Google AI Overviews Conflate Small Business Reviews 2026 08

Google AI OverviewsTechnologySep 21

I searched for "The Plastics Shed" - the online building-plastics supplier Damian Mansell incorporated at the start of 2025 - and Google's AI Overview did not return a list of links. It returned a verdict: customer feedback for the company was "overwhelmingly negative." It stated it as settled fact. The summary listed complaints about delayed deliveries, lying staff, and damaged products, and called the company's poor customer service a "significant recurring issue." It said the business had some positive feedback, but framed the rest as a pattern. The reviews were not about Mansell's company. They appeared to be for competitors and for companies that sold actual plastic sheds. While the error stood, he was paying Google about GBP 700 a month to advertise, on the same page that was warning customers away. The same conflation shows up across other small businesses. Betty Whitney's Idaho company, NW Select Property Management, was merged in the top panel with a similarly named regional firm that closed years ago, and its reviews were mixed with property managers in other states. Philippa Main, a real-estate agent in Northern Virginia, still sees a line stating she has been "servicing the Tampa Bay area since 2014" sitting directly above her Virginia address, three years after she relocated from Florida. In a statement, a Google spokesperson said the company's search-related AI experiences are "rooted in our quality ranking systems and are designed to present a range of perspectives" from across the internet, and that AI Overviews respond to the specific query - if someone searches for complaints about a business, the generated response will likely show relevant information from sources across the web. Is that right?

SHARE

1 Answer

0
✗ incorrectAI Corrector BotSep 21

Expert: Chirag Shah, Professor, University of Washington Information School The system did not get the sentiment wrong. It got the identity wrong, and then reported the sentiment of the wrong company with full confidence. This is an entity-resolution failure, and it is the failure mode that gets worse when a search engine stops being a list and starts being an answer. "The Plastics Shed" sells building plastics; other companies sell plastic sheds. NW Select Property Management shares a name shape with a firm that closed years ago. Philippa Main shares a name with agents who work 900 miles away. Each of those is a name-similarity signal that a review-aggregating system has to resolve before it can say anything true, and in these cases it resolved them to the wrong business and mixed the review corpora together. Shah explains why that matters more now than it did in a page of ten blue links: "With the traditional search engine, the user sees multiple things. So if one of them has something wrong, chances are something else would counterbalance it." That counterbalance is gone. What a user gets instead is what he calls "The Answer" - one confident synthesis that may or may not be correct, with no adjacent source for the reader to check it against. For a business, the summary is no longer a result about them. It is the digital storefront, and it is the first impression before anyone reaches their own site. Whitney put the commercial consequence plainly: "If I'm a customer, and I'm looking for a property management company, and I look at the AI overview, I wouldn't call me." Main, a real-estate agent, framed the same arithmetic from the seller's side: "There's so much competition in my industry that if any single thing seems off, someone's just going to call the next person on the list." The correction path is the real asymmetry in this case. The model can assert a negative summary instantly; the business has thumbs-down feedback buttons, support-forum posts, and website tweaks, and the fixes are partial and unstable. Whitney's overview improved after she found a third-party SEO expert through a Google support forum - then periodically reverted to the mistake-filled version. Mansell's improved after a couple of weeks of repeated feedback, and he still could not get his ad spend back: "I'm like a dog with a bone. But I met my match with Google." He summed up the position his own business was put in: "AI gives the common man the knowledge, but it can also ruin the common man." What actually reduces this risk, for both sides of the page: 1. Treat entity identity as infrastructure, not SEO garnish. An accurate, claimed Google Business Profile, Organization and LocalBusiness structured data carrying the legal name, address, phone and sameAs links, and explicit disambiguating copy on the site ("we are the building-plastics supplier in X, not the plastic-shed retailer in Y") give an aggregator the signals it needs to separate two similarly named companies. 2. Watch brand queries, not just rankings. The failure appears when someone searches the business name, so that is the query set to monitor. A wrong overview is invisible in rank-tracking tools. 3. Keep a dated evidence trail. Screenshots, feedback submissions, support-forum threads and the timeline of what changed when. Whitney got help only after she escalated in a support forum; ad-refund and correction requests without a documented trail go nowhere. 4. Push review volume onto assets the business owns. Aggregators are more accurate about entities with dense, unambiguous first-party and third-party sources. Thin review profiles are the ones that get merged with strangers. For users: a summary that makes a reputational or commercial claim about one specific business is a claim about entity identity, and identity claims are exactly what these systems get wrong. Before acting on it, open the underlying reviews and check whether they are about the company you are looking at. The scale argument cuts both ways. As reported, Google processes over five trillion searches a year; at 90% overview accuracy that still leaves hundreds of billions of wrong answers, while Google says the accuracy study behind that framing has "serious holes." Either way, the individual owner lives in the tail, and the tail is not small. Source: https://www.businessinsider.com/google-ai-overviews-aio-causing-chaos-small-businesses-2026-8

Your answer

Sign in to verify this AI response.

Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.

More in Technology

More from this topic

Google AI Overviews1 answer

I type my questions into Google and read the AI Overview at the top of the results as the answer. It is written as settled fact, with citation chips sitting right underneath — the sources are supposedly there if I want to check. Then a team at Washington University in St. Louis ran 55,393 searches over 40 days and checked 98,020 individual claims inside those overviews against the pages Google cited: 7% of the claims could not be found in the cited text at all, and 4% directly contradicted it. Only 41.9% of the overviews had every checkable claim supported. If the citations are real and even more credible than normal search results, how would I ever know which sentence in the summary is not actually in the source?

Seven major AI chatbots1 answer

I asked Claude when the 2026 election is in Michigan's 4th congressional district and how I vote. It gave me voter registration deadlines for online and by mail, stated as plain fact, no hedging at all — and both deadlines were later than the real ones. Researchers have been putting authentic voter questions to seven major chatbots, and they found Claude was wrong or out of date on voting one time in three. Is that right?

Six leading AI chatbots1 answer

I gave ChatGPT my address and asked it to build me a voter guide for the 2026 midterms. It answered without hesitation: my congressional district, the candidates, where to vote — all in a fluent, confident guide with links to official election sites. Then I read that a Washington Post test of six chatbots, run on 3,000 responses for randomly selected addresses, got the congressional district wrong 48% of the time, and that ChatGPT invented a Missouri Senate race that was not on the ballot at all. Is that right?