Should we be worried about the emergence of AI technologies?

Are we in the midst of an AI revolution? Is AI going to take my job? How long before I have to take up arms to fight killer robots that look like Arnold Schwarzenegger?

All very valid questions. When I look through any of my social media feeds recently, I cannot seem to escape the consistent stream of Chat GPT tips, information, how-to guides, or money-making schemes.

The impact of ChatGPT is clear – figures show the software has hit 100 million active users in record-breaking time, making it the fastest-growing consumer application in history. For context, it took TikTok nearly 9 months to hit a similar number, and Instagram over 2 years[1]. The rise in popularity of ChatGPT forced Google’s hand, announcing their AI chatbot Bard and subsequently whopping 100 billion off their valuation after a muddled launch with fumbled answers (whoops) [2].

[1] https://www.reuters.com/technology/chatgpt-sets-record-fastest-growing-user-base-analyst-note-2023-02-01/

[2] https://www.cnbc.com/2023/02/08/alphabet-shares-slip-following-googles-ai-event-.html

During the announcement, Google Bard fluffed its lines when asked, “What new discoveries from the James Webb Space Telescope (JWST) can I tell my 9-year-old about?” Google should be ashamed.

But wait, does this mean that ChatGPT always gives the correct information? Or have we been caught up in the sci-fi fantasy overlooking errors, ethical issues, and data bias? Maybe…

Recently, the use cases of AI and machine learning have come under scrutiny. With most users asking it wacky questions or getting help with their homework, there are reports of ChatGPT being used to create malware[1]. A worrying thought that this tool can allow those with little programming experience to create malware just by asking it the right kind of questions.

This goes further than just ChatGPT – a recent Vice article shows how AI-generated voice technology was used to hack into a bank account that used voice recognition software[2]. Scary stuff.

This throws up a lot of ethical questions – how do we stop AI from being used for deviant purposes? Who polices or manages these open-source systems? And so on. It seems like we have a lot of things still to work out.

We are only touching the surface here. We have data bias (highlighted brilliantly in the Netflix documentary Coded Bias[3]) and rogue AI’s such as the Twitter chatbot in 2016 which generated racist content before being shut down, a CNT news site which produced factually incorrect/plagiarised content, or the Google Photos tagging feature which grouped photos under racial stereotyped names[4].

We can dig further into AI issues, but what really astounds me is the data inaccuracies in ChatGPT and how we are largely ignoring them in favour of promoting the platform. You can see some of the inaccuracies here for yourself[5].

This is why at SYZYGY, we do not see the emergence of ChatGPT as a job killer or creativity killer, and I am pretty confident I do not need any sort of firearms training (at least in the near future). But we do see the productivity benefits of the software, and I want to briefly go through how we at SYZYGY are using it to benefit us and our clients.

[1] https://cybersecuritynews.com/chatgpt-build-malware/

[2] https://www.vice.com/en/article/dy7axa/how-i-broke-into-a-bank-account-with-an-ai-generated-voice

[3] https://www.netflix.com/gb/title/81328723

[4] https://www.bbc.co.uk/news/technology-33347866

[5] https://medium.com/@chiayewken/mistakes-of-microsofts-new-bing-can-chatgpt-like-generative-models-guarantee-factual-accuracy-5ec82a7453f4

Ah, remember those times when you knew what you wanted the output from an Excel formula to be, but you had absolutely no idea what formula you needed or how to even Google the question? Before long you have spent the good portion of an hour trying and failing to find a formula that, for example, would split the domain out from the referring page URL in an Ahrefs export.

Well never fear! ChatGPT has you covered.

Here at SYZYGY and for me personally this has been a game changer, ChatGPT, in our experience is able to create accurate Excel formulas based on simple details.

Of course, to get the right output you need to ask the right questions, so before you start cursing ChatGPT and its creators, why not consider how you ask your questions and the details you provide.

A few examples of Excel formulas we have used ChatGPT to create are;

• A formula to isolate the domain from a URL.
  • Very useful for backlink analysis especially if you want to create pivot tables based on linking domains for example.
• A formula to categorise keywords by common words.
  • Great for managing larger keyword lists when you know that certain keywords include a specific intent (how, what, why who keywords are questions or information terms for example)
• A formula to split URLs into categories and subcategories based on their URL structures.
  • Great for analysing website and URL structure/hierarchy.

We have all been there, we are trying to use a filter in Google Search Console, and we get a screen like this. We know that this collection of keywords has driven clicks and impressions so why oh why is our filter returning zero results?

Again, ChatGPT is easily able to create regex which can be copied and pasted into Google Search Console to segment data.

Use cases for this could be to;

• Segment keywords by questions
  • Is our content working hard enough for us?
• Split brand and non-brand keywords
  • A firm favourite in any/all SEO reporting.
• Segment by facets on e-commerce websites
  • Which colour facets are driving clicks and impressions for example.

Chat GPT can return answers in multiple different formats including.

• Plain text,
• Tables,
• HTML
• JSON

Straight away we can spot some uses cases here.

ChatGPT can be used to create;

• Schema, this is very well documented across the industry so I am not going to keep going on about it.

But it should be noted that Schema can be updated or changed and as ChatGPT only has a data base up until 2021 some Schema information may be missing.

• Hreflang code and Hreflang sitemaps, most impressively ChatGPT seems to the use indicators in the URL to make assumptions on the country and language a given URL targets without the need for data input.

But this is limited, it worked well for homepages but as soon as I added a subfolder to my imaginary website (please forgive the lack of creativity in the name) it hit a snag.

Again, the emphasis here is on the request rather than the machine learning model.

• Creation of redirect rules
  • Redirects can often make or break a websites performance after a migration and URL change. So why not leverage ChatGPT? We can use Chat GPT to create.
    • Bulk re-direct rules when URLs are changed following a subfolder move for example.
    • 121 redirect mapping for anything which is slightly more complex.

• Finally, how about using ChatGPT to update and test robots.txt directives.
  • Want to create and sanity check new rules in the robors.txt file. No problems!

At SYZYGY we do not feel like ChatGPT will be writing content for us any time soon. A few core problems such as a lack of an up-to-date data base, some fact fabrication, issues around intellectual property and so on mean that in its current form we use this tool for expansion rather than creation.

What ChatGPT can do very well is expand on topics, pulling out additional details and being a ‘digital sounding’ board to ensure we may not have missed any core information or angles relating to a topic.

But be warned!

ChatGPT is not an SEO tool, the topics, questions, and information it provides is not based on search volume or search interest but rather on the data that it was fed and therefore the information will still need to be checked and verified.

Additionally, given the issues we have touched upon multiple times in this post, we suggest not using this tool for anything health or finance related, in essence YMYL (your money or your life) as it could well be surfacing incorrect or historic information.

That being said. Here is a good use case example.

Like many people, my summers are filled with weddings and this leaves with the conundrum, what do I wear?

So, asking ChatGPT for a content outline for what I should wear to a wedding throws out a comprehensive guide on what I should consider when choosing.

We can use this content outline as a basis to build out topic clusters (formal, semi-formal, casual, accessories, seasonal wedding etc.) and can easily lift questions from the content outline (what to avoid wearing for example) to build out our own comprehensive content list.

We can then cross reference this information with our own briefs to ensure we have covered all basis.

I may have overreacted at the start. I do not think an AI Armageddon is coming our way, but I am very wary of all the overly positive information regarding ChatGPT especially when I hear about SEO use cases.

I would urge everyone to take a step back and consider how they are using the software. Yes, it might have short term growth opportunities such as creating content at scale and some of it will rank, but I do believe Google and other search engines will target future algorithms towards this content (Panda for the new age) to combat this.

For now, we are revelling in the productivity benefits ChatGPT has brought and long may that continue.

…Oh, and one more use case. I asked ChatGPT to proofread this article, so please direct any issues to them.