AI answers are ripping traffic
AI “fetcher” bots are pulling huge amounts of publisher content away from original sites, and that extraction is already measurable: bot traffic to publishers has surged about 300%, creating a big revenue and attribution problem for news organisations. (searchengineland.com) At the same time, Google’s AI Overviews are producing large volumes of incorrect answers — analyses say that, at Google’s scale, those errors add up to millions of wrong responses every hour — which heightens the trust-versus-monetisation trade-off for any product that intermediates news. (tech.yahoo.com)
A news site can spend money sending a reporter to city hall at 9 a.m., and by 9:01 a.m. an artificial intelligence bot can copy the facts, hand them to a chatbot, and keep the reader from ever visiting the site that paid for the reporting. Akamai says artificial intelligence bot activity rose about 300% in 2025, with media and publishing among the hardest-hit targets. (searchengineland.com) (prnewswire.com) Akamai splits these bots into two groups. Training crawlers vacuum up old archives to teach models, while fetcher bots grab fresh articles in real time so an artificial intelligence product can answer the user directly. (markets.businessinsider.com) (searchengineland.com) The second group is the one publishers fear most, because it shows up at the exact moment a story becomes valuable. A sports score, court ruling, earnings report, or weather alert is worth the most in the first few minutes, and that is when fetcher bots can strip out the answer and send the audience somewhere else. (searchengineland.com) (markets.businessinsider.com) That breaks the old search bargain. For two decades, publishers let search engines index pages because links sent people back, and those visits turned into ad impressions, subscriptions, and brand loyalty; now the machine can take the useful sentence and leave the publisher with the server bill. (blog.google) (searchengineland.com) Google pushed this model into the center of the web when it rolled out AI Overviews to everyone in the United States in May 2024. Google said people had already used the feature billions of times in Search Labs before the wider launch. (blog.google) Google also said in March 2025 that it now sees more than 5 trillion searches a year. At that scale, even a system that is right most of the time can be wrong an astonishing number of times. (searchengineland.com) A New York Times analysis with the startup Oumi tested 4,326 searches on a standard factual benchmark called SimpleQA and found Google’s AI Overviews were correct about 91% of the time in February, up from 85% in October. That still leaves roughly 1 in 10 answers wrong. (searchengineland.com) (tech.yahoo.com) If you apply that error rate to more than 5 trillion searches a year, the arithmetic gets ugly fast. Multiple reports put the implied total at more than 57 million incorrect answers an hour, or nearly 1 million a minute. (tech.yahoo.com) (popsci.com) So publishers are getting hit from both sides at once. The bot takes the audience before the click, and the answer it gives can still be wrong, which means the platform captures the attention while the original source loses both the traffic and the credit. (searchengineland.com 1) (searchengineland.com 2) Akamai says publishing accounts for 63% of artificial intelligence bot triggers inside the broader digital media sector. The company also says fetcher bots make up 24% of all artificial intelligence bot activity hitting media, and publishing accounts for 43% of that fetcher segment. (prnewswire.com) (markets.businessinsider.com) That is why publishers are moving from quiet complaints to meter-and-toll tactics. Akamai says companies are tightening bot controls, while newer services such as TollBit are trying to turn scraping into a paid transaction instead of a free extraction. (searchengineland.com 1) (searchengineland.com 2) The fight is no longer just about copyright in a courtroom. It is about whether the web keeps rewarding the people who gather facts first, or whether the reward shifts to whichever machine can repackage those facts fastest. (searchengineland.com) (tech.yahoo.com)
