ChatGPT, the OpenAI chatbot that’s taking the internet by storm, is very good at impressions – and that presents both opportunities and risks.

It does a convincing impression of how a human naturally writes, even though it’s not a human. (No potential for exploitation there…)

It does a highly convincing impression of an authority that knows what it’s talking about – even when it really doesn’t. (Not exactly what you want in a search engine, Bing.)

And if you ask, it can attempt an impression of your brand. And whether that impression is convincing or not tells you something about how you’re coming across online.

What are ChatGPT’s responses based on?

ChatGPT was trained on about 300 billion words of online text, including articles, books, Wikipedia and other websites, and then fine-tuned by human operators.

When you give the chatbot a prompt, it puts together its response word by word, gauging the probability of different words coming next in the sentence, based on patterns in those 300 billion words of online writing.

This is what makes it a less-than-trustworthy source of facts. It’s not actually looking up people’s queries – or factchecking its answers (even if that’s what you explicitly ask it to do). Instead, it’s making an informed guess about what an answer to your question should look like. It is often right, but you should still always factcheck its answers against other trusted sources.

The way ChatGPT works also makes it an unoriginal writer. While it can put words together in novel combinations – rephrasing rather than copy-pasting its training data – it can’t come up with new insights or ideas.

So if you’re considering asking it to write your next thought leadership piece, bear in mind that what you’ll probably get is a solid impression of a thought leadership piece based on lots of existing thought leadership pieces … which sounds like the opposite of thought leadership.

ChatGPT prompts give you brand insights

ChatGPT may be unoriginal – but to put it a more positive way, it’s an excellent mimic.

How does that help your brand? Well, a good impression can reveal things to the subject that they might not realise about themselves. You know the tone of voice and personality your brand is supposed to have; ChatGPT can do an impression of what your online content is actually like, allowing you to judge whether the aspiration and reality align or mismatch.

Try giving ChatGPT these prompts:

  • Write three tweets in the style of [your brand].
  • Write an elevator pitch for [your brand].
  • Write an article about wellbeing in the style of [your brand].

Don’t worry too much about what the answers say – just look at how they say it. Do ChatGPT’s tweets sound like your brand, or could anyone have written them? Is the personality you aspire to coming across in the article? Is ChatGPT’s idea of your elevator pitch punchy and compelling, or generically corporate?

If you like what you see, that suggests your online content has a distinct and consistent personality. If not, it could be time for a more in-depth content audit and better tone of voice guidelines for your writers.

Innocent Drinks (@innocent): We're not saying our smoothies have superpowers, but they do make you feel pretty super! 💪🏼🌟 #SuperSmoothie #FeelGood #InnocentDrinks

One of ChatGPT’s responses to “Write three tweets in the style of Innocent Drinks.”


This is a plausible Innocent tweet – it’s got that breathless exclamation mark and some playful hyperbole. These elements must come across clearly and consistently enough in Innocent’s tweets that ChatGPT has picked up on the pattern.

Metropolitan Police (@metpoliceuk): We're committed to keeping London safe and secure - if you see anything suspicious or have any concerns, please don't hesitate to contact us. 🚨👮‍♀️👮‍♂️ #LondonSafety #MetPolice #StayVigilant

One of ChatGPT’s responses to “Write three tweets in the style of the Metropolitan Police.”


Other than the cheeky emoji use – apparently ChatGPT thinks every tweet ends with some emojis and three hashtags – this is significantly more sober, urgent and weighty. The Met Police’s personality is miles away from Innocent’s, and comes across strongly enough for the bot to nail the distinction.

Beyond the middle of the road

Don’t despair straight away if ChatGPT’s impression of your brand is no good. This technique works best for big brands with lots of online content; if your brand has a smaller online footprint, ChatGPT will be less able to predict the kind of language usually associated with your brand name.

The chatbot’s main training dataset cuts off at 2021, so this technique is also less useful for newer brands or those who have recently changed (though it can continue learning from the people chatting with it now).

Redhouse (@redhousebrand): At Redhouse, we help brands tell their stories and connect with their audiences in meaningful ways. Let us help you make a lasting impression! 🚀📈👍 #BrandStorytelling #Marketing #RedhouseAgency

One of ChatGPT’s responses to “Write three tweets in the style of the London brand agency, Redhouse.”


Our positioning used to centre around creating meaningful connections with audiences, and you can see elements of that language in this response. We must have been consistent about it – which is good! But our positioning has evolved since then, and ChatGPT’s impression of us is still stuck in the past.

ChatGPT’s talent for mimicry can still be useful for newer or smaller brands, though. Say you’re a challenger bank: try prompting the bot to “write an article about mortgages in the style of a bank”. (Not any specific bank; just “a bank”.) This will show you the kind of language and style ChatGPT associates with your sector in general: the middle ground you need to stand apart from. This can help you define a tone that’s distinctly yours. The bot’s unoriginality can help you be more original.

Understanding how ChatGPT and other machine learning apps do what they do is the key to finding ways they can be useful.

If you try our suggestions and don’t like ChatGPT’s responses, we’re here to help with the next step.

Tell us what you found