How To Use AI To Stop Sounding Like a Robot in Your Writing

In this post, I'll look into one of the most common writing tips – write how you speak. I'll use tech to run a quick experiment that takes this advice in the most literal way possible. 

Let's get to it.

What does it mean to write how you speak?

 If you have ever looked for writing tips online or elsewhere, this must have been one of the hardest to grasp. 

It’s vague when said alone. So what really is this write how you speak? Sure, it sounds like something that could help you sound more genuine and more who you are. But then again, that depends on how well can you speak.

Looking for a more structured approach

Let's do a quick Google search to see what else there is.

One of the first results point us to marketingexamples.com where point number eight is write how you talk.

There’s not much. The post suggests us being casual, colloquial, and full of pronouns.

Is there anything else to it?

Can ChatGPT come up with something better?

Let's start with something super simple. For example, how to write how I speak.

This is not too helpful. 

Let's see what happens if we ask it to write a blog post outline with separate points.

Looks decent on the 2nd try

We can skip the intro, there’s no tangible advice in there. The same applies for the conclusion in this outline.

With those out of the way, there seem to remain 2 main groups of tips.

Group 1 – advice you have to know yourself for

Identifying the elements of your voice and the words and phrases you commonly use is nice advice if you know those. If you don't, you can find out by asking your friends about it.

Then your tone and demeanor. The way you express your thoughts and ideas. Using structure and grammar that accurately reflects how you speak. All are good tips, but you have to know yourself well to take this advice.

Further down there’s also practicing writing in your own voice. It suggests writing a short paragraph or two in your own voice. That’s nice and also exactly the thing we’re trying to figure out here.

Group 2 – immediately actionable advice

The 3rd section in ChatGPT’s outline is even more useful.

Avoiding overly formal language is a good one for sure. Uprooting some of the things we learned in school or other place of formal education can help a lot.

Using contractions and informal language for a conversational tone is very close to the last tip.

Paying attention to the rhythm and flow of your sentences is a little harder, but still easy to spot when editing yourself. If it doesn’t feel right – it probably isn’t the way you’d say it.

And of course, reading it aloud and listening for elementsof your voice is a great tip. You should absolutely read out aloud what you’ve written, especially, if you aren’t writing regularly.

Reading it aloud might even help you figure out a few of the group 1 tips from the previous section.

To sum up ChatGPT’s advice

The outline from ChatGPT is decent. It has some valid points and covers what I’d say is the most important tip – reading your writing out aloud. But nowhere does it mention another approach that I’ve decided to experiment with.

Experiment: The most literal way to write how you speak

If you think about it, there’s only one true way to sound like you speak. It’s speaking. Duuuh.

That’s why I decided to do an experiment that’s either novel or stupid – record myself speaking, auto-transcribe it, and then try to turn that into a blogpost.

Structure of this experiment

Here's what I'm going to do:

1. Record something that could be a blog piece.

2. Transcribe it automatically with some software.

3. Edit and turn it into a blog post.

4. Determine if it's actually worth doing.

 This is going to be a little bit meta since I'll be showing the process on the material that you have just read.

So how did this meta experiment turn out?

The first realization is that I'm nowhere near as comfortable speaking as I'm writing. It comes as no surprise and is a matter of practice for most of us.

In terms of writing that came out of this, here's a side by side comparison of what I spoke and what the edited post looks like:

Visit this page to inspect the text.

There are some obvious changes like video replaced with post, also the addition of some section headers.

Some areas were not picked up at all despite being the same exact thing. E.g. let's do a quick Google search to see. Then there are also blocks of text which I decided to remove in writing and later removed in the video itself too.

Overall, quite a lot of red and green indicating changes which reflect how much editing I had to do to turn it into writing despite meaning staying the same.

Is that writing how I speak?

I would say so.

There are some recurring phrases and elements of speech that I use. I'm not necessarily proud of all of them, but they're there and the sentences are pretty short and digestible.

Is that good writing?

Let's check it with Pro Writing Aid for a more objective look.

The summary report shows 67% of goals as achieved. That’s not horrible, but less that I would typically publish.

If you look at where this document needs work, some of the areas can be ignored or fixed almost immediately.

• Grammar score – a technical matter, easy to fix, doesn’t take away from the purpose of this experiment.

• Sentence length – on the low side, which is expected given that I spoke this. It’s not even bad if we have enough sentence length variety.

The main pain points are these:

• Sticky sentences (glue index) – too many sentences have too many words not really meaning anything, just like this one.

• Too many adverbs – should be fixable.

The good parts which were likely helped by the approach:

• Readability grade

• Sentence length variety

• No extra long sentences

• No passive voice

Let’s give this text a fix and see how that changes the report.

Post-editing writing quality

After some polishing up, the summary report shows 79% of goals as achieved. I could root most of these out with a fair deal of editing, but that would defeat the purpose of this experiment.

The goal was to see how things turn out if I record and transcribe myself. So these things like style and glue index should be fixed at the source, meaning that I should fix my speaking instead of fixing the aftermath i.e. the transcript.

So I'll just leave most of these writing “defects” in and try to speak better the next time.

How long did it take?

For the part until the section named So how did this meta experiment turn out? it took:

• 25 minutes to correct the transcript into something that I like. It would have taken less if I only corrected the words instead of also adding some general formatting like paragraph breaks.

• 1 hour to edit the transcript into something that I felt was okay to be posted as a blog post.

I'm not adding any images, PDFs, and what not to this 1 hour. I’m also completely omitting the amount of time it took to record myself speaking.

That took considerably more, but I already mentioned before that speaking is a matter of practice and I’m nowhere near as comfortable doing that as I am writing.

Can this be used for other types of writing?

If your aim was to write something short form, like ad or web copy – I don’t think this would work. When writing short form copy, you bash out variation after variation for the same use. Speaking into camera for this wouldn’t be too useful.

One case of writing where you might benefit from it is writing presentation material. If you have an idea of what you want to say in your head and just record it, it might work as a shortcut to sort out a bunch of quirks in your presentation.

Alternative non-writing uses

Where I see the approach and this experiment being most useful is in a completely different scenario than help with writing.

Once you record something like this and have the source video and audio files, you have virtually endless possibilities to reuse this content for other channels.

To name the most straightforward uses:

• The audio recording can become a podcast

• Shorter video bits can become social media posts

• The long video can go on YouTube

Of course, editing the above will take time and effort, but it’s worth it. I’ll test the concept out in one of my future experiments and see if there’s any tech to make it more efficient.

Let’s give this the obsolete marketer score

Ease of use 🟢🟢🟢🟢🔴

For ease of use, technology-wise, it's four out of five.

There's barely any learning curve, and the automated transcript is pretty good the moment you get it.

Productivity 🟢🟢🔴🔴🔴

As a productivity boost in terms of time savings, it's two out of five.

You most likely won't save time with this unless you're really good at speaking, hate writing, and will delegate editing to someone else.

And as I mentioned, the real advantages of this might be hiding in content repurposing instead.

Usefulness 🟢🟢🟢🔴🔴

In terms of relevance and usefulness to you as a marketer, it's three out of five.

You can definitely use the tech to transcribe something that has been recorded anyway and save a bunch of time. You could also repeat the approach in this experiment once just to learn the typical phrases that you use and some other things that will help you write how you speak.

Obsolescence 🟢🔴🔴🔴🔴

For likelihood to make you obsolete. It's one out of five.

Unless transcription is a huge part of your job, I don't see this replacing you anytime soon. If you offer transcription services, this can actually help scale your business instead of taking it away.

That’s a wrap!

I hope you found the experiment at least mildly useful or entertaining. And if you'd like to suggest your own experiments or just have a quick chat, join in on Discord.

It's just starting out, so I can't promise you a horde of like-minded people, but you'll surely get to nag me with questions.