Neil Patel

I hope you enjoy reading this blog post. If you want my team to just do your marketing for you, click here.

Yes, ChatGPT-4 Still Requires a Heavy Human Touch

An image saying "Yes, ChatGPT-4 Still Requires A Heavy Human Touch.

This blog is written by Austin Cosler, Content Production Manager at NP Digital, and Jeff Caspersen, Copy Editor at NP Digital.

It’s no surprise if you’re enamored by the capabilities of ChatGPT-4. 

We are, too. 

While the rollout of ChatGPT-4 has been impressive — some of its highlights include multimodal capabilities, longer memory, multilingual functionality, and more “personality” and factual accuracy — it’s still a glorified search engine that relies on the training data OpenAI fed it. 

Our content writing team used ChatGPT-3.5 on a regular basis prior to the release of ChatGPT-4, and a significant professional writer-editor touch was essential to shaping what it produced into something usable for our clients. 

ChatGPT-4 is no different. 

Trained writer-editor supervision is necessary to properly prompt ChatGPT-4 and use the copy it generates to produce strong, well-optimized content. There’s even a burgeoning profession related to using AI tools like ChatGPT-4: AI prompt engineering. This shows that understanding the inner workings of these tools and how to use them efficiently is critical for success.

OpenAI even emphasizes that human review is essential when using their tool, especially in high-stakes situations:

Despite its capabilities, GPT-4 has similar limitations as earlier GPT models. Most importantly, it still is not fully reliable (it “hallucinates” facts and makes reasoning errors). Great care should be taken when using language model outputs, particularly in high-stakes contexts, with the exact protocol (such as human review, grounding with additional context, or avoiding high-stakes uses altogether) matching the needs of a specific use-case.

The bottom line: ChatGPT-4 is a useless tool without a professional content writer-editor behind the scenes accurately prompting it and heavily massaging the results it provides. 

Let’s look at some of the pitfalls of ChatGPT-4 for a content writer-editor, as well as how you can use this AI tool to help boost productivity and create great copy. 

The Pitfalls of ChatGPT-4 for a Content Writer-Editor

It’s a glorified search engine.

No writer-editor can simply settle for what ChatGPT-4 gives them, as it can provide similar encyclopedic information to different people (just like a search engine does). Rethinking, retooling, rephrasing, rewriting, and going more in-depth with the copy ChatGPT-4 provides is paramount to avoiding duplicate content scenarios and engaging readers. 

Take the ChatGPT-4 “What is content marketing?” query below.

A query asking ChatGPT-4, what is content marketing?

We ran these results through Copyscape, and the plagiarism software pinged two different sources that featured similar content — and in some instances, verbatim content. In fact, the top hit in Copyscape featured 45% matching text to the ChatGPT-4 results and even featured duplicate clauses. 

It can’t account for E-E-A-T.

ChatGPT-4 simply can’t account for the experiential and first-hand expertise factors of great content that hooks audiences — which is critical during a time when Google is laser-focused on AI and search with  E-E-A-T (the new Search Quality Rater Guidelines acronym that includes “experience”) when evaluating AI-generated content.

People are tired of reading regurgitated content, and ChatGPT-4’s widespread adoption will continue to flood the web with clunky, derivative information. Readers want engaging content written by those who have real-world experiences and subject matter expertise, not an algorithm with no concept of reality, tangible first-hand knowledge, or soul. 

Here’s Google’s most recent advice on using AI-generated copy:

However content is produced, those seeking success in Google Search should be looking to produce original, high-quality, people-first content demonstrating qualities E-E-A-T.

It can’t cite sources or give timely statistics.

ChatGPT-4 still can’t cite its sources, unlike Microsoft’s Bing Chat which provides citations as it scrapes the web for answers. You have to fact-check everything it produces, which can sometimes take twice as long as researching and writing simultaneously — not to mention it can provide misinformation more frequently and persuasively than ChatGPT-3.5. 

Check out ChatGPT-4’s answer when I asked it where it got the results from for its answer to “What is content marketing?”

A query asking ChatGPT4 where it gets its info from.

ChatGPT-4 is still trained on an outdated data set (its training data is capped at September 2021) which means it’s not useful for new topics or breaking news because it can’t give timely supporting statistics or figures. This will certainly change in the future — but that doesn’t help the now. 

A query asking Chat-GPT-4 what's new with content marketing in 2023.

It often scratches the surface and disregards writing best practices.

Although you can ask it to write in a specific tone or voice, it typically provides stiff copy or goes off the deep end with its interpretation (see example below). And while you can also ask more granular follow-up questions to your original query or command, it often produces highly surface-level copy. It also has a penchant for passive voice and an aversion to contractions, which are plagues to quality conversational writing when used incorrectly. 

Check out this total tone and analogy miss:

A query asking ChatGPT to describe content marketing in a certain tone.

It’s a liability landmine.

Using a generative AI writing tool like ChatGPT-4 can be a colossal liability landmine when it comes to plagiarizing content or copyright infringement.

Because it’s trained on a universe of written content created by humans, ChatGPT-4’s outputs will inevitably resemble much of what’s already out there. And, as we touched on above, its widespread adoption means it’ll invariably provide uncomfortably similar content to different users — opening the door to all sorts of intellectual property shenanigans.

And let’s not forget ChatGPT-4’s tendency to “hallucinate” facts and commit reasoning errors. That poses serious red flags if a writer-editor’s using it to craft high-stakes Your Money or Your Life content

The fallout from publishing faulty or false financial or medical information or advice could be disastrous. From both a liability and ethical perspective. 

Consider this take from Microsoft Vice President Peter Lee, author of “The AI Revolution in Medicine: GPT-4 and Beyond.” Asked whether ChatGPT-4 provides trustworthy medical information, he told USA Today:

I personally would not trust models at that level of sophistication for my own personal health advice or information and I would be concerned if doctors and nurses were making it for medical decision-making.

That’s a technology industry insider’s view. One whose company is at the forefront of AI’s evolution. If he wouldn’t trust AI to generate reliable medical advice, no writer-editor should trust it to create reliable medical content — or any level of content with the potential to significantly impact a reader’s life.

It can stunt writer-editor growth.

What’s the best way to sharpen your writing and editing skills? Writing, editing, and researching. Nothing fuels progress like practice, something a novice writer-editor can skimp on if leaning too heavily on an AI writer.

A novice writer-editor evolves into a seasoned writer-editor by learning, researching, and doing — going down grammar rabbit holes, experimenting with turns of phrase, taking courses, gritting their teeth at editors’ annoying edits before realizing, “Oh, they have a point.” 

There’s so much to gain from trial, error, and implementing lessons learned from all that trialing and erroring.

Relying solely on an assembly line of AI tools robs developing writer-editors of the work that polishes their skills. Encourage writers to turn off AI every once in a while and flex their creativity. That might include:

  • Taking the time to write a killer intro hook.
  • Freewriting to play with language, loosen those writing muscles, or to light the way out of brain fog.
  • Putting an emotion-evoking spin on otherwise flat subject matter.
  • Turning out one of those “I wish I’d thought of that”-level turns of phrase.
  • Taking time to read others’ work and learn from different approaches.
  • Seeking out courses to spur growth.

Don’t lose your way amid the chaos of cranking out deliverables. Don’t sacrifice growth, development, and strong content for the sake of speed.

Strike a balance that promotes writer-editor growth and higher-quality writing.  

It poses an ethical problem.

AI labs like OpenAI are in an arms race to release the most advanced algorithmic minds — and this has led to unchecked releases, management, and safety protocols that must exist to keep these tools safe. There’s a growing concern among some of the world’s most influential minds that, if AI labs continue on their trajectory, they could pose monumental ethical risks to society and humanity. 

In fact, a group of more than 500 technologists, engineers, and AI ethicists — including Elon Musk and Steve Wozniak — recently signed an open letter that calls on AI labs to pause all AI system training for at least six months for four main reasons: 

  1. AI systems are becoming human-competitive at general tasks.
  2. AI labs are not taking the proper safety precautions when releasing updates.
  3. There are currently no regulatory authorities overseeing, auditing, or approving AI systems before their release. 
  4. AI creators are unaware of what their systems are capable of due to the race to release advanced systems faster than their competitors. 

While this open letter is a call to action, some countries are already taking matters into their own hands. 

Italy has banned ChatGPT over privacy concerns

For example, Italy recently blocked access to ChatGPT-4, citing that it’s illegal for OpenAI to collect and store personal data to train their system based on the General Data Protection Regulation (GDPR)

Italian privacy regulators also noted that ChatGPT-4 doesn’t always provide factual data based on the real data it was trained on, which could result in the incorrect processing of personal data — another illegal act under the GDPR. Regulators also noted that there’s no age verification associated with the tool. 

Italy gave OpenAI 20 days to notify them of “the measures implemented to comply with the order, otherwise a fine of up to EUR 20 million or 4% of the total worldwide annual turnover may be imposed.”

The Benefits of ChatGPT-4 for a Content Writer-Editor

ChatGPT-4 has several tangible benefits as long as a content writer-editor treats the copy it creates as foundational. Adding expertise, depth, character, sourcing, supporting statistics, use cases, and key elements of the targeted style guide is a must no matter the copy ChatGPT-4 provides.

Let’s look at a few key benefits of ChatGPT-4 for writer-editors:

  • It’s a research tool that can provide information about a subject faster than sifting through the SERPs. 
  • It can help uncover major points or sections to include in the outline of an article.
  • It can help expedite copy production for isolated sections of “encyclopedic” article copy, such as social posts, video scripts, alt-text descriptions, or copy for landing pages or blogs. 
  • It can help repurpose existing content into unique guest blog posts.  
  • It can be a thought starter that helps you find your way out of a fog.

When a writer-editor uses ChatGPT-4 correctly, it can be a productivity boost that still hits all the right notes regarding quality and ethics. But you must have a firm grasp on how to use and what to look out for — and that starts with understanding how to prompt it. 

How to Feed ChatGPT-4 as a Content Writer-Editor

An infographic explaining how to feed ChatGPT4 as a writer/editor.

To adequately feed ChatGPT-4, a content writer-editor must follow these five steps:

  1. Be direct and specific with intent: Input prompts like: “I want you to act as a…,” “You will …,” or “My first request is …” You can also provide examples of content you want it to emulate in your command or query, such as, “Use the following examples to…” For instance, you could say:
    1. “I want you to act as a bank that specializes in mortgage loans.” 
    2. “I want you to act as an IT expert providing tips to a trusted friend.” 
    3. “You will create an outline for an article on the benefits of Kanban boards in project management.”
    4. “You will create an outline for an article on the best grilled cheese recipes.”
  1. Befriend a thesaurus and choose your verbs wisely: If your initial query or command didn’t provide the results you were looking for, changing up your wording could unlock the correct results. Also, be sure the verbs you use in your query or command adequately express your intent. For example, telling ChatGPT-4 to “summarize this” rather than “make this section shorter” could provide better results. 
  2. Chunk things out: Asking ChatGPT-4 to write entire content pieces is generally bad​, even with its new memory and word count capabilities. Instead, focus on generating content for specific, isolated sections of an outline. 
  3. Give it a voice: Feed tone- and voice-related prompts, such as: “Write in a conversational/elegant/academic/fun tone.” Do your best to give it granular tone and voice follow-up directions if it misses the mark, but don’t spend too much time trying to get it to produce exactly what you want. A human touch is often necessary to nail the proper tone and voice.
  4. Ask follow-ups: Depending on the subject matter’s complexity, don’t try to cram too much into your initial prompt. Instead, start small, break out concepts, and get more granular as you go.​ Here’s an example string of questions you could ask that start simple and get more detailed: 
    1. Write a paragraph comparing a 30-year mortgage to a 15-year mortgage.
    2. Write these comparisons as if you were a loan officer giving advice to a trusted client. 
    3. Go more in-depth when comparing interest rates on 30-year and 15-year mortgage loans.
    4. Now describe which type of loan is better for different financial situations.

What to Do After You Feed ChatGPT-4

An infographic explaining what to do after you feed ChatGPT-4.

After feeding ChatGPT-4 and getting initial results from it, a writer-editor must shift gears to focus on giving the copy the credibility and human touch it needs. Here’s what to focus on:

  1. Get granular: A good rule of thumb is to get more granular with depth and specificity, but that doesn’t always mean adding a ton of extra words. Elevate the copy with use cases, examples, statistics, or quotes from subject matter experts to check all the E-E-A-T boxes. If you’re using it to create outlines or to uncover angles for a piece, don’t stop there. Rack your brain for more. Draw inspiration from competitive research and personal expertise. 
  2. Source all claims: Fact-check the f*&# out of every claim. We all know AI plays loose with the truth.​ Is ChatGPT-4 providing points that aren’t common knowledge? Find the sources, corroborate the information, and tweak the copy while including all citations.
  3. Edit with a heavy hand: Approach the copy ChatGPT-4 generates as a professional editor would. Look for issues with grammar, spelling, punctuation, sentence structure, or redundancies in language. Weave in conversational language to smooth out any clunky, stiff copy.​ Keep in check ChatGPT-4’s penchant for passive voice and aversion to contractions.
  4. Add voice and character: Add in personality (in accordance with client brand guidelines, of course). The bottom line: Humanize AI content based on your brand voice and target audience.​
  5. Do an SEO check: ChatGPT-4 is no SEO expert, so it’s imperative to ensure all the correct SEO elements are there, including on-page keyword targeting, internal and external linking, header insertion and hierarchy, writing with the inverted pyramid in mind, and including short paragraphs. 

Now Let’s See These Steps in Action

We prompted ChatGPT-4 with “How do you implement unattended remote access devices for better security?” Here are the results, as well as what we’d recommend to get the copy it generated up to par. 

A breakdown of areas to improve a ChatGPT-4 response in.
  1. Introduction: The introduction ChatGPT-4 provided featured passive voice in “…are often used to…”, generalizations such as “…and other IT equipment,” and a lack of use cases to support “why” securely implementing remote access devices is essential. It also features a dry reading experience and lacks personality, which could go against a potential client’s brand tone and voice.
  2. Bulleted list formatting: ChatGPT-4 generated eight bullet points covering best practices, of course, without sources. This list was far too long, and several of the points it gave could’ve been consolidated into a shorter list for easier reading and comprehension. We also must source all of these “best practices” claims to ensure they’re true.
  3. “Use strong passwords” section: We could go more in-depth here by recommending a password generator, as well as general password-creation best practices and password organizational tools like LastPass. This could also be a good spot to include a quote from an IT thought leader about the importance of creating strong passwords, as well as an example statistic related to a security breach caused by a weak password.
  4. “Use encryption” section: What the heck is AES-256? The reader likely wants to know more. We could provide some additional context around this algorithm and add potential internal or external links to provide them an outlet for even more information.
  5. “Keep software up to date” section: Depending upon the client, we could include its products and services related to cybersecurity or automatic patching services. We could also mention operating system-specific recommendations here to get more granular.

FAQs

What is ChatGPT-4?

In its own words, ChatGPT is “a large language model trained by OpenAI.” It’s based on the generative pre-trained transformer (GPT) architecture — a deep learning model that generates human-like language. Its primary function is to turn natural language inputs into natural language outputs. It’s trained on text from the internet, including websites, articles, and books.

What can ChatGPT-4 do?

ChatGPT-4’s main function is to take natural language inputs (prompts) and turn them into natural language outputs. Through prompting, it can: 

  • Answer questions.
  • Generate written content in multiple languages.
  • Assist in ideation.
  • Summarize longer texts.
  • Translate text from one language to another.
  • Process images to find relevant information (such as explaining what’s in a photo).
  • Generate code in various programming languages.

What are the limitations of ChatGPT?

While it is a robust tool that can significantly speed up content generation, ChatGPT-4 does lack in certain areas, including:

  • It’s a glorified search engine: ChatGPT-4 provides similar encyclopedic information to different people just as a search engine would.
  • It can’t account for E-E-A-T: ChatGPT-4 can’t hit the experiential and first-hand expertise notes that only humans can provide.
  • It can’t cite sources or give timely statistics: Because it can’t cite sources and is trained on an outdated data set, ChatGPT-4 outputs require heavy fact-checking.
  • It often scratches the surface and disregards writing best practices: ChatGPT-4 often provides stiff, surface-level copy that requires a heavy-handed edit from a writer-editor.
  • It’s a liability landmine: Because it’s trained on written content from the internet, ChatGPT-4’s outputs will provide much of what’s already out there, opening the door to potential plagiarism and copyright infringement issues.
  • It can stunt writer-editor growth: If a novice writer-editor leans too heavily on ChatGPT-4 as a content generation tool, they may fall short in developing their writing skills.
  • It poses an ethical problem: Some worry that, if unchecked, AI could pose ethical risks to society and humanity. 

AI Is What a Human Makes It — For Now

Time will tell how advanced AI systems will become. Some speculate that ChatGPT-5, which is set to release by the end of 2023, will achieve artificial general intelligence (AGI) — the holy grail for AI labs seeking to make AI indistinguishable from human capabilities. Check out this tweet from Siqi Chen, a thought leader and entrepreneur in the tech space:

A tweet from Siqi Chen.

But as ChatGPT-4 stands today, it’s only as good as the writer-editor using it. To make it a practical and efficient tool in your content marketing arsenal, you must understand its limitations, how to accurately prompt it, and what to do with the content it spits out.

These are best practices for any AI writing tool, for that matter.

The key is keeping our use of generative AI in check as content marketers. Treat it as an assistant — don’t be its assistant.

Don’t just lightly polish up the words it gives you and call it a day. Always go further.

How are you implementing ChatGPT-4 into your workflow? If you’re not, what are some of your concerns?

About The Authors:

A headshot of Austin Cosler

Austin Cosler: Austin is a passionate creative writer, editor, and people manager with more than 6 years of professional digital marketing experience. Austin is currently the Content Production Manager at NP Digital, where he ensures copy crafted for some of today’s most prominent brands – including Adobe, LinkedIn, and DIRECTV – tells a story, provides an exceptional reading experience, and is optimized for search and user intent alike.

A headshot of Jeff Caspersen.

Jeff Caspersen: Jeff Caspersen is a copy editor at NP Digital. He began his writing and editing career as a sports writer and has since accumulated extensive experience in the digital marketing and newspaper industries. As an editor, he takes pride in helping writers (and AI bots) unlock their best work.

Consulting with Neil Patel

See How My Agency Can Drive More Traffic to Your Website

  • SEO - unlock more SEO traffic. See real results.
  • Content Marketing - our team creates epic content that will get shared, get links, and attract traffic.
  • Paid Media - effective paid strategies with clear ROI.

Book a Call

Are You Using Google Ads? Try Our FREE Ads Grader!

Stop wasting money and unlock the hidden potential of your advertising.

  • Discover the power of intentional advertising.
  • Reach your ideal target audience.
  • Maximize ad spend efficiency.
Ads Grader

Share

Neil Patel

source: https://neilpatel.com/blog/chatgpt-4-requires-human-touch/