Everyone is talking about ChatGPT, and it’s with good reason. Bill Gates said in an interview earlier this month that generative AI like ChatGPT would “change our world,” and search giants are in on the action—Bing is partnering with OpenAI (ChatGPT’s creator) while Google is on the brink of launching its own generative AI search companion, Bard.
Practically daily, there is a new advance or innovation, far too much to cover here in this blog post.
Clients are curious about ChatGPT and other generative AI applications. Do we use it? Should we use it? How do we use it?
What’s our philosophy on AI?
At seoplus+, we use any (white-hat, legal!) tool at our disposal to help our clients achieve great results. We also use any resource at our disposal to make a better experience for our team members, whether it’s taking the headache out of finding available calendar slots, automating meeting note capture, or using AI to recommend the next best task to work on.
We subscribe to many different platforms, from SEO keyword research tools to grammar review tools to paid ads budget pacing software, all of which have some kind of AI component built in to save time, increase accuracy, uncover more insights, and generate better results for clients.
When it comes to AI, we’ll try anything once. We’ll test it, evaluate it, and incorporate it into our tech stack if it makes sense for our clients and our team. If it produces poor output, it’s unreliable, or causes frustration for our team, clients, or end-user, we scrap it.
Since ChatGPT has dominated the conversation, it may seem like this is new ground (which it is, to an extent) or that it’s the only AI tool out there (which it most certainly is not).
What’s our philosophy on ChatGPT?
As with almost any other tool we come across, we have been testing ChatGPT and seeing its applications and if and where it might make sense to incorporate this into our workflow.
This is the exact same as it would be for people in any role; finding instructions on how to do something and getting a direct answer vs. having to comb through Google search results, rewording an email to best communicate your point, or coming up with a list of fun ideas for the next social event.
In terms of client work, it can also be a powerful tool that supports the brainpower of our human team. For example, coming up with user personas for a very niche industry might take a lot of time for a team member, and only be based on that team member’s best guess. If you can ask the right prompt of ChatGPT, you can get a detailed and specific result that pulls on trillions of data points.
Where does ChatGPT miss the mark?
Sourcing: At the time of publication, ChatGPT cannot search the internet. Anytime you need it to include links, or ask it to navigate to a URL, it is not able to do so.
Timeliness: At the time of publication, ChatGPT is limited to a data set that is cut off at September 2021. At the light speed the world is going, this is effectively ancient history. For example, ask it a simple question like who won the 2022 Super Bowl, and it’s unable to provide an answer.
Accuracy: ChatGPT doesn’t “know” anything. It is good at producing a facsimile of human conversation. Beware of ever relying on ChatGPT as factual or 100% accurate. This is intimately revealed when you ask it to write about something you don’t know about, versus something where you are a subject matter expert. Its limitations are quickly revealed.
Redundancy: ChatGPT seems to have an “optimal” way to phrase things. It sounds perfect the first time. Then you realize it repeats the optimal phrasing, which makes for a very poor reading experience. For example, I asked it to list the pros and cons of advertising on Google, Facebook, and LinkedIn. The final three “cons” for each platform were word-for-word.
Uniqueness: If I’m asking it to write a paragraph on a certain topic, do you not think one or 1,000 other people are also asking the same question? Unless you are exceptionally creative with your prompts, it would be a mistake to assume the content being generated is custom-tailored and exclusive to you.
Human Touch: AI doesn’t have opinions, feelings, perspective, insight, wants, needs, discretion, or anything else a human possesses, for better or worse, no matter how much it generates outputs that make you feel like it does. AI is at its most useful when it is harnessed by humans and accompanied by human oversight and framing.
Ownership: There’s a debate about the data set that ChatGPT is trained on. If it creates poetry trained on the copyright poems of others, where does this fall in terms of intellectual property? This is a big issue in the visual arts space, with people selling AI-generated art that is trained on human-created art. This is definitely a grey area and something to watch closely.
Where does ChatGPT excel?
I appreciate ChatGPT’s content generation for internal purposes, like if I need help with a script for an internal training video.
For public-facing, it has rarely generated something I feel I can use, at least not without extensive editing and rewrites. This will probably change, maybe even in a matter of weeks.
That said, I’ve been happy so far with my experience with the following prompts:
- Write a creative blog title on [this subject] (PS this blog title is all me, for better or worse!)
- Come up with a creative headline that includes [this keyword]
- What’s a better way to phrase [this]
- Write a 160-character meta description based on [this intro] that will draw clicks
- Can you read [this excerpt] and let me know if there are any issues? Are there any typos or grammatical issues? Is it clear?
- Can you read [this excerpt] critically? Would you be persuaded by this if you were [target audience]?
- Here is a checklist. Can you cross-reference it against [this excerpt] and see if I have hit all the points?
- Is there anything I could add to [this excerpt] to make it stronger?
- Help with writing conclusions (does anybody like writing conclusions?)
Again, to be clear with some of these prompts, ChatGPT does not actually read content “critically,” it is not capable of human thought. But it can generate a response that mimics feedback well enough to point you in the right direction, versus having no critical review of your writing at all.
You can also be exceptionally creative with your prompts. Ask basic and generic prompts, get basic and generic output.
Where does generative AI go next?
We’re only scratching the surface, and this article might well seem quaint in a few months. I’m personally most excited for when you can train these generative language models on your own content. We’ve had this vision of a central repository where team members can pick up everything they need to know for their roles, from how to make a benefits claim to how to edit a certain PHP module on a client’s site. For our clients, this could be a chatbot that answers everything a visitor needs to know to make an informed purchase. These seemed like pipe dreams a few months ago, and now it feels like this kind of functionality is just around the corner.
At seoplus+, we explore any and all tools that can benefit our clients and their end-users. I believe when you combine the speed, breadth of data, accessibility, and computing power of AI with human expression, creativity, and insight, you can achieve almost anything. ChatGPT, Bing’s AI-powered search, and Google’s Bard are only the tip of the iceberg.
Disclaimer: ChatGPT was NOT used in the production of this article. 😅