Recently, some social media users have been complaining about a trend with Midjourney.
If you aren’t familiar with it, Midjourney is a tool that allows you to generate images using artificial intelligence. Of course, the way it works is by drawing from other, human-made images that have been posted on the Internet to compile a new image; one that is entirely unique. An original work of art.
The issue, then, that arises, is that (after Midjourney being available for months now) the Internet has become already flooded with artificially generated art, much of it from Midjourney. So now, when the software tries to draw source images from online, it ultimately ends up self-plagiarizing. It uses AI-generated images as its source to create AI-generated images, and, over time, with fewer original images to draw from, the images become worse and worse in a sort of AI-inbreeding, also known as model collapse. (I didn’t invent this idea.)
The Google Update
Perhaps you’ve already seen it, but Google has been incorporating a search feature that uses AI to auto-generate a response to your search query. It does this by scanning common answers in the articles written in response to that keyword and compiling them into one brief answer.
When this feature was announced in May, an article written in Forbes declared it “a nuclear bomb [for] an online publishing industry that’s already struggling to survive.”
But what will it mean for content writing? What will it mean for the people who write the keyword-optimized articles designed to appear under a Google search query?
Well, a lot of content writers were already a bit shaken up by Chat GPT and GPT-4 coming onto the scene. While Chat GPT still produces rather surface-level, robotic-sounding content, there are many websites that don’t care as much about the quality of the writing on their site, and just need a large quantity, so AI does the trick. Just like AI-generated art, AI-generated writing is already filling up the web.
For many Google searches, even without the AI-generated answer up top, the issue has already started. The above photo is of a Google search that brings up an article written by ChatGPT with clearly incorrect information in it.
Google’s AI-generated answer to search queries was likely released hastily as a response to Chat GPT. When Chat GPT started becoming popular, many people even stopped using Google for their search queries and simply asked the robot. While the robot is far from perfect (it still hallucinates, it still gives false information, it still sometimes plagiarizes word-for-word), there were clear reasons to use it over Google.
In fact, Google was already being criticized for compromising the integrity of its product in order to make the product more profitable.
And they weren’t in an easy spot. Not only is the search engine completely free, but it depends on content made by users, and people don’t like to make content for free either. So how do you incentivize people to create free content?
The answer is that, well, they’re often not really making it for free. They have to be making money off it somehow. Maybe the article is making money through affiliate links (they get a kick-back from the products they recommend), maybe it’s through ad space around the article (not always super lucrative), maybe they’re leaving cookies on your browser and collecting your data, or maybe they’re engaging in content marketing for a product. Regardless, if they’re filling up the internet with content that isn’t behind a paywall, then they need some incentive to keep doing so.
The fear with this new feature that uses AI to summarize all answers at the top of a SERP is that it will take away the incentive for websites to create new content.
If people get all their answers at the top of the page, then they don’t need to enter into the articles below, they don’t need to click on affiliate links, they don’t need to open the page and exchange cookies and look at ads, and many of us probably won’t.
It will probably hurt traffic for many websites that depend on that traffic to make money, and many websites will either conclude that it’s no longer financially viable or rearrange their digital marketing efforts elsewhere.
AI, of course, doesn’t need to be incentivized. This may seem like great news: Just give the search queries to the robot, and the robot spits out answers. But we should recall that the AI draws its answers from the human-written content below. And if we take the incentive to write new content away, and throw in the fact that many websites are already publishing Chat-GPT-written content, then the AI starts drawing from outdated or AI-written content, inevitably making its answers worse. Inbreeding.
In order to ensure that this new AI-generated answers feature (becomes and) remains relevant and useable, it will have to find ways to continue to incentivize SEO article creation and content writing, at the very least so that their AI-generated answers have something to harvest.
Fortunately, there are a few ways Google could encourage SEO marketers to continue writing content.
Promote Websites with Blogs/ A Lot of Written Content (Even if the Articles Themselves Aren’t Ranking)
Maybe a website has a button on its homepage that generates the vast majority of the business’s online sales. In that circumstance, the business wants their home page or their sales page to rank well organically and they may ask how to do that. Google can answer them with this: “We ‘reward’ sites with more content with a higher ranking. If you want your home page to rank better, build a robust and accurate blog in your niche.”
Note that in that circumstance, the individual blogs themselves wouldn’t necessarily be bringing in traffic and converting. They would instead be supplementary; allowing for the home page to rank better because of their existence. If they themselves begin to rank and convert, or even just offer more brand awareness higher up in the sales funnel, then that could be an added benefit.
What Google gets out of it: More recent content online (for their AI to reference).
What the businesses and websites get out of it: Their home page gets a better ranking for when people Google commercial-intent keywords.
What the writers get out of it: Websites paying for writers to create blogs in hopes of gaining a better home page ranking.
Announce a Preference for Human-Written Content
AI detectors are famously inaccurate and prone to giving false positives. We also don’t believe that the answer to all of AI’s problems is to not use AI at all. (You can read a more in-depth post about our agency’s AI policy here.) For that reason, in terms of actionable policy, it would be difficult to “punish” AI-written content or to “promote” human-written content.
I think an announcement would be all it would take to get fingers on keyboards.
Google has said that they don’t penalize AI-written content but that they can tell when content is spammy or keyword-stuffed. I would take that a step further and note that (and I have run many, many articles through plagiarism checkers) AI tends to plagiarize, often without you noticing.
Generative AI doesn’t plagiarize like a person. A person might copy/paste from a single source and try to switch the words up. But many articles written by AI come up as plagiarized from 15-20 different sources. Perhaps it took no more than a sentence from each source, but when the article is completed, sometimes it could be 20-25% plagiarized.
Google should not be afraid of saying the word “plagiarism” in the conversation about generative AI, and discouraging the overuse of AI in response. We know that plagiarized content online is punished in Google search rankings, so Google should directly tell people to be very careful about publishing AI-written content, especially without first having a detail-oriented human editor go through it with a flea-comb.
Google should announce that, while they aren’t anti-AI, they have a preference for human-written content. That announcement alone would be enough to increase the quality of online material.
Offer Public Recognition or an Award to Scientists and Researchers who Share Online Content
Right now, Google is still a place where people go when they want to find trusted information. Unfortunately, the way that Google is currently approaching AI can easily erode that trust at a time when the promotion of accurate information, especially in topics related to health and science, is more important than ever.
Educational and research institutions recognized by Google Scholar should be further incentivized to publish understandable content in layman’s terms through a blog or another medium that isn’t behind a paywall.
With academics facing burnout and packed schedules, a major incentivizer for them could be some form of public recognition; something that can be featured on a resume or celebrated by their community.
A monthly or yearly award offered by Google (potentially in partnership with a prestigious academic institution and potentially with a cash prize) to the academic with the most notable, engaging and persuasive blog could encourage thousands of scientists and researchers to publish large swaths of useful web content.
Kate Maximov, a Ph.D. candidate in the Johns Hopkins Neuroscience program, put it this way: “Many [academics] want to become more involved in scientific communication and outreach, but are overbooked with commitments that may be tailored to landing highly competitive academic roles or trying to achieve tenure. Academics may be more incentivized [toward outreach] if they could gain some recognition or reward for their efforts, which could be added to a CV or resume.”
She added that it would be helpful if a prize highlighted the skills that writing these articles requires and how those skills are important for a variety of roles in industry or in academia.
While there are many ways for scientists to partake in outreach, Google can create an award focused on accessible online content, like blogs.
I don’t believe that it’s hyperbolic to say that if Google doesn’t find ways to incentivize the creation of higher-quality written online content, then it could become obsolete. The answer won’t be throwing more AI on top of the AI and hoping that makes the problem go away. The answer is finding ways to keep human beings writing and posting online. That way, we work in tandem with the robot, and not against it.
Awards, announcements, and ranking manipulation are good places to start. But overall, Google must first recognize that in any problem presented by generative AI, its most important allies are writers.