers multiple versions of text, Writesonic could come in handy for generating ideas to help overcome writer’s block.

Writesonic was the only tool we tested that highlighted its ability to translate text into multiple languages, which could be a helpful feature for companies that serve customers globally.

Conversion.ai

+ ease-of use

– writing quality

Conversion.ai’s mission is to “help people break through writer’s block to finally get their content done.” Its GPT-3-powered AI, called Jarvis, writes their website copy.

Based on what we saw in our tests, if Jarvis wrote Conversion.ai’s copy, it must have gone through an intensive amount of editing before it was publishable.

Who it’s for

Over 10,000 copywriters, marketers, and entrepreneurs use Conversion.ai.

What it writes

Conversion.ai offers templates for a wide variety of writing tasks, including marketing frameworks, copywriting formulas, blog outlines and introductions, product descriptions, emails, video scripts, social media and ad copy.

Sample Content

Jordan Miller caught several instances of factual errors and illogical messaging:

“This one has a few glaring red flags. First, I believe GPT-3 only produces text, not audio or video. And the comparison with Photoshop doesn’t work here. GPT-3 creates while Photoshop — at least how they describe it — only alters. The distinction between simulating and approximating doesn’t really land either; what’s the meaningful difference? And finally, the Twitter implication doesn’t really make sense. GPT-3 couldn’t hack the president’s phone, it could only create a fake Tweet. I could do that, too. The issue is that neither of us has his password. And the awake/asleep bit is a miss… I wouldn’t give this the go-ahead.”

The Pros of Conversion.ai

+   Conversion.ai is easy to use. The minute you log in for the first time, it provides clear next steps.

Many of the writing tools within Conversion.ai invite users to suggest what kind of tone Conversion.ai should write in, from “witty” to “professional.” The AI can even try to mimic the writing style of famous authors, celebrities, and fictional characters.

The Cons of Conversion.ai

−   As our editors noted, Conversion.ai’s output is spotty at best. At first glance, the copywriting seems human-written, but upon a deeper read, it’s clear that the AI is putting together a jumble of related words without meaning.

−   Unlike the other contenders we tested, Conversion.ai doesn’t offer a free trial. The lowest-tier monthly plan costs $29 up front, and it costs a whopping $109 per month to access all of the capabilities.

Recommended use cases

Like the other GPT-3 copywriting tools, Conversion.ai is handy for unlocking writing ideas. Based on the results of our test, we can’t recommend the product for producing long-form written content unless you have an editor and fact-checker on hand to fix anything the AI creates.

GPT-3 is a writing tool (not a replacement)

GPT-3 is a big step toward automating content creation, but can’t come close to replacing human writing talent— yet. Until AI copywriting tools like those we reviewed can perform at the same level as professional writers, think of these tools as writing helpers. You can provide your creative team with GPT-3 copywriting tools to help plant seeds of creativity, speed up the content creation process, and repurpose existing content.

No matter how human-like AI writing gets, companies will always depend on creative thinkers like you to drive and prove the impact of content strategy and hold your company’s content accountable to the values and quality standards of your brand.