01 The evidence
Readers ranked expert wording first, AI wording last
The way you explain something is the part people trust, and it survives even when your name is not attached.
Three versions of one piece of financial advice went to 285 adults, and only the wording differed.
Three researchers published a study on September 17 that separated two things usually tangled together: who gives advice, and how the advice is written. They recruited 285 American adults and put eight real financial decisions in front of them, covering housing, credit card debt, emergency savings, long-term investing and graduate education.
The underlying recommendation was held identical across every version. Only the style changed: one written the way an AI assistant writes, one the way a human expert writes, one the way an online forum answers.
Expert-style wording was ranked first by 47.4% of readers. Online community wording took 29.1%, and AI-style wording came last at 23.5%.
Then the researchers removed the source labels, so nobody could see which was which. The expert style still came first, at 47.8%, essentially unchanged. People were not responding to a credential they could see, but to how the answer was built.
That distinction matters for anyone who sells expertise. A franchise advisor explaining why one territory is worth more than another, or an accounting firm explaining a filing position, is being judged on the explaining, not only on the license behind it. The study states its own limit plainly: these were written scenarios, not real money moving.
What to do about it
Take the last written explanation you gave a customer and read it beside what an AI assistant produces for the same question. The study says the wording is doing more of the work than your credentials are.
Source
arXiv, Aryan Ramchandra Kapadia, Eshwar Chandrasekharan and Koustuv Saha, “Trustworthy FinAInce: Unpacking How AI-Mediated Financial Advice is Judged,” September 17, 2026 · arxiv.org
02 The evidence
ChatGPT raised coding scores and cut what students remembered
Work produced with AI can look better and leave your team knowing less about how it was done.
Fifty-five students were tested twice, once straight after a programming task and again two days later.
Four researchers ran a controlled experiment with undergraduate computer science students, published September 18. Fifty-five students completed three programming tasks, half with ChatGPT-4.5 and half with ordinary web search and no AI.
The output improved sharply. Students using ChatGPT scored 89% on the coding tasks against 69% for the group without it.
Then the researchers asked them what they had just done. Immediate recall was 41% for the ChatGPT group against 53% for the others, and after 48 hours it was 39% against 52%. The gap opened at the start and did not widen, which means the knowledge was never taken on rather than forgotten later.
The last number is the most important one. Students who used ChatGPT attributed 45% of the submitted code to themselves, against 81% for the comparison group. They could tell the work was not quite theirs.
For a business whose product is knowing something, that is the cost that does not show up in the deliverable. A tax practice or a real estate brokerage can ship better documents this year and have a team that understands them less well next year. The authors flag their own limits: a small group, three short tasks, and physiological measures they say lost too much data to interpret.
What to do about it
Pick one piece of AI-assisted work your team produced this month and ask whoever submitted it to walk you through why it says what it says. If that explanation is thin, the output was the easy half.
Source
arXiv, Christian Bergh, Benjamin Tag, Alexandra Vassar and Jake Renzella, “Your Programming Students’ Cognition with ChatGPT: Higher Performance, Lower Retention, and Reduced Ownership,” September 18, 2026 · arxiv.org
03 The ad change
Google Ads can now send your paid click to Google
You can now pay Google to deliver your customer to a Google page instead of your own website.
A new setting in Performance Max campaigns offers two destinations, and one of them is not your site.
Performance Max is Google’s automated ad product, the one that places a single campaign across Search, Maps, YouTube and the rest without the advertiser choosing where. A new setting appeared in it on September 15 asking advertisers where people should go after clicking.
There are two options, and Google’s own wording is worth reading exactly. The first reads “Your website: Take people to your website. This might be your homepage or a specific page.” The second reads “Google Business Profile (optimized for ads): Take people to a Google-hosted page with built-in conversion measurement capabilities.”
The phrase doing the work there is “Google-hosted page”. Choosing the second option means the click you paid for ends on a page Google owns, not one you control, and the measurement that tells you it worked is Google’s too.
There is a real argument for it. A Google Business Profile loads fast, works on a phone, and shows hours and directions and a call button without the customer hunting. For a medical practice or a luxury home builder, that page may convert better than a slow website.
The trade is that the visit never reaches anything you own, so nothing about that person carries forward. The person who found you becomes a contact Google holds rather than one you do.
What to do about it
If you run Performance Max, open the campaign and check which destination is selected. This is a new setting, so confirm it says what you intended rather than assuming the default is your own site.
Source
Search Engine Roundtable, Barry Schwartz, “Google Ads PMax: Where Should People Go After Clicking Your Ads?” September 15, 2026, reporting a finding by Natasha Kaurra · seroundtable.com
04 The thing to check
Sites that fell on September 4 recovered on September 13
If your traffic dropped in early September and came back, that was Google moving, not your website breaking.
Google changed search rankings around September 4 and reversed the change nine days later, announcing neither.
Search Engine Roundtable reported on September 14 that websites which lost rankings in early September had got them back. Barry Schwartz wrote that “a number of the sites that fell off a cliff around 9/4 completely surged back yesterday. They dropped heavily in Search and some in Discover as well.”
Google did not announce the September 4 change and did not announce the reversal. Neither appears on its published list of confirmed updates.
That is the opposite of what most businesses assume when traffic falls. A drop is normally read as something broken on your end, which is when somebody gets paid to fix a site that was never the problem. Google’s ranking systems move constantly, and most of that movement is never announced.
There is a practical test: check whether the drop and the recovery line up with early September and around the 13th. If they do, the cause was outside your business and nine days of worry was the only damage. If your traffic fell and has not returned, that is a different question and worth a real look.
What to do about it
Open your analytics and look at the first two weeks of September. If the line dips around the 4th and recovers around the 13th, you have your answer and there is nothing to fix.
Source
Search Engine Roundtable, Barry Schwartz, “Google Search Ranking Update Around September 4th That Reverted September 13th,” September 14, 2026 · seroundtable.com
05 The precedent
Google removed free product listings across Europe
A regulator can turn a free placement into a paid one, and the businesses relying on it get no notice.
Unpaid product blocks vanished from Google search results across five European countries in about two days.
Google’s “popular products” carousels, the unpaid product blocks that appear on shopping searches, have almost entirely disappeared across the European Economic Area. Search Engine Roundtable reported it on September 18.
The measurement comes from one analytics firm and is worth reading as such. Hugo Huijer of Productrise reported “~90-100% drops in EEA countries, among which we have sufficient data for Germany, France, Belgium, Sweden and the Netherlands.” Adriaan Dekker posted the same pattern separately.
The cause is not in dispute. Ginny Marvin, Google’s Ads Liaison, confirmed that Google removed the free product listings because of the Digital Markets Act ruling in Europe. The space is going to price comparison services instead.
This is Europe, and a business selling in the United States is not affected today. It is here as a precedent. A free placement that businesses had relied on for years was removed in about two days by regulation, and the businesses relying on it were not consulted and got no warning.
What to do about it
Nothing to do today if you sell in the United States. The point worth carrying is that free placement on somebody else’s platform is a permission rather than an asset, and permissions get withdrawn.
Source
Search Engine Roundtable, Barry Schwartz, “Google Search Drops Free Product Listings From European Economic Area,” September 18, 2026 · seroundtable.com
