Nieman Foundation at Harvard
HOME
          
LATEST STORY
Journalism scholars want to make journalism better. They’re not quite sure how.
ABOUT                    SUBSCRIBE
Aug. 29, 2024, 1:58 p.m.
Aggregation & Discovery
Audience & Social

Here’s how 7 news audience directors are thinking about Google’s AI Overviews

Google’s generative AI search feature is here to stay, but will it actually impact how digital outlets do business?

Earlier this summer, AI Overviews made headlines across the U.S. — mostly for the wrong reasons.

The new Google search feature offers up AI-generated answers to queries, synthesizing sources on the web and filtering them through Google’s own large language model, Gemini. In the days after its U.S. roll out, journalists quickly took to social media to pan the product, screenshotting Overviews that were filled with factual inaccuracies and outright delusions (among many fails, Google prompted users to run with scissors for cardio and bathe with a toaster to relieve stress). Overviews became the new face for rushed AI adoption and the ways generative technologies are being stretched beyond their limits.

Lurking beneath the viral tweets, though, was a genuine anxiety among journalists and news publishers who worried that Overviews was using (and often misrepresenting) their editorial content to output its answers, all while pushing links to outlets out of view and disincentivizing users to click through to article pages.

Despite an initial statement from Google that it would scale back the feature in late May, three months post-launch it’s clear that AI Overviews are here to stay. On August 15, Overviews was released in six more countries, including the U.K., Japan, India, and Brazil. Searches in Japanese, Hindi and Portuguese have also begun surfacing Overviews.

News publishers around the world are now faced with figuring out how Overviews will impact their organic search traffic and whether or not they should adjust their search strategy accordingly. To begin to answer that question, I asked seven leading audience strategists and SEO experts in the U.S. how they’re thinking about this new phase in search and whether they’ve already made changes to their audience strategy since May.

Their responses have been lightly edited for length and clarity.

Veronica de Souza, director of digital news and audience, New York Public Radio:

I wouldn’t say that the introduction of AI Overviews has meaningfully changed WNYC and Gothamist’s audience strategy, but alongside other Google updates, it has made the implementation of this strategy more urgent. We are a member-funded newsroom. Our strategy is centered on building direct, addressable relationships with our audience, and lessening our reliance on platforms we don’t own. Though our off-platform audiences are still an important part of our strategy, we’ve doubled down on converting people to our O&O (owned-and-operated) platforms like our app and newsletters.

While I believe that Google wants to make its products more helpful to its users, I also think Google wants its users to spend more time with its product and less on other sites. Each new “helpful” feature pushes external links further down the search results page, which is dangerous for publishers who have too many eggs in the SEO basket.

More transparency about which categories of search queries surface AI Overviews would be a good start. I would also like to see Google commit to not using AI Overviews on search results pages for news content and to limit the “stacking” of their other products.

I’ve seen more changes in the Google algorithm in the last year than ever in my career, so it’s hard to attribute traffic changes to one particular product. We, along with many other publishers, have seen a shift in the referral source for their Google Traffic from Search to Discover, which is even more concerning than AI Overviews. While you can’t control how or when you appear in search results, you do have some influence over it. You can be strategic about your content choices, packaging, and the technical build of your site. Google Discover is basically a black box.

Moving forward, we’ll continue to track our Google traffic very closely, looking at share of visits from different Google products and our visibility and click-thru rate for our set of key search terms (like “nyc news”).

Our newsroom is committed to providing New Yorkers with the local news they need to navigate this city and go about their day. Changing the kinds of stories we cover just because of a Google product update would do our audience a disservice. We will, as always, continue experimenting with story packaging, formatting, and page speed, all of which have a positive impact on the user but also help with visibility in search results. We’re also doubling down on sticky, habit-building digital products like our app and newsletters.

Bryan Flaherty, deputy head of audience strategy, The Washington Post:

Existing Google search features [like “knowledge panels” and “people also ask” modules] extract text and link directly back to the source of the information. Users can easily assess the veracity of the information and go to the source page to get more context. But AI Overviews are pulling from multiple sources without that clear citation, and not only is the aggregate answer not always accurate, it is presented in such a way that makes it nearly impossible for users to know if what they are reading is based in fact.

AI Overviews that tell users to eat rocks or use glue on pizza make headlines, but there are others that appear believable — and that is the real danger to the public here.

It’s hard to overstate the problems with the inaccuracies, misinformation, and hallucinations, and what that means for users’ trust in Google and Google as a platform for news publishers to reach audiences. If Google loses users due to the quality issues in its results and AI Overviews, users could continue to turn to non-traditional search platforms that don’t have as direct a tie back to sites, like YouTube and TikTok, which will have an impact on traffic.

There is also no data provided by Google around AI Overviews — what search queries these show up on, what traffic is driven by AI Overview links vs. other features, etc. — which makes it difficult for publishers to assess performance and make strategy decisions.

Whenever a platform, search or otherwise, makes a substantial change it’s right to be worried about the impact. There are still so many unknowns about AI Overviews — including what topics and types of queries it will trigger on moving forward — that we’re watching and evaluating any changes in our site performance. That said, it is our mission to provide our audiences with original, rigorously reported news, analysis, perspectives and service journalism to help them understand the world. It is not often that a zero-click answer will be enough for the audience’s we are trying to engage.

Mike Dougherty, director of digital strategy, Vermont Public:

We haven’t seen a noticeable impact on our search traffic yet. In part that seems to be because as a local news site covering a small state, the specific topics we’re covering day to day don’t seem to rise to the level of triggering AI Overviews. I expect this could change as the tools evolve and as more people start using them.

So far we’ve only identified one example of a story that got pulled into an AI Overview. A search for “are unicorns real?” brings up an AI Overview that draws from an episode of our nationally distributed kids’ podcast, But Why. Search referrals on that story declined slightly after this May. According to Google Search Console, impressions were steady while clicks declined — which seems to indicate that people are reading the overview and not feeling as much need to click through.

This product could so easily put clickable citations into or above the text. It could even write, “According to [publisher],…” the way one news outlet might credit another. We all know generative AI can be instructed to do that. Why isn’t Google doing that here? I’d also like to see Google provide better data, in Search Console or elsewhere, on how publishers’ content is appearing in AI Overviews. We’re doing a lot of guesswork to understand how our content is showing up, if at all, and what impact the overviews are having.

I do think this will cut into our organic search traffic, but it’s too early to tell how much. There are a lot of SEO experts that have more capacity to track these developments than individual publishers do — we just have too many other things on our plates. Some experts are saying that guardrails around hard news topics seem to be working. Other studies show plenty of news-related queries yielding AI results, in some cases without any links or citations. It’s really hard to get an accurate read right now.

Outside of search, we want to use every opportunity to highlight the humans behind our content. It’s going to be more important than ever for audience members to understand that a piece of reporting came from a real local journalist, someone who lives in their community and cares about it.

Scott Brodbeck, founder and CEO, Local News Now (ARLnow, ALXnow, FFX Now):

I’m not too worried about the Google AI summaries at this point, for three reasons. First, featured (non-AI) snippets in the search results already exist and were not the end of the world. Two, the kind of users who would read a two-sentence AI summary of an article and not click through were likely just to read the headline and blurb anyhow. Thirdly, I am confident that our publications add enough value to readers — through engaging writing, additional context, interesting visuals, reader comments, etc. — to incentivize people to click through

If you as a news publisher cannot out-compete a brief AI-written summary, I think you have a big problem that’s not just being caused by Google and AI.

One area that does worry me with AI is the ability of unscrupulous websites to systematically scrape news content from legitimate publishers, publish AI re-writes, and out-rank the legit sites on Google search results and Google News. I’ve seen examples of that happening already. I hope that Google can be proactive about finding and blacklisting such sites going forward.

Marat Gaziev, vice president of audience growth, IGN:

AI Overviews differentiate themselves in one key aspect, they are designed to engage users in a conversational manner. While not fully realized yet, the intent, I believe, is for these overviews to evolve into a chat-like experience, prompting users to ask follow up questions directly within the search results. This marks a shift from the traditional ‘search and click’ behavior to a more interactive ‘search and converse’ model.

I certainly share the industry’s concerns regarding the factual inaccuracies of AI Overviews, particularly for YMYL (your money or your life) queries. These issues underscore critical aspects of Google’s mandate to organize the world’s information and make it universally accessible and useful. The ‘usefulness’ of information is tightly tied to its accuracy, and accuracy cannot be reliably achieved without the RAG (retrieval augmented generation) framework.

RAG requires a deep and symbiotic relationship with content publishers and the media industry to ensure that only credible sources are utilized during retrieval and augmentation. These accuracy problems during the rollout have only reaffirmed my belief that audiences will continue to seek out credible sources of information, and that a symbiotic relationship between Google and content providers must be maintained.

I want to be realistic about the fact that Google, as a business, needs to implement features and products that satisfy their users. I also believe that both Google and the media industry need to stay in lockstep with changing consumer behaviors. Rather than trying to tweak the product artificially, I’d like to see Google continue to evolve with user preferences but do so in a responsible manner. To me, this means fostering an even deeper relationship with reputable sources of information to ensure that consumers aren’t misled by factual inaccuracies.

Ryan Restivo, YESEO app founder, 2022 Reynolds Journalism Institute fellow:

If Google’s stated mission is still to organize the world’s information, then they must know how valuable contributions by journalists are. Stories produced by journalists, whether written or video or otherwise are some of the most highly vetted records of our history that get categorized every day. Anything that changes how that information is interpreted is going to be consequential.

We have started to see more AI Overview answers coming from the top ranked search items, which is a good thing. But the biggest problem, in my opinion, is the competition entering this space. Google has so far committed to not having as many AI Overviews in news searches, but competitors are entering the space, and this market pressure could change that policy. When consumers see overviews, we need to make sure they understand that they may not be seeing the best representation of the information.

I also think the one point that doesn’t get talked about a lot is the environmental impact of these types of rolled out AI Overviews. The amount of compute needed to produce these at scale is hurting our environment, because the amount of power to these highly advanced chips are emitting more carbon than hundreds of households in a year and consuming more clean water. We have to really think about the cost of progress, how it affects all of us as this space evolves.

Seth Liss, director of audience engagement, The Los Angeles Times:

It’s possible that Google might make more money from search results than AI answers and change their strategy. But if they continue going in the current direction, I imagine Google will continue to evolve its search product to produce better answers, rather than better search results.

Search has long been a major way people discover new websites, and a step on their way to becoming devoted readers. If Google decides its best way forward is to keep all of those readers on its own site, there will be a lot of sites that have to figure out other ways to find new audiences.

When it comes to search, our audience strategy has pretty much remained consistent. We continue to watch, track, experiment and try to figure out the best way to bring L.A. Times content to both our existing and potential audience. We are always talking about new ways of delivering content to our readers and how we can best serve them on our own platforms and through social and other sites where people get information, but I can’t say that is in response to AI Overviews.

Andrew Deck is a generative AI staff writer at Nieman Lab. Have tips about how AI is being used in your newsroom? You can reach Andrew via email (andrew_deck@harvard.edu), Twitter (@decka227), or Signal (+1 203-841-6241).
POSTED     Aug. 29, 2024, 1:58 p.m.
SEE MORE ON Aggregation & Discovery
Show tags
 
Join the 60,000 who get the freshest future-of-journalism news in our daily email.
Journalism scholars want to make journalism better. They’re not quite sure how.
These past five years have made it clear just how labor-intensive it can be to bring people from journalism studies and practice together, and how challenging it can be to maintain the discussions between these sides.
Congress fights to keep AM radio in cars
The AM Radio for Every Vehicle Act is being deliberated in both houses of Congress.
Going back to the well: CNN.com, the most popular news site in the U.S., is putting up a paywall
It has a much better chance of success than CNN+ ever did. But it still has to convince people its work is distinctive enough to break out the credit card.