Be proud of what you are putting out into the world

Anna says: “As AI becomes part of our workflow, ethical SEO means that you should be creating content that you're proud to put back into the world.”

Does ethical SEO mean human-created content?

“Not necessarily. I'm a huge believer in AI for helping us do things. It's more about human influence and putting a lens on content, so that we have oversight and can give it checks and balances.

As clever as AI is, sometimes it doesn't understand nuances and the issues that it has.

When you're looking at AI output, you need to remember that it picks up biases and strange ideas as it’s going along. It’s really good to have a diverse range of humans involved to understand what kind of content we need to be creating to put back into the world.

I work a lot in disability forums. I've got ADHD, so I am neurodivergent, and my son's autistic. I am also classed as disabled because I have fibromyalgia. Therefore, I understand when language may be ableist or isn’t taking people with disabilities into account. That means that I can put a human lens onto my content that AI misses. It doesn't understand these nuances of humanity.”

What’s the commercial value of creating content that you're proud to put out into the world?

“This is the problem, and it’s why the ethical side of things gets overlooked –particularly by the people who create AI algorithms. Ethics does not always equal profit in short-term scenarios.

You can't measure it very quickly. You have to start by measuring things like sentiment and how people feel about your business. I worked at a business a few years ago, and the sentiment towards them was very negative. They were thought of as Big Corporate, their customer service was shocking, and people generally disliked them. We started doing digital PR around building the community, being more ethical, and promoting sustainability.

It was really hard to measure, I'm not going to lie, but we started using things like YouGov and other different methods to understand how people thought about that business – whether they thought we were ethical, whether we weren't, etc. Then, we tried to track that against our growth.

It is very difficult, but if you look at all the stats, everyone (particularly the younger generation) says that ethical brands are the ones that they look out for, and it’s a value they look for. If they're looking for it, that's going to lead to more sales. A lot of small brands that are very ethically aware are getting good traction because of how ethical they are. Patagonia is a good example of that.

From an SEO point of view, people are very aware of AI these days. There's a lot of controversy around it, so we're trying to use it ethically to do good in the world. It can help you no end. It's brilliant at helping you create content, but you've got to think about, when you put that content on your site, is it actually going to benefit anyone? Is it accurate? Is it truthful? Is it good quality content?

When it eventually gets back into the algorithm, when someone searches your name and a bit of content surfaces, are they going to see an accurate piece of content or something that sounds like a ChatGPT hallucination? That could be quite damaging for your business.”

Is it possible to be proud of AI-produced content, and where do humans fit in to ensure that the content reflects those key values?

“I think humans need to be involved at almost every stage, even in the prompting that you use.

I recently did a talk called Amplifying Inequalities: The Hidden Bias in Data, and I did a test, where I went to ChatGPT and said, ‘Create an image of a startup founder.’ It created this image of a mid-twenties guy who looked like a tech founder. I asked this question across several chats, and it created the same image every time. I asked Gemini and Copilot, and they created the same image as well.

At my talk, everyone in the room put in the same prompt, and it created the same image for all 25 of us. There is an inherent bias (and sometimes untruths) in AI, so you have to be careful with how you prompt it. I could have put, ‘Give me a diverse range of images.’ For text, you can say, ‘Make sure that I'm including neurodivergent voices and considering different cultures.’

If we know that there's an inherent issue with a system, it's about us, as humans, to push and craft it to produce something that is less biased. It's being aware of the issues that are there.

Then, once you've got the content, you have to edit it and tweak it yourself. Make sure there’s a human in the loop at every stage: the prompting, the reading, etc. Get someone to edit it or have a look over it so that you can get a few pairs of eyes on it.

Also, the input that you give it. I put a lot of my own generated content into ChatGPT, so it uses that to generate stuff. If you're uploading images, make sure that they're a truthful, accurate representation. If you're uploading content, make sure that it's high-quality. That way, you can influence the output as well.

Then, afterwards, you can do regular content audits to make sure that your stuff reflects the brand and its audience. Also, get your own data to feed into it, rather than using ChatGPT to find stuff. Do surveys and feed that back into it, so that you've got a representation of your data rather than the AI’s.

I also do regular searches on the different LLMs to see what comes up when I'm mentioned, so I know that the information being picked up is accurate. It’s a multi-stage process.”

With the prevalence of AI-generated content, is it becoming more difficult to produce content that stands out and isn’t just a generic reflection of the majority of situations?

“It’s interesting. I quizzed ChatGPT on why it came up with that same image of a startup founder, and it said that it was just what it perceived a startup founder to be. When one of the attendees asked the same question, it gave the same answer.

I know very well that a large number of startup founders are not in the tech industry, and their average age is about 35-40. Therefore, statistically, that picture did not even match the average startup founder in the world. It was more of a perceived average. It was a young person (which is what we all perceive a startup founder to be) but statistically, they are older than that.

It was interesting that it wasn't using statistics; it was using some sort of societal perception of what a startup founder would be, which is based on the current content that exists out there. The problem is, if we all use AI to generate that same image of a startup founder, and we put it on our blogs, and then AI reads those blogs and sees those images, we're perpetuating that myth. It’s a self-fulfilling prophecy.

You've got to be really careful. You've got to think, is this statistically true? For example, I know that 1/3 of UK doctors are from an Asian background and 50% of GPs in the UK are female, because of research I have done for a pharmaceutical client. I did a general Google image search for UK doctors, and there was very little representation of female doctors and a very small percentage of Asian doctors.

That did not actually represent the doctors that we have in the UK, yet ChatGPT is probably seeing those images and thinking, ‘This is what UK doctors look like.’ This is why we need industry experts, because they have an awareness of what's accurate and truthful. If I looked at that list of doctors and I didn’t already know the stats, I would have thought, ‘This looks a little bit diverse, so it’s probably accurate.’ However, because I knew the details and the stats, I knew that it wasn't representative.

This is why we need people to understand their industries and their customers. In Matthew Syed's book, Rebel Ideas: The Power of Diverse Thinking, he calls it a frame of reference, where everyone has their own frame of reference and their own history. If a doctor had looked at that list, he might have gone, ‘That doesn't represent me or my colleagues. That looks very different.’ However, if someone's not in the medical industry, they may not have noticed.

It's important that you have people with this knowledge of what your customers look like and what your industry looks like, so that you can make sure people are represented – not just in imagery, but in tone, in mentions in the text, etc. It's really, really important.”

By using real-life experts and customers, can you feed AI with correct information to improve the accuracy of what it produces?

“Exactly. I do that with my podcast. For all of my podcast recordings, I take the transcriptions and feed them into ChatGPT. That is real-world advice from real people in the industry, which fills the AI with high-quality content.

If I'm going to write a blog, I've got a specific ChatGPT project that I use, which has all of these different stories and pieces of advice from founders and experts. If I need to produce AI content, I will use that project, so I know that it's coming from a place of accuracy.

NotebookLM is really good for that. You can shove all of your content in and create your own mini database, and then create content and get ideas from it, so you're producing something that you know is of quality. As they say, ‘garbage in, garbage out’. AI is a bit of a black box. You don't know what's gone in there. Do you trust what it's putting back out? Possibly not.

I'm really careful. If I do use it without my own content, I quiz it, I click through to every link, I have a quick read of the article, and I make sure that I'm happy with this content. From a professional reputation perspective, I don't want my content (or the company’s content) to be a poor representation of us. It's got to be high quality.

I'm reading Mark Schaefer’s new book, How AI Changes Your Customers: The Marketing Guide to Humanity's Next Chapter, and he mentions that he got his first customer through AI, when someone put in, ‘top 10 marketers in the world,’ and he came up. If someone then searched some of his stuff on ChatGPT and it was of poor quality, they may feel like they don't trust him.

Interestingly, from an SEO point of view, Muck Rack has analysed AI responses, and they say that approximately 95% of them derive from non-paid content sources. The three main areas are corporate blogs/owned media, journalism, and academic research. That shows you how important this content is, because it's directly feeding those AI responses.”

Anna, what's the key takeaway from the tip you shared today?

“Be careful. Put good stuff into AI to make good stuff come out.

Don't just rely on the black box of AI. It’s a brilliant tool, but you need to put the human into it. Make sure you're producing quality content, using your own stuff and your own sources, that you know is high quality.”

Anna Bravington is Business Advisor and Founder at Creating the Curve, and Innovation Director at Oxford Innovation. Find out more over at CreatingTheCurve.co.uk.