01 The consumer data
Google’s spam update moved rankings hard over the weekend
If your traffic dropped this weekend, the cause was probably Google’s update and not anything you changed on your site.
Google announced the update on Thursday and said it would take about two weeks, and fewer than five days have passed.
Google announced the September 2026 spam update on Thursday, September 24 at about 12pm Eastern, and said it would take about two weeks to roll out. Search Engine Roundtable’s Barry Schwartz reported on Monday that the movement began in a big way on Friday the 25th, grew on Saturday the 26th and eased on Sunday the 27th.
A spam update is not the same thing as a general ranking update. It targets pages Google judges to be breaking its spam policies, so the sites that move furthest are usually the ones on the wrong side of that judgement, plus whatever happened to be ranking alongside them.
The number worth holding on to is the calendar. Fewer than five days of a two-week rollout have passed, which means whatever your traffic looks like today is not the number this update finishes on. Schwartz’s own read is that these spam updates are landing harder than the ones from years ago.
Glenn Gabe, who tracks this closely, reported drops spread across different industries and different countries rather than concentrated in one kind of site. No tracking tool has published a figure for how large the movement was, and the thirteen trackers Search Engine Roundtable follows are registering movement rather than measuring it.
So if you run a veterinary practice or an architecture studio and your enquiries went quiet on Saturday, this is the likeliest explanation, and it is not a verdict on your website.
What to do about it
Do nothing to your site this week. A mid-rollout reading is not a result, and a change made now cannot be told apart from the update once it finishes. Write down today’s traffic and rankings so you have a before and an after, then look again after the second week.
Source
Search Engine Roundtable, “Google September 2026 Spam Update Has Big Weekend Impact,” September 28, 2026 · seroundtable.com
02 The research
Researchers watched 240 people follow advice they judged wrong
Your customers can be moved toward a recommendation by an assistant they do not even rate as credible.
Style made no difference, but an assistant that knew something about the person did, and the least expert people moved most.
Hasibur Rahman, Benjamin R. Cowan and Smit Desai ran a preregistered experiment with 240 participants, published on September 21. Each person ranked three stocks the researchers had chosen to be equally defensible, discussed them with an AI assistant, then ranked them again.
The assistant varied in two ways. It could change how it talked, matching the person’s style. Or it could change what it said, using something it knew about the person. Only the second one moved anybody: changing the style had no reliable effect on the rankings at all.
Here is the finding that matters. Participants felt more influenced without judging the assistant as more correct, trustworthy, intelligent, likeable or high-quality. The people who started furthest from its recommendation moved furthest toward it while rating its advice less correct.
The authors’ own phrase is that a personalized assistant can steer decisions among defensible options with only a minimal evaluative trace. In plain terms, the person cannot tell afterwards that they were steered, because nothing about the exchange felt like persuasion.
An exploratory part of the analysis found the effect was stronger among people with less expertise in the subject. That is the opposite of reassuring for anyone selling to first-time buyers. If you are a wealth advisor, an accountant or an independent insurance broker, the customer who knows least about your field is the one most likely to arrive already tilted.
What to do about it
Ask an assistant the question a new customer asks, then ask it again after telling it one thing about yourself: a city, a budget, a job. If the second answer names a different business than the first, that is the mechanism in this study working on your market. Write down what changed.
Source
arXiv, Hasibur Rahman, Benjamin R. Cowan and Smit Desai, “Annie, Are You Okay? How Style- and Context-Based Personalization Shape AI-Assisted Decision-Making,” September 21, 2026 · arxiv.org
03 The platform change
Google confirmed its local ads can show the wrong number
Calls your ads paid for can reach you unlabelled, so the report you judge that spend on undercounts them.
Google’s own explanation is that a location asset can win over the tracking number you set, and one advertiser could not override it.
Local Service Ads are the ads that sit at the very top of Google’s results for service businesses, above the map, with a phone number attached. Google has confirmed that those ads can show the plain business number from a Google Business Profile instead of the tracking number set in Google Ads.
Google confirmed the behaviour twice. Its explanation: “If you are utilizing both location and call assets, ads featuring specific business locations may direct calls to the phone numbers associated with those locations rather than those set within your call assets.”
Advertisers could not override it. One reported having the Google Ads number set as the call asset and in the Business Profile advanced settings, and still seeing the Business Profile number on every active local ad format.
The calls still arrived. What went missing was the label on them, so a month of calls generated by paid local ads may be sitting in your reporting as ordinary calls, or as nothing at all. For a business judging its ad spend on booked calls, that is the number the decision rests on.
On the dates, be careful about what is actually established. The advertiser who raised it, Anthony Higman, says it began on September 1 and told Search Engine Roundtable on the 25th that it is working correctly again. That timeline is one practitioner’s account rather than anything Google has published, and Google has published no explanation and no account of what the gap did to anyone’s reporting.
What to do about it
Pull your Local Service Ads call report for September and compare it against the calls your front desk actually logged. If the two do not match, the gap is this and not a drop in demand. Do not cut a budget on September’s numbers.
Source
Search Engine Roundtable, “Google LSAs Showing Google Business Profile Number, Not Google Ads Number,” September 25, 2026 · seroundtable.com
04 The deadline
Google is upgrading your legacy search campaigns this month
If you run Google search ads, settings you never chose are being switched on for you before the month ends.
Two legacy settings are being auto-upgraded with new matching behaviour on by default, and September ends on Wednesday.
Google’s own post, written by Brandon Ervin, a director of product management on Google Ads, says that campaigns using Automatically Created Assets or the campaign-level broad match setting “will continue to be auto-upgraded starting in September 2026”, and that “remaining eligible Search campaigns with legacy settings will automatically upgrade to AI Max”.
AI Max is Google’s newer bundle of automated matching and creative settings for search campaigns. The upgrade is not a like-for-like swap. Search Engine Roundtable reported in August that campaigns using Automatically Created Assets arrive with search term matching and text customization enabled by default, and campaign-level broad match campaigns arrive with search term matching enabled.
Search term matching means Google can show your ad against searches you never added to the campaign. Switched on by default, in a campaign you did not touch, it changes which searches spend your money.
Two honest limits on what is published. Neither of the pages I could open names an end date for the September window, and neither describes a way to opt out. Trade coverage and Google’s own help documentation put the window at September 1 to 30, and I could not reach either page to confirm it, so treat the end of the month as the practical deadline rather than a stated one.
Google’s advice is the clearest line in the post: “We strongly recommend transitioning to AI Max now to maintain full control over your campaign setup.” Dynamic Search Ads got a reprieve in that same post and do not move until February 2027, which is a separate matter from the two settings above.
What to do about it
Open Google Ads today and filter your search campaigns for Automatically Created Assets and campaign-level broad match. If any are still on legacy settings, upgrade them yourself before Wednesday so you pick the matching and creative settings instead of inheriting the defaults. If none are, you have nothing to do here.
Source
Google, Brandon Ervin, “We’re upgrading Dynamic Search Ads to AI Max,” April 15, 2026, updated June 11, 2026 · blog.google · Search Engine Roundtable, August 7, 2026 · seroundtable.com
05 The quality question
404 Media says OpenAI fired contractors for using AI
The layer meant to catch wrong answers about your business is thousands of contractors, working to a standard nobody outside can read.
Thousands of contractors rate ChatGPT’s answers, and reviewers concluded some of that rating work was itself machine-made.
404 Media reported last week that multiple contractors hired to improve OpenAI’s models have been fired for using AI to train the AI, as summarised by Search Engine Roundtable on September 25. That write-up notes OpenAI uses “thousands and thousands of contractors” on this work.
This is the quality layer. When ChatGPT answers a question about your industry, people in this pipeline are the ones reading sample answers and judging whether they are any good. Some of that judging was itself produced by a machine, and the people doing it were dismissed for it.
Nothing here says a single answer about your business was wrong because of it. What it says is narrower and more useful: the human check behind these answers is contract labour at large scale, and its standards are not published anywhere a business owner can read them.
Forever Cited is in the business of AI visibility, and the honest disclosure is that nobody outside OpenAI can read the standard its raters work to, and that includes us. Saying so is better than repeating a figure we cannot check.
The practical consequence is the same one as last week. You cannot appeal to a quality process you cannot see, so the only thing that reliably changes what an assistant says about your business is what it can find published about you.
What to do about it
Ask ChatGPT what your business does, then check every factual claim it makes against your own website. Anything it gets wrong it read somewhere else, and the fix is publishing the correct version where it can be found, not reporting the answer.
Source
404 Media, “People Training OpenAI’s AI Fired for Using AI to Train the AI” · 404media.co · Search Engine Roundtable, September 25, 2026 · seroundtable.com
