The Online Optimism logo
ServicesSchedule a Call
See All Posts
Google's preview image of their AI Overview feature, showing it beginning to generate an answer for the query "what are good options for a..." on a mockup phone screen with a light green gradient around it.

Google’s AI Overviews Are Introducing Dark Patterns

Ivey Barr
Ivey Barr
Ivey Barr is the Senior Developer at Online Optimism. Learn more about them here.

At this point, everyone has seen the AI Overview section at the top of their Google search results: a panel of instant answers, magically generated from various search results somewhere further down the page. As of May 2024, the feature left Google’s experimental Search Labs and began rolling out to all users — intentionally or unintentionally introducing new dark patterns into the user experience.

A screenshot of a Google search for “why is ai in everything now,” with Google’s AI Overview of the results displayed at the top.

What is a dark pattern?

The term “dark pattern” was coined by UX designer Dr. Harry Brignull, founder of the Deceptive Patterns Initiative, to describe interface decisions that manipulate, trick, or force users into taking an action. Familiar examples include confirmshaming (putting “No thanks, I like paying full price” on your pop-up ad’s close button), obstruction (making it hard for a user to perform the action they want), and hard-to-cancel aka “roach motel” (making it easy to sign up, but hard to quit).

Dark patterns explicitly benefit the company at the expense of the user, and they’re not exactly a great experience for anyone. Software engineer and responsible tech advocate Selam Moges gave an excellent presentation at the recent International Javascript Conference on how these patterns show up in common scenarios, and the ethical alternatives — which can range from simple wording changes to more empathetically rethinking how the business interacts with its customers.

Let’s take a look at how this shows up in Google’s AI Overview feature, why it’s a problem, and what Google could do about it.

Issues with Google’s AI Overview

When Google announced they were rolling out AI Overview to all users, their blog post made it clear that they saw it as a win for their users: get answers more quickly, ask more complex questions, or even search with a photo or video. Google CEO Sundar Pichai speculated that many people may not even realize they’re using AI at all when they see the overview, normalizing the interaction to the point of invisibility.

Not everyone is thrilled with the feature, though. Users are constantly asking how to turn it off. (Search Engine Land notes that while Google says users love AI Overview, they have yet to present any data to back that claim.)

A user asks how to turn off AI results in the Google support forums in May 2024, saying the feature ruins their browsing experience and they’re considering switching browsers. The question has been marked as a duplicate and replies have been locked, and the “I have the same question” button has 4,954 upvotes.

It frequently “hallucinates” misinformation and absurdly incorrect answers, leading to viral coverage across multiple outlets as it advised people to run with scissors and eat rocks, reported cats on the moon and dogs playing professional sports, and gave misleading medical advice. (It also spoils TV shows and games, since AI hasn’t quite wrapped its emerging mind around the concept of not wanting certain information.) Newer versions are even worse than the originals, with OpenAI’s latest error rates ranging from 33% to a whopping 79%.

Beyond its questionable utility, many people are deeply uncomfortable with the environmental impact of AI: a ChatGPT request is estimated to consume more than 9 times the electricity of a regular Google search query, and Google itself cited its AI ventures as the reason for its 48% increase in emissions over their baseline in 2019 (the year that they set a goal of net zero emissions by 2030). If current AI development practices remain unchanged, then instead, by 2030, AI is predicted to use more energy than the entire human workforce and significantly offset our already way-off-track efforts to reach net zero worldwide.

The virtual world isn’t benefiting, either. As far back as 2019, 50.33% of Google searches resulted in zero clicks to another website, and a few months after introducing AI Overviews, that number has reached 58% in the U.S. and 59.7% in the E.U. That impacts businesses trying to attract visitors to their sites, as their content gets pulled into the answer panel and shown to people before anyone even notices their website is the source. In the words of SEO consultant Duane Forrester, “the best you can hope for is to be used.” (If you’re worried about how to adapt your own content to show up in this new frontier, our team has excellent practical tips for you.)

That’s a lot of different concerns. The simplest way to address them? Don’t like it, just don’t use it.

And that’s where the dark patterns come in.

Dark patterns in Google’s AI Overview

Google’s AI Overview is now, according to their own FAQ, a built-in feature that can’t be turned off. Their official advice to users who dislike the feature is to use the Web filter after running a search —click “More” in the filter bar, then click “Web”.

A screenshot of a Google search for “why is ai in everything now,” with Google’s AI Overview of the results displayed at the top. Underneath the search bar, the “More” dropdown has been expanded to reveal an additional three filter buttons, including the highlighted “Web” filter.
The same Google search, now with the Web filter turned on. The AI Overview is gone, and the first two regular Google search results can be seen instead: a Reddit post and an article from Fast Company.

Note that this multi-step workaround doesn’t disable the AI Overview at all: the user still has to perform their search, see the AI Overview of the results, and then switch to the filtered view. So for the ethically-opposed segment of unhappy users, this solves nothing, and users who just don’t like the experience still risk misinformation, spoilers, and general unreliability before they can navigate over to the other view.

In its current state, AI Overviews demonstrates dark patterns including:

Dark Pattern #1: Forced Action

Forced action is a dark pattern in which the user wants to take an action, but is required to do something else undesirable in return.

  • Simple example: A popup that asks the user to provide their email address, without making it clear that the step is optional.
  • In AI Overviews: A visitor to Google wants to search the web, but is required to perform an AI request and view its response before seeing their search results.

Dark Pattern #2: Obstruction

Obstruction is a dark pattern in which obstacles are placed between a user and the task they want to complete or the information they want to access.

  • Simple example: An opt-in process where the opt-in option is presented clearly with a one-click button, but the opt-out option requires multiple clicks and confirmations.
  • In AI Overviews: The user must click “More” and then “Web” in order to see the search results they want to see.

In Google’s case, this also constitutes visual interference: the desired action is hidden and not clearly labeled. (Neither “More” nor “Web” clearly communicate “View results without AI Overview.”)

Dark Pattern #3: Preselection

Preselection is a dark pattern where the user is presented with a default option that has already been selected for them.

  • Simple example: A donation form with a pre-checked checkbox to make the donation recur.
  • In AI Overviews: The overview is shown by default on whichever searches Google chooses. The user has no access to unselect this option.

Fixing the dark patterns

Google’s gotten in legal trouble for dark patterns before, and it cost them $95 million. In order to stay on the safe side of UX law, they have plenty of options:

Make it optional

The simplest answer is still the simplest answer — Google just needs to implement it.

  • Allow users to opt out of AI Overviews at the account level, and respect the setting across all Google properties.
  • Add a small “Turn off AI Overviews?” link at the top or bottom of the AI Overview panel, so users can go straight to their account settings page to opt out.

Opt into AI Overviews

It would take only slightly more effort to go the other way around, which also eliminates the preselection dark pattern.

  • On search, default to showing the regular search results without the overview.
  • Add a “Show AI overview of these results?” button at the top of the search results page.
  • When a user clicks to show the AI Overview panel, show the panel, and a link to “Always show AI overview?”
  • If a user clicks that link, remember their preference and continue to show them AI overviews!

Rethink where AI is most important

The last option is far more difficult, but could have important future repercussions as we continue to develop and regulate the usage of AI. Google is throwing its efforts into Project Astra, a “universal assistant” designed to make any request as simple as conversation, among dozens of other AI initiatives including image generation, music generation, video generation, code generation, organizing research materials, asking your photo gallery to answer questions — the list seems never-ending. They’re also aware of its environmental impact, and have joined several other companies looking into ways to use AI to protect the planet. And they still have to watch their budget: Google’s projected 2025 spend on AI is “approximately $75 billion,” according to Pichai, of which a significant chunk goes to servers, data centers, and other infrastructure.

Researcher Alex de Vries points out that, logistically speaking, every single Google search literally can’t become an LLM interaction — in order to make that happen, Google would have to consume as much electricity as the entire country of Ireland, spend somewhere around $100 billion to purchase more servers than NVIDIA could produce, and spend another $33 billion or so over the next 3 years just to operate and maintain all those servers. Given so many other areas to focus their energy, and the incredible resource drain of a feature that’s badly designed to the point of introducing dark patterns to their everyday users, Google has a clear opportunity to rethink their AI strategy.

Dropping AI Overviews entirely, at least until they’re able to refine it to a more reliable state, opens up new opportunities for Google to expand its AI work in ways that benefit their users far more than a buggy answer panel.

Why continue wasting time, money, electricity, and developer efforts that could instead be invested in improving MedGemma’s ability to identify tumors in CT scans or Astra’s ability to describe surroundings to low-vision users? Google has some incredible initiatives in development, and AI Overviews isn’t worth the energy it’s stealing from them.

|

Tags: Conference, Resources

More Posts by Our Team