01 The demand signal
Your client talked it over with something that agrees
A quarter of American internet users bring personal problems to a chatbot, and many prefer its company to a person’s.
Elon University’s Imagining the Digital Future Center and the Elon University Poll published “The Rise of AI Companions” this month, and the headline number is the plain one. 27% of internet-using US adults now have social, emotional or personal conversations with a large language model such as ChatGPT, Gemini, Claude or Copilot. The researchers then surveyed a representative sample of 1,000 of those users, fielded through YouGov in May 2026, to ask what the conversations are for.
The answers are closer to your practice than they first look. 53% said they turn to AI for advice on how to handle difficult situations with other people. Half use it to talk through personal problems. 39% said they occasionally tell AI things they would not tell another person, and the same share agreed that AI understands them better than most people do.
Then the part that should give any advisor pause. 31% of these users call the chatbot a friend, 36% feel emotionally connected to at least one, and 20% said it is at times more reliable or supportive than their own friends and family. A person who has spent six weeks confiding in an agreeable machine does not arrive at your office neutral.
Two limits. Fieldwork ran in May 2026, so this is the spring rather than this week, and everything here is self-reported by people who already use AI this way. The 27% covers internet users, not every adult.
What to do about it
Assume the first conversation about a matter happened before you were called, and that the other participant had no duty to be right. Ask what the client has already read or been told, early, while it is still cheap to correct.
Say plainly what a chatbot cannot know: the file, the jurisdiction, the other side, the deadline that already passed. That is a more useful answer than telling a client the tool is unreliable, which they will not believe.
Source
Imagining the Digital Future Center and the Elon University Poll, “The Rise of AI Companions,” September 2026 · imaginingthedigitalfuture.org · announcement: Elon University
02 The evidence
Same product, different seller, higher price
Two million listings say a spot on the results page tells you almost nothing about the answer.
Productrise, a company that sells organic product tracking for Google Shopping, ran the same product searches in AI Mode and in ordinary Google results on the same day, in the US and the UK, from August 9 to August 31. It logged more than two million product listings across more than 100,000 results pages and AI Mode answers, matched items by Google’s own product ID, then compared the first offer shown on each side. Search Engine Journal wrote it up on September 3.
The overlap is the finding. Where both surfaces returned products for a query, the carousel showed 27.8 products on average and AI Mode showed 3.9. Averaged across the 23 days, only 1.28% of carousel listings also appeared in AI Mode for the same search that day. Among the products that did appear in both, the seller listed first was a different merchant 49.6% of the time.
Prices moved with the sellers. They differed on 38.1% of matched products, and AI Mode carried the higher one 68.4% of the time. Across every matched product, AI Mode averaged 21.6% higher than the carousel.
Now the caveats, because they matter more than the headline. Google told Futurism it has not verified the accuracy of the report and says both surfaces draw on the same Shopping Graph. Productrise queried products it already monitors for clients rather than a random sample of Google searches, it sells a tool that this finding argues for, and the data records only what appeared on screen. Nothing here measures a click, a purchase or a price actually paid. The founder’s reading, that price matters less to AI Mode, is an interpretation of a pattern, not a measurement.
What survives the caveats is worth carrying anyway. Two surfaces on the same search engine, on the same day, disagreed about who should be listed first in about half of overlapping cases. If that holds for products, ranking well on one surface is weak evidence about the other, which is the whole premise behind most visibility dashboards sold to service firms.
What to do about it
Run your three most valuable queries twice, once in ordinary results and once in AI Mode, and write down who appears first in each. Two minutes gives you your own version of this study, on the searches you actually care about.
When a vendor reports your visibility, ask which surface was measured and on what date. A report that does not say is measuring one of several answers and calling it the answer.
Source
Search Engine Journal, “Google AI Mode Prices Differ From Product Carousel For Same Items,” by Matt G. Southern, September 3, 2026 · searchenginejournal.com · underlying data: Productrise · Google’s response: Futurism
03 The platform change
Google’s newest model is not citing anybody
Gemini 3.8 Flash reached AI Mode on Wednesday, and the early answers arrive with no links in them.
Robby Stein, VP of Product for Google Search, announced on September 2 that Gemini 3.8 Flash “has landed in Search.” It is live in AI Mode for Google AI Pro and Ultra subscribers worldwide, selectable from the model menu. Nick Fox, who runs Search, noted the timing: this arrived less than three weeks after 3.7 Flash.
By Wednesday evening the reports had turned. Gagan Ghotra posted a series of broad, early-stage AI Mode searches run on the new model, and none of the answers he showed carried a link or a citation of any kind. Search Engine Roundtable covered it Thursday morning under a question mark, which is the right punctuation for it.
Treat this as a sighting. These are individual screenshots, not a measured sample, the model is behind a paid subscription, and Google has not said whether the missing citations are intended or a defect. Citation behaviour has changed between Gemini versions before and moved back.
The reason it is worth your attention is the arithmetic underneath. A firm gets nothing from an AI answer except the citation, and if the surface stops linking, the pipe from answer to website narrows to nothing while your visibility report keeps counting impressions. Two model swaps in three weeks also means anything you measured in August was measured against a model that is no longer running.
What to do about it
If you pay for an AI visibility tool, ask which model it queries and how often that changes. A number that moved this week may be describing Google’s release schedule rather than your firm.
Keep the channels that do not depend on a citation working: the phone number on your profile, referral relationships, your own list. Being named in an answer with no link is brand exposure, not a lead source.
Source
Google, “Gemini 3.8 Flash,” September 2, 2026 · blog.google · rollout: Search Engine Roundtable · citation reports: Search Engine Roundtable, September 3
04 The one with a clock on it
Check whether your Local Services Ads came unlinked
Google has confirmed a bug that detached LSA accounts from their Business Profiles, in the middle of a migration.
Ginny Marvin, Google’s Ads Liaison, wrote on LinkedIn on September 1: “We identified a bug and are actively working to resolve it.” She added that it is unrelated to the migration described below. The complaint she was answering came from an advertiser, Jack Hepp, who reported that his longest-running Local Services Ads account was unlinked from its Google Business Profile over a weekend and that he needed support to relink it. He also could not see the LSAs in any associated Google Ads account, and hit repeated access errors.
The context is a move that has been scheduled all year. Local Services Ads are becoming a Performance Max campaign type inside Google Ads, built for pay-per-lead goals. Google’s help documentation puts the first phase in August 2026 for selected US home and storefront service categories, broader groups including service-area businesses later in 2026, and non-US accounts plus remaining categories in 2027. Budgets, settings and creative assets carry over automatically.
Two details in that documentation are the ones to act on. Account administrators get a notice 14 days before their migration and a reminder seven days later, so the window to prepare is short and it starts with an email somebody has to read. And historical Local Services Ads performance reports do not move to Google Ads.
For a firm running LSAs, an unlinked profile is not a reporting nuisance. The link is what ties the ad to the verified business, and this is the channel that still carries a call button in searches where organic listings have lost theirs.
Do this today
Open each Local Services Ads account and confirm the Google Business Profile link is intact. If it is not, contact support to relink rather than waiting for the migration to create a link for you.
Export your Local Services Ads performance history now, before your migration notice arrives. Google says the historical reports do not transfer, and a lead-cost baseline you cannot produce is a baseline you cannot defend later.
Source
Search Engine Roundtable, “Google Working On Fixing Local Service Ads Bug,” by Barry Schwartz, September 1, 2026 · seroundtable.com · migration documentation: Google Ads Help
05 The money
The judge changed Google’s conduct, not its shape
A second breakup attempt fails, and the remedy is disclosure rather than divestiture.
On September 2, US District Judge Leonie Brinkema declined the Justice Department’s request to force Google to sell AdX, its exchange for publisher ad inventory, a year after she found that Google had illegally monopolised that technology. She instead granted most of the behavioural remedies both sides had proposed, including sharing more information with publishers, and ordered Google to stop practices that depress the rates publishers earn.
Reuters put the stake in plain terms: publishers pay Google roughly a 20% fee to sell ads through auctions that run as a page loads. The New York Times reported that Brinkema has not released the full opinion and previewed it in a short docket filing. Google’s regulatory affairs VP, Lee-Anne Mulholland, said the company was “very pleased the Court rejected the DOJ’s proposal.”
This is the second time Google has avoided a divestiture under a monopoly finding, after the separate search case remedy last year. Nothing here touches search results, AI Overviews or the ad auction your firm buys clicks in.
What it does settle is the direction of the next few years. The remedy for a market a court has already called monopolised is more reporting to the people selling inventory, not a different owner. Expect better data about where display money goes and no particular reason to expect it to get cheaper.
What to do about it
Nothing changes in your account this week. If you buy display or run ads on legal publications, watch for new publisher-facing disclosures over the coming months and ask your media buyer what the reported take rate is.
Keep the state bar rules attached to whatever you buy. Advertising rules apply to a machine-assembled ad exactly as they apply to one you wrote, including claims about results and anything that reads as a testimonial.
Source
Search Engine Roundtable, “Judge Rules Google Ad Tech Business Won’t Be Broken Up,” by Barry Schwartz, September 2, 2026 · seroundtable.com · reporting: Search Engine Land
