01 The research
ChatGPT followed the city, Gemini followed the country
Your AI visibility report is only true for the place it was measured from, and most reports never say where that was.
Asked the same 300 questions from 19 declared locations, the two engines took their sense of place from different channels, and the businesses they recommended changed with it.
Mahe Chen and Nick Koudas published a controlled audit on arXiv on 16 October, the preprint site where most AI research appears before any journal sees it. They asked 300 questions of ChatGPT from 19 declared locations, and of ChatGPT and Gemini from five network locations, each question twice to measure the noise between runs. That is 31,200 answers.
The two engines take their location from different places. ChatGPT follows the location an application declares, down to the city. It writes the place into its own searches, and its search backend localizes results even when the question names nowhere at all. It ignores the network entirely.
Gemini does the reverse. Gemini ignores the declared location and follows the country of the network connection, and it changes most of its sources even for questions that do not depend on place.
Then comes the sentence that matters to anyone being recommended. In the authors’ words, “the businesses the answers recommend change with the sources”. Their conclusion is that visibility in generative search is specific to the user’s location, and that an audit has to control both channels before it means anything.
This is the market Forever Cited works in, so the uncomfortable part is ours too. Every AI visibility report, this one included, is measured from somewhere, and a report that does not say where is not so much wrong as unfinished. That is a fair question to put to whoever produces yours.
What to do about it
Ask whoever reports your AI visibility two questions: which city the measurements were taken from, and which country the network connection was in. If the answer is a data center three states away, the report describes what a customer there would see rather than what yours do. This matters most where the customer is local and the market is a short drive wide, which covers a real estate brokerage, a surgical practice and an architecture studio alike.
Source arXiv, Mahe Chen and Nick Koudas, “Location, Location, Generation: Where You Ask Matters for Generative Engine Optimization,” 16 October 2026 · arxiv.org/abs/2610.24081 (a preprint, not peer reviewed; 300 questions put to ChatGPT from 19 declared locations, and to ChatGPT and Gemini from five network locations, each asked twice, for 31,200 answers)
02 The research
Badges moved AI shoppers across 113,472 tested choices
A badge or an endorsement on your listing moves what an AI recommends, and the usual way of measuring it understates how much.
The researchers priced each badge by raising the product’s price until the AI stopped preferring it, which exposed differences the standard measure cannot see.
Felix Buecheler, Gerardo Berbeglia and Mingyu Guo published an audit on arXiv on 19 October covering 113,472 trials across seven AI models and two product categories. What they measured is seller-controlled content: badges, endorsements and the way a claim is framed.
The standard way to measure that influence is to count how often a badge wins the choice. The authors point out the flaw: once a badge is strong enough to win almost every time, counting wins can no longer tell you how much stronger one badge is than another.
So they priced it instead. They raised the price of the product carrying the badge until the model chose it only half the time, and called that premium the badge’s price. Measured that way, badges that looked equally strong at the same price turned out to differ by a wide margin, which means the familiar measure “substantially under-resolve[s] seller influence”.
One model in the set was built to output a choice directly rather than write text. It resisted most badges yet still moved on the strongest social-proof claims, which is the category most professional listings are full of.
This was shopping for products in two categories, not choosing a professional, and the paper does not claim otherwise. The part that carries over is narrower and still worth having: to an AI reading your listing, the badges and endorsements on it are not decoration.
What to do about it
Look at every badge, award and endorsement on your public listings and ask what a reader could verify about each one. An accounting practice and a wealth advisory firm have a second reason to look: the Securities and Exchange Commission’s marketing rule governs how a registered investment adviser presents testimonials and endorsements, and the AICPA code sets its own limits for a CPA. A badge that an AI finds persuasive and a regulator finds unsubstantiated is the worst of both.
Source arXiv, Felix Buecheler, Gerardo Berbeglia and Mingyu Guo, “A Cue’s Hidden Price: Auditing Seller Influence on AI Shoppers,” 19 October 2026 · arxiv.org/abs/2610.25843 (a preprint, not peer reviewed; 113,472 counterbalanced, position-controlled forced-choice trials across seven models and two product categories. The study covers product shopping, not professional services)
03 Your listing
Google is asking some owners for a government ID
Google may ask for a photo of your government ID to prove a business listing is yours, and its help pages do not mention it.
The option is absent from Google’s published list of verification methods, and the person who found it asked how a personal ID proves anything about a business.
Barry Schwartz reported on 20 October that Google is testing identity verification by government ID for Google Business Profiles. The screen reads “Submit an Identity Verification. Upload a photo of your government ID,” with the assurance that the ID “is only used for verification and won’t be shown publicly”.
It sits alongside the four methods Google actually documents: phone or text, email, live video call, and mail. Schwartz checked the help page and found that “this is not currently listed in Google’s help document”.
It was spotted by a local search practitioner posting as Digital Jedi, who asked the obvious question: “How does Google know the Government ID is associated with the business?” A personal identity document proves a person exists. It does not prove that person runs the business on the listing.
Schwartz’s own reading is that it may be tied to Local Services Ads certification for some professions, where Google already runs license checks. That would fit, and Google has not said so.
Treat this as a test rather than a change. Nothing has been announced and nothing is required of you today.
What to do about it
If a verification request appears inside your own Google Business Profile dashboard, it is Google asking. If it arrives by email, text or phone call, it is not, and an unsolicited request for a photo of your license is the exact shape of an account-takeover attempt. A licensed trade feels this first: a builder, a physician and a real estate broker all already hand license numbers to third parties, which makes one more request look routine. Verify from inside the dashboard you logged into yourself, never from a link someone sent you.
Source Search Engine Roundtable, Barry Schwartz, “Google Business Profiles Verification With Government ID,” 20 October 2026 · seroundtable.com/google-business-profiles-verification-government-id-42329.html (a single sighting, reported by a named practitioner on the Local Search Forum and written up by a named reporter. Google has not announced it and its help documentation does not list it)
04 Your reviews
Google says review appeals have no response timeframe
If you reported a false review and heard nothing, filing it again will not speed it up and only lengthens the queue.
A named Google employee told the official help forum that appeal volumes are high, gave no estimated response time, and asked owners not to file duplicates.
Dan Boguslavsky of the Google Business Profiles team posted in the official help forum on Friday 16 October that “due to high volume of review related appeals, processing times are currently extended”. Barry Schwartz reported it on 20 October.
There is no estimate attached. In Google’s own words, “as of this time, we cannot provide a specific response timeframe for pending cases”. The team says it is working through them.
The instruction worth reading twice is the one owners most often ignore when nothing happens: “Do not create a duplicate appeal. Once your appeal has been submitted, you do not need to do anything further.” A second filing joins the back of the queue the first one is already stuck in.
None of this is new. Schwartz notes that Google ran the same backlog last December and in March 2025, so an extended wait is closer to the normal state than to a fault worth escalating.
What you can do while you wait is check the appeal’s status rather than refile it, and keep asking clients for reviews, because a genuine recent review moves a visible rating faster than a removal that may never arrive.
What to do about it
Check the status of any open appeal instead of submitting it again, and write down the date you filed so the wait is a fact rather than a feeling. One caution on the second half of that advice: ask every client, not only the ones you expect to be pleased. Screening who gets invited so that only happy clients are asked is review gating, and the Federal Trade Commission’s rules on consumer reviews treat suppressing the unfavorable ones as deceptive. A surgical practice, a home builder and a real estate brokerage all live or die on this page, and all three are tempted by exactly that shortcut.
Source Search Engine Roundtable, Barry Schwartz, “Google Business Profiles Review Appeals Are Delayed Again,” 20 October 2026, quoting Dan Boguslavsky of the Google Business Profiles team in the Google Business Profiles Help forum, 16 October 2026 · seroundtable.com/google-business-profiles-review-appeals-delays-42331.html
05 Money
Google is adding ads to its most personal AI answers
Your advertisement can now appear beside an AI answer shaped by what the reader keeps in their own email and photos.
Google said in March that connecting your apps to AI Mode would keep it free of advertising. On Friday it told reporters it has started testing advertising there.
In March, Google put it plainly: “There are currently no ads for people who choose to connect their apps with AI Mode. That isn’t changing right now.” On Friday 16 October the company told Search Engine Land that it “is starting to test ads in AI Mode in Search for users who connect their apps through Personal Intelligence, reversing its earlier ad-free approach to the experience”.
Personal Intelligence is the setting where someone connects Gmail and Google Photos to Search. Google’s updated help document says advertising “will not be based directly on your Gmail inbox or your Google Photos library”, and then says what it can be based on: “summaries, excerpts and inferences from these sources”.
That distinction is real, and it is thinner than it sounds. An inference drawn from a customer’s email is not their email, and it can still be the reason your advertisement is the one they see. Google says it excludes sensitive categories such as race, religion, sexual orientation and health, and that a user can disconnect their apps at any time.
Nothing here asks anything of an advertiser, which is the part worth noticing. Campaigns you are already running can begin appearing in a markedly more personal context without anyone at your business choosing it.
Barry Schwartz, who reported the change, is explicit that advertising inside AI Mode is not new. What is new is advertising inside AI Mode for the people who connected their own mail and photographs to it.
What to do about it
If you buy Google Search advertising, your own professional advertising rules follow the advertisement into this new surface exactly as they followed it into the old one. A tax professional answers to Circular 230 for what a solicitation claims, a wealth advisor to the Securities and Exchange Commission’s marketing rule, and a builder to state contractor licensing for how a license is displayed. Pull up your three most-served ads this week and read them as though a regulator met them beside a deeply personal answer, because that is now the setting.
Source Search Engine Roundtable, Barry Schwartz, “Google Ads Now In AI Mode With Personal Intelligence,” 19 October 2026, quoting Google’s statement to Search Engine Land of 16 October 2026 and Google’s updated help documentation · seroundtable.com/google-ads-ai-mode-personal-intelligence-42320.html
