Tuesday, 19 December 2023

Inside the first 'SEO heist' of the AI era

A robot hand grabbing a webpage
  • Over the last year, website Exceljet was the victim of an AI-powered "SEO heist."
  • A provocative online marketer used AI to clone thousands of Exceljet's articles for a competitor.
  • The plot shows how generative AI is transforming the web, and posing thorny questions for Google.

Traffic was crashing, and David Bruns couldn't figure out why.

Exceljet, his website for navigating Microsoft Excel, had been a hit since 2012, quickly becoming a lucrative full-time job for the Utah resident.

In late 2022, though, visits to Exceljet started slipping — and then kept going down. The site relies heavily on Google search traffic, and with the tech giant periodically updating how it ranks websites, Bruns wondered whether he'd fallen out of the company's favor. He'd also embarked on a technical rebuild; perhaps that had something to do with it?

Then in the fall of 2023, a friend flagged a social media post to him, and it all made sense. Bruns been the victim of an "SEO heist," one of the first of the new generative AI era. It was an audacious plot to copy his website and siphon away hard-earned Google referral traffic using powerful new AI tools.

The heist was part of a wildly successful campaign that sent traffic to the perpetrator's client up an estimated 60-fold, even as Exceljet's online visits halved.

"It's one thing to get outranked by an article that is arguably better than the article that you wrote, but it's something else to get outranked by an article that was written by a machine that no human ever reviewed," Bruns told Business Insider. "It's just wrong to make money on that."

Exceljet's struggles raise big questions about the future of the internet. Content on any topic can now be generated with the press of a button, and it's all but impossible to spot the difference between this synthetic stuff and what a human might produce on the same subject. That's quickly upending the existing rules of the web — and it's particularly thorny challenge for Google.

"The usual signal methods of what's true are no longer clear, and trusted sources like Google searches don't turn up the right things anymore," said Ethan Mollick, an associate professor at Wharton who studies the impact of AI. "We're going to have to fundamentally change how we relate to information. I just don't know whether we're going to do that as fast as we need to."

SEO on steroids

The central battleground where this is playing out is "Search Engine Optimization." SEO professionals help websites improve their ranking in search results, using tactics ranging from official Google trend analysis to more dubious trucks to game the system. The basic idea is to find out what people are searching for, and deliver relevant content for those queries.

SEOs were quick to see the commercial potential of generative AI when ChatGPT burst onto the scene over a year ago. Once you know which Google search terms (aka keywords) to target, there's no longer any need to ask humans to laboriously type up a relevant article. Now, AI models can knock that out in seconds, thanks to new startups such as Jasper and SEO.AI.

It's SEO on steroids, and Jake Ward is has been more audacious than most industry players in taking advantage of this. The Exceljet heist is his AI-powered handiwork. And he's unapologetic.

Ward argues that what he's done is no different than any other SEO content strategist. Generative AI has just made it way faster. "It's unusual that something acceptable, if done once, suddenly becomes unethical when reproduced at scale," he wrote on X earlier this month.

"Most websites, especially large-scale projects, will hire cheap writers who know nothing about the topic they're writing about," he added. "They copy and regurgitate content from the top few Google results. AI is smarter than 99% of the people writing those articles."

The Exceljet heist, explained

An expat from the north of England now living in Dubai, Ward and his business partner worked to boost business-planning software startup's Causal's presence on Google — in part by using generative AI to feed off Exceljet's existing content.

First, they downloaded Exceljet's "sitemap" — essentially, a directory of the roughly 1,800 pages on Exceljet's website. They took the URLs for each page and fed them all into AI content generation tools. Using those URLs as guidance, the tools promptly spat out a stream of articles on the same subjects, no human author required.

It's not clear exactly when Causal began posting the Exceljet-inspired articles, but its traffic has been climbing steadily for more than a year. It's also not Causal's only attempt to use AI to boost its ranking: Ward's business partner Mack Grenfell previously wrote about using generative AI to work with Causal as early as October 2021.

At the start of May 2022, Causal was averaging less than 10,000 organic traffic visits a week, according to estimates by web analytics tool Ahrefs. Around the start of 2023 it was reaching 200,000, and it peaked in October at more than 600,000 per week.

Over the same period, Exceljet's traffic plunged. It averaged around 2 million users a month in 2022, Bruns said. By 2023, that was down to just over 1 million. (Changes to Google's algorithm, tweaks to Exceljet's website, and other non-AI-augmented competition may have also played a part in that decline.)

The backlash

On November 24, Ward posted on X about his targeting of Exceljet, saying he "stole" traffic from a competitor.

The post went viral, and the backlash was swift. George Orosz, a high-profile software engineer and newsletter writer, argued that it posed a significant threat to Google.

"Generating absolute junk with AI just to make some $$ is happening at alarmingly fast pace. First sites/articles, then images and soon videos," Orosz wrote. "Internet search like Google is becoming much more useless, even as they try and battle all this. Trusted sources become more valuable."

In a conversation with BI, Bruns echoed Orosz's criticism. "This is a big problem for Google," he said. "If people keep finding crappy articles at the top of the search results, they're going to end up questioning whether Google's doing a good job, but the fact that the articles are near or they can be at the top of the result makes it seem like they're legitimate."

Even more galling for Bruns: When he reviewed the AI-generated Causal posts, he found factual errors in some articles, with one describing a purported feature that doesn't exist in Excel. Google even selected this AI howler as a "featured snippet" and placed it prominently at the top of relevant search results, according to Bruns.

"I started to look at the quality of the articles, and realized that we're losing traffic to stuff that doesn't even make sense," he said.

When reached by BI, Ward initially said he was open to an interview, but subsequently declined to talk on the record. "The client in question doesn't want any more exposure with what we did," he said in a voice message.

Causal did not respond to multiple requests for comment.

Google responds

Google isn't fussed about AI-generated content. The $1.67 trillion search giant won't penalize synthetic articles in its search results — so long as it deems the content useful to users.

"AI can be a useful tool to help people with creative endeavors, including the development of great content," Google spokesperson Jennifer Kutz said in a statement. "But like any tool, it has the potential to be misused, so we have a number of efforts to protect the integrity of Search and ensure we're taking a responsible approach."

If Google determines AI is being used primarily to manipulate Search rankings, though, that's not allowed. It's a nebulous judgement that may prove difficult even for a company with the AI prowess of Google.

Shortly after Ward posted his thread, Causal's organic web traffic dropped off a cliff. In early October, it was accruing more than 610,000 organic traffic visits per week, according to Ahrefs' data; by mid-December, that had plummeted to around 190,000. It looked, said some experts, suspiciously like Google had taken a "manual action" — deliberately penalizing the website in its search rankings. (Google declined to comment on specific websites.)

"I can't confirm there's a penalty," said Patrick Stox, an SEO expert at Ahrefs. "But there's basically no other thing besides deleting a bunch of pages that would cause a drop like that."

Google conflicted

While Google passes judgement on other websites' synthetic content, it is hard at work building its own AI models to churn out automated text and other information.

It's trying to sell some of these generative AI tools to major news publishers, according to The New York Times. And there are bigger changes coming. Google is building out SGE (Search Generative Experience), an experimental AI question-answering tool baked directly into search. It could be the biggest shake-up in Google's 25-year history, delivering customized answers directly to users.

AI-powered search experiences like this may lead to traffic declines of as much as 25% for many websites, according venture capital firm Insight Partners.

"As with every great disruption, you either adapt to that disruption or you ultimately perish," said Gary Survis, an operating partner at Insight.

It's a one-two punch for publishers and websites dependent on Google for traffic: Their links may be demoted below Google's AI-generated content, and they'll be competing in any remaining tradition search results with other websites that use AI to write content automatically. 

Dipping their toes

Media companies are already using dipping their toes in the AI water, with sometimes calamitous results. Gizmodo publisher G/O Media was slammed for publishing AI-generated stories containing numerous factual errors earlier this year. Sports Illustrated's publisher recently fired the magazine's CEO following a brouhaha over its publication of articles that featured fake writers and AI-generated profile pictures.

Still, other publishers aren't dissuaded. Michael Nunez, editorial director at tech news site VentureBeat, encourages his writers to experiment with AI tools. Nich Carlson, global editor-in-chief of BI, said in April that the newsroom would carefully learn how to use this new technology.

"The most important piece of that for us is that we never rely on AI blindly or uncritically, and we use it as a tool to augment human skill or judgment rather than to replace it," VentureBeat's Nunez told BI. (Disclosure: Nunez and the author of this article both sit on the board of the San Francisco Press Club.)

Even Exceljet's Bruns is starting to use AI in his workflow, despite Ward's SEO heist.

"A lot of times when I finish an article, I will feed it back into ChatGPT and ask it to review the article. I'm mostly interested in typos, or logical errors," Bruns said. "I sometimes use it to generate ideas," he added.

"Like occasionally, I'll ask it: Can you tell me how you would solve this problem in Excel?"


Got a story? Contact reporter Rob Price via encrypted messaging app Signal (+1 650-636-6268), email (rprice@insider.com), Telegram / Wickr / WeChat (robaeprice), Twitter/X DM (@robaeprice), or mail: Rob Price, Business Insider, 44 Montgomery St, 3rd Floor, San Francisco, CA 94104, USA

Read the original article on Business Insider


from Business Insider https://ift.tt/DQwInOF

No comments:

Post a Comment

From hours to seconds: A short demo on parallel computing shows why Nvidia is the world's most valuable company

Nvidia CEO Jensen Huang Chip Somodevilla/Getty Images Parallel computing speeds up data crunching. Projects that took hours now take se...