You saw a number on a screen, trusted it, and acted on it. Later you found out the screen was wrong, and the money was already gone. Amanda Herrfort owns Anna Mae’s Bakery in Millbank, Ontario. On a Saturday in July 2026, she was on her way to a wedding when somebody sent her a screenshot. Someone had Googled her bakery, and the AI summary at the top of the page said it was closing for good on July 31. “That’s when I panicked,” she told CBC News a few days later.
The bakery was open, and she had no plans to close it. It’s a popular spot with tourists and locals, CBC says. She told CTV News that if you looked up the bakery’s name and website, everything came up normal, with nothing about closing. Herrfort put the word out online. People kept calling and emailing anyway, and they kept walking in. They’d tell her employees that Google said they were closing, and the employees would tell them they weren’t.
So Google had an open bakery listed as closing. My list of thirty thousand companies had the opposite problem. Companies that had stopped trading still had a price, so they still looked open.
Why a dead company still has a price
When a company stops trading, it doesn’t disappear from the data feed. Its last traded price stays in the record, and depending on who sends you the data, it keeps showing up as if it were today’s price. The accounts from the years before it left are still there too. They’re complete, and they look like perfectly good numbers to work with. So the math runs. The cash, the number of shares and the price were all real on the day the company left the exchange. And the result says the business is selling for six cents on the dollar.
I sorted the list by how far the price was below what the math said each company was worth. The ones at the top were 94 percent below. That’s the six cents. And every one of them had stopped trading. When I went looking for the rest, I found 4,041 of them on the list. Every input was real, and the math still broke. A price on a screen is supposed to be a price you can get, and no feed tells you whether you can. If you sort a list by how good things look, the broken rows pile up at the top. So when I test a screen, I read the top first.
Herrfort ran into the same thing from the other side. The date at the top of the Google page looked like a fact, the same way a price on a screen looks like a price you can get. People who saw July 31 came in and asked about it. She’d panicked too when she first saw it.
Where the errors come from
A Google spokesperson told CBC that the vast majority of its AI summaries are high quality. The spokesperson also said that mistakes can happen with billions of searches every day, and that problems come up when there isn’t much good information about a topic on the web. At the bottom of every summary, in small print, Google says its AI answers can include mistakes. Google’s mistakes show up where the web is thin. Mine pile up at the top of my list, because that’s where sorting puts them.
In the end, Herrfort fixed it by typing into Google’s AI search bar and asking for a correction. She told CTV she wrote that they were the owners and they weren’t closing. She told CBC that Google’s AI wrote back that it understood the harm and would get it fixed as soon as it could. Now she tells people to contact the business directly, or check its website or Facebook page, before they take the AI’s word for it.
Where this breaks
A rule keeps the dead companies off my list now, and it’s a pretty blunt rule. It looks at one thing: whether the price has moved. Nothing in the feed says a company left the exchange, so that’s the only sign I get. The cost is the live company that hardly ever trades. To the rule, it looks the same as one that stopped trading, and some real businesses got caught. When one of those starts trading again, it comes back on its own, because the price keeps getting checked. I can’t go through the whole list by hand and decide which prices are current. So a rule that’s a bit too wide beats deciding by eye, even if it catches some companies it shouldn’t.
You can still find a dead company if you search for it by name. In the app, its value stays a gray square, with a red note next to it that shows when its price stopped. It’s the same gray a company gets when any other number is missing, like a currency with no government rate I can use (I wrote about that one in the Toyota piece). If you look up a company that stopped trading, the app should tell you it stopped trading, and that’s all the square should say. On September 23 the opposite case turned up. A company in Tokyo that had only just joined my list looked dead on my screen, with a flat line and no description. It had come in after my monthly data load had already run, and I fixed the load that day.
The other Anna Mae’s
Google didn’t make up the closing. CBC reported that the search engine had mixed up her Anna Mae’s with a bakery of the same name in Illinois, and that one is closing.
My list had a mix-up like that too, and I found it a day before the dead companies. Sorting had put 505 other things at the very top, around 97 percent below value, and none of them had stopped trading. They were preferred shares and listed debt. A preferred share is a different piece of paper from the company’s ordinary shares, and in a bulk feed it often comes in carrying the parent company’s name and accounts. So the model worked out what the whole company was worth per share. The parent was a real company with real accounts. The price was for a different piece of paper.
Key factsThe numbers and where they come from · 7 September 2026
- In July 2026, a Google AI summary said Anna Mae’s Bakery in Millbank, Ontario, would close for good on July 31. The bakery was open, its owner, Amanda Herrfort, told CBC News.
- A Google spokesperson told CBC News that problems with its AI summaries come up when there isn’t much good information about a topic on the web.
- When a company stops trading, its last traded price stays in the data feed. Bufetico found 4,041 such companies on its list of thirty thousand.
- A preferred share is a different security from a company’s ordinary shares. In a bulk data feed, it often arrives carrying the parent company’s name and accounts.
