Lily Ray, founder of the search consultancy Algorythmic and vice-president of SEO and AI search at Amsive, argued in an essay published on Sunday, October 4, 2026 that Google is close to releasing one of its larger ranking updates. Her case rests on four spam updates in 2026, documentation Google rewrote on October 1 and October 2, remarks by Google staff at a developer event in Barcelona, and her own measurement of 100 business software queries, in which brands that ranked themselves first on their own "best of" pages were left out of the recommendation in 83% of the Google AI answers that cited those pages.

In Short

A well-known search consultant wrote that Google is probably about to make a big change to how it ranks websites, because it has just rewritten its rules on AI-written pages and fake authors and has been running spam clean-ups more and more often. That matters to any company that churns out large numbers of AI pages, invents expert writers, or pays to be named in "best of" articles so that chatbots recommend it. Nobody outside Google knows when, or whether, such an update will land, but Google's own help pages now spell out what it treats as low quality, so sites doing those things can no longer say they were not told.

A forecast built on paperwork

Ray's essay, published on her Substack under the title "Prediction: the next massive Google update is around the corner", does not claim inside knowledge. "Nobody outside Google knows when the next big update will land or exactly what it will target," she writes, describing the piece as an informed opinion drawn from years of watching what she calls the cycle of SEO. A tactic is found. It spreads. Eventually the search engine, and now the AI companies, build systems to stop it working. In her telling, "it's usually not a question of if Google will respond, but when."

Her conclusion is firm all the same. "I think by this point, it's fair to assume from these recent communications that something big is on the horizon," she writes.

The forecast comes from a practitioner with a commercial stake in the subject, and that is worth stating plainly. Ray says she has spent much of 2026 working with companies Google has already hit, some algorithmically and some through manual actions, and that she has spent most of the year reverse-engineering patterns across hundreds of other affected sites. Audit and recovery work of that kind is part of what consultancies in her field sell. She has also made a version of this call before: in December 2025 she predicted a major enforcement action against AI-generated content in 2026, three months before Google released the March 2026 spam update.

What Google changed on October 1 and 2

The generative AI guide

On Thursday, October 1, Google added three sentences to the accuracy section of its guidance on using generative AI content. The third, quoted by Ray, reads: "It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing." A reworded sentence extends that review to title elements, meta descriptions, structured data and the alt text attached to images, all of which can appear in search results. The edit was confined to one section of a page Google first added on May 21, 2025, and Barry Schwartz observed that "critical" is a word Google rarely uses in its documentation.

Ray treats this as the strongest of her signals, because scaled AI content is, in her account, the most common factor behind the demotions and penalties she has examined this year. Such content is often full of errors, she argues, and Google's own AI products can then repeat those errors as fact in AI Overviews and AI Mode.

One detail in her reading is contested. Ray writes that the page now points readers to the sections of the quality rater guidelines on scaled content abuse and on content made with little or no effort. Comparisons by Search Engine Journal, against a version dated December 10, 2025, and by Relevant Audience, a Bangkok-based agency, against an archived copy from September 20, 2026, found the paragraph citing raters' sections 4.6.5 and 4.6.6 already in place before the edit. What changed was the accuracy language. Google's changelog gave its reason as wanting "to get our documentation in sync with our presentations we use at our developer events," without saying which new passage drew on the raters' material.

Main content and effort

On Friday, October 2, Google added two sections to Creating helpful, reliable, people-first content, the document Ray describes as the clearest explanation of what Google's ranking systems try to reward. Schwartz reported the additions at 06:00 Eastern that morning. The first defines main content broadly enough to include user reviews, forum posts, comments, interactive tools, images and video, which means a page can be judged on material its publisher did not write. The second sets out four qualities raters are trained to look for: effort, originality, talent or skill, and accuracy.

Main content quality, Google wrote, is "one of the most critical factors" in assessing a page. The definition of effort leaves little room for interpretation where automation is concerned: "...using generative AI to produce large amounts of text without manual oversight or curation represents little to no effort." For topics touching health, money and safety, which Google groups as YMYL, content "must be highly accurate and consistent with established expert consensus."

Authors who do not exist

The same page gained a paragraph that Marie Haynes spotted, Cyrus Shepard circulated and Glenn Gabe flagged with one word, "BEWARE." It reads, in part: "Fabricating creator profiles (such as by using AI-generated headshots, made-up names, or false credentials to make content appear as if it was written by human experts) is a form of deception." Google continues: "Any form of deception makes a page untrustworthy to both users and our automated quality systems, and is a signal of a low-quality page."

Raters were already told to treat invented authors as deception. The change, as Ray reads it, lies in the audience and the mechanism. The language now addresses site owners directly and names Google's automated systems, not human raters alone. According to Ray, Search Engine Journal found the paragraph was absent from the page in June. She cites an AI-generated "art therapist" who was quoted more than 30 times in the media before anyone noticed, and says invented authors were a common trait among sites hit by the Helpful Content Update and by core updates since, including in 2026.

Barcelona

The documentation changes coincided with Search Central Live Deep Dive Europe, held in Barcelona from September 30 to October 2. Drawing on a recap by John Campbell of the agency ROAST, Ray reports that Gary Illyes said scaled content is becoming more of a problem than link spam. Duy Nguyen, a search quality analyst at Google, attributed the recent frequency of spam updates to the sheer volume of new content, said the team now uses AI to catch more of it, and defined quality with the same four attributes that appeared in the documentation that day.

The same recap carried a figure Illyes put on screen: Google filters out 40 billion spam pages a day, though Google's earlier public material framed the same number around pages detected rather than removed. Illyes also said core updates assess content at page level rather than domain level, which is why some pages on a site rise while others fall.

Ray's interpretation of the timing is pointed. "From my perspective, Google puts this guidance in writing so that when an update hits and site owners are caught off guard, it can point back to the documentation and show that the warnings were there all along," she writes.

A precedent list with no denominator

Her historical argument is that Google tends to put its intentions in writing shortly before it acts. The list begins in January 2011, when Matt Cutts wrote that spam had increased and people were asking for stronger action on content farms; Panda followed in February. In March 2012 Cutts said at SXSW that Google was working on a penalty for over-optimised sites, and Penguin arrived in April. On July 20, 2018, the rater guidelines added emphasis on the reputation of sites and their creators, and the "Medic" core update followed on August 1, 12 days later. In September 2023 Google replaced "written by people" in its guidance with "helpful content created for people"; that November, Danny Sullivan, then Search Liaison, said at BrightonSEO San Diego that he did not want to say "buckle up" because people would panic.

The March 2024 core update is her central precedent. Released on March 5, 2024 alongside three new spam policiescovering scaled content abuse, expired domain abuse and site reputation abuse, it ran for 45 days, the longest core rollout Google has on record. Google said afterwards that low-quality, unoriginal content in its results had fallen by 45%, a figure drawn from its own evaluations with no published methodology.

Two more recent cases fit her pattern. On January 23, 2025, Google added 11 pages on spam and low-quality content to the rater guidelines; the March 2025 core update began on March 13, seven weeks later. On May 15, 2026, Google stated that its spam policies apply to AI Overviews and AI Mode and published a guide on optimising for AI features; the May 2026 core update started six days later, on May 21, and the June spam update followed the next month. Google even gave a week's notice of the Helpful Content Update before its rollout in August 2022.

What the list cannot show is a rate. It records documentation changes that were followed by updates and omits any that were not, so it establishes sequence rather than probability. The lead times it does record run from six days to roughly six months. And Google has drawn no connection between the October edits and any forthcoming release: its changelog attributes them to developer-event material, and the October 1 change came with no ranking change, penalty or enforcement mechanism attached.

Spam updates on a shorter clock

Ray's most measurable evidence concerns cadence. Google has run four spam updates in 2026: March 24 to 25, June 24 to 26, August 18 to 21, and a fourth that began on September 24. In all of 2025 there was one, the August 2025 cycle, which ran for 27 days. The intervals between starts have narrowed from 92 days to 55 and then 37; the first two gaps were also recorded when the September update began.

Durations have moved the other way. March's update finished in 19.5 hours, June's in about 48 and August's in three days. The first three were each expected to take "a few days." September's came with a warning that the rollout may take up to two weeks, the window Google normally attaches to core updates. Search Engine Roundtable recorded heavy movement from September 25 to 27, and on September 28 Ray posted on X that early signs pointed to "highly templated, programmatic, (probably) AI-generated pages, like one page for every phone number/area code." The post had drawn 35,400 views when Ray captured it. As of her writing, Google had not marked the update complete, and its window could run to October 8.

There is a small tension in Google's own figures. The recap of Illyes' Barcelona session gave spam updates a rollout of one to two days, against a September cycle still open after ten days.

Spam updates refresh the classifiers behind SpamBrain, Google's machine-learning detection system, and the rulebook those classifiers enforce has grown during the year. On April 13, Google made back button hijacking an explicit violation, with enforcement from June 15. That rule has already carried a price: the withdrawal of a back-button feature took $20 million out of Taboola's second-half profit. On April 14, according to Ray, Google said spam reports could now be used to take manual action; ten days later it stopped processing reports containing personal information after complaints that reporters' identities could be spoofed. On May 15 came the extension of the spam policies to AI Overviews and AI Mode.

Ray also cites SAFE, a Google research system for detecting AI-generated spam, which Search Engine Journal reported on in September and which the paper says is in early deployment. She notes that it appears focused on video. That caveat matters: every reference in the three-page paper concerns YouTube channels and videos, with no stated connection to web search or to SpamBrain.

The data on self-ranking pages, and the data that disagrees

The most specific new figures in the essay concern pages on which a company ranks itself first in its own category, a tactic built to be repeated by AI answers. Ray first measured it in June, finding that "the brand that ranked itself #1 was left out of the actual recommendation 69% of the time." Her update, run across 100 "best B2B software" queries in September 2026, puts that share at 83%. Compared with June, she writes, Google's AI Overviews now cite 38% fewer of these vendor-written "best of" pages, and in 70% of cases the page helps competitors get recommended while the brand that published it is left out.

These are a consultant's own measurements, on a small sample from one vertical and one AI surface. The essay does not set out how recommendations were classified or how the June and September query sets were matched.

Other datasets point in different directions. In January 2026 Ray had already observed dozens of companies taking site-level hits, one of them holding 2,000 articles that each claimed first place in its category. SISTRIX later reported visibility losses of up to 49% for software sites built on such pages. Kevin Indig's study of 5.32 million organic result rows, by contrast, found that 62% of vendor-authored list pages had gained estimated organic traffic since January, with a median rise of 38% - a traffic measure with no connection to Ray's citation figure despite the identical number. Scrunch, an AI visibility vendor, analysed roughly 10,000 URLs cited by seven AI assistants and found that a cited self-ranking page lifted its author's recommendation rate from about 4% to roughly 7%, with a competitor recommended instead in 24.3% of answers citing such a page.

Can 24.3% and 70% both be right? Possibly. Scrunch measured seven assistants over a few weeks in mid-2026; Ray measured Google's AI Overviews alone, three months later. Neither study has been independently audited, and both come from parties that sell services around AI visibility.

Mentions for sale

Ray reserves her sharpest language for paid brand mentions, which she calls the 2026 version of paid link building, aimed at generative AI responses rather than rankings. Her inventory of what is on offer is specific:

  • placements in third-party "best X" and "top X" lists, sold per placement and priced by the host site's Domain Rating or Domain Authority, often from about $135 to $250 each;
  • packages of "AI brand mentions" inserted into existing lists, reviews and buying guides, such as five mentions for $1,500;
  • list placements that guarantee a dofollow link;
  • agency playbooks noting that, if outreach fails, "many of these sites accept paid placements", sometimes for one-off payments of $100 or $200;
  • fully managed "offsite" services from AI visibility platforms that handle outreach, content, negotiation and payment;
  • integrations that pass the sites AI tools cite to a placement network, which automates outreach, drafting, negotiation and payment at fixed prices or through what are sold as "reverse mentions".

The rules these run into are older than the market. Google's link spam policy lists "exchanging money for links, or posts that contain links" as a violation; paid links are permitted only when marked rel="sponsored" or rel="nofollow". The FTC generally requires paid endorsement relationships to be clearly disclosed, which Ray says the sales pages rarely mention. On mentions without links, Google's guide to generative AI features says "seeking inauthentic 'mentions' across the web isn't as helpful as it might seem," and its spam policies now cover "attempting to manipulate generative AI responses in Google Search." At a Search Central Live event in May, Illyes reportedly compared buying mentions to buying links. Demand for this kind of influence is not new: in January 2026, a Wall Street Journal investigation documented businesses paying to shape what ChatGPT recommends.

Reddit, a favoured venue for such campaigns, has been acting on its own account. In March it began labelling permitted automated accounts and asking some suspicious ones to verify that a human is behind them, saying it removed an average of 100,000 bot accounts a day. In June chief executive Steve Huffman put the number of spam views blocked at up to 23 million a day. Ray adds a third figure, from July: automated systems catching 25,000 spammy posts and comments a day as Reddit tries to stop brands planting posts meant to win recommendations from ChatGPT and Gemini. The three numbers count different things - accounts, views and posts - and cannot be compared directly. Reddit also banned four subreddits used in a GEO experiment that relied on automated upvotes and comments, according to Ray.

Three other analysts, one conclusion

Ray presents Gabe, Haynes and Shepard as corroboration. Gabe, who has posted running "Spam Update Notes" through the September cycle and documented the August update hitting scaled, programmatic, AI-generated and thin affiliate content, wrote after the documentation changes: "For those still wanting to scale via AI or programmatic content, Google is clearly trying to tell you something." The following day he added: "...with all of the doc updates, and Google presentations from the past year, I feel like we are going to see a very big update from Google." He and Schwartz also discussed the likelihood of a large update in minutes eight to ten of "SEO for Paws", a charity webinar.

Haynes, who has studied Google's algorithms since 2008, ties the timing to Gemini 4 Argon, Google's newest model, arguing that large jumps in model capability have often been followed by large ranking changes. Her conclusion carries an exclamation mark: "I predict we will soon have a significant core update!" In July she argued that a growing share of pages sitting in Search Console's "crawled - currently not indexed" state were there because their content was ordinary, not because anything was technically broken.

Shepard has just surveyed more than 130 practitioners on ranking factors; the published results, drawn from 131 respondents and 13,665 data points, placed content quality among the three strongest positive signals. On October 2 he called Google's changes "significant" on X, in a post that reached 20,000 views. When Haynes asked whether Google had written them "to prepare us for the aftermath of an upcoming update," Shepard replied: "As Hamlet would say, 'that is the question.'"

Agreement among four people reading the same public documents is, however, not the same as four independent sources of evidence. None claims knowledge of Google's release schedule, and all are responding to the same handful of pages.

What Ray says the affected sites had in common

From her client work and her wider analysis, Ray describes six recurring traits among sites hit in 2026:

  • Scale: hundreds or thousands of pages added in a short window, often on topics where the site had no history, frequently after long spells of low output.
  • Templates: one page per city, phone number or product comparison, with a few words changed between them - a version of doorway pages, and in her account a common pattern among affected sites for at least 15 years.
  • Nominal review: sites that claimed a human editor checked every AI draft were devalued anyway, because the problem Google appeared to judge was the overall pattern of publishing rather than the quality of individual sentences.
  • Invented authors: made-up names and AI-generated headshots, including on health and finance topics.
  • AI-facing self-promotion: self-ranking and comparison pages written mainly to be cited by ChatGPT and AI Overviews, which many sites hit by Google's unconfirmed January 2026 update had published by the hundred or the thousand.
  • Spillover: when these sites lost organic visibility, she writes, their citations in AI Overviews, AI Mode and often ChatGPT fell as well.

Ray says she has moved most of her clients to a "less is more" content strategy. The essay closes with six recommendations for site owners, from auditing author pages to having a person fact-check every AI-generated title, description and alt text, while conceding that the timing is uncertain. "I do believe it's a matter of when, not if," she writes.

Why this matters for the marketing community

Organic enforcement no longer stays organic. Since December 2024, Google Ads has disapproved ads pointing to destinations removed from Search through manual action, and sites found in breach of the scaled content policy can lose eligibility for Google-served ads. A crackdown on templated and mass-produced pages is therefore also a question about the landing pages advertisers pay to send traffic to, and about the ad-funded inventory programmatic buyers purchase on the open web, where DoubleVerify traced a network of more than 200 AI-generated domains to tens of millions of impressions.

The paid-mention market Ray describes is sold into marketing budgets, often by vendors positioned between SEO and AI visibility. If Google treats those placements as link schemes or inauthentic mentions, the cost lands twice: in the placement fee, and in ranking credit that, once spammy links are neutralised, does not come back. Self-ranking pages carry an exposure of their own. ClickUp's blog lost 97.6% of its traffic while carrying 249 competitor-alternatives pages that ranked ClickUp first, though the decline tracked a sequence of core and spam updates rather than any single enforcement action.

For agencies and in-house teams, the October documents change the record more than the rules. A sentence requiring manual review of generated text does not alter ranking systems, but it does change what a publisher or an agency can later claim about its process when a client disputes a traffic loss.

Ray's own recent record shows why AI answers are the contested ground. In January 2026 she published a fabricated core update article on her blog as a test, and AI Overviews repeated it as fact within 24 hours, with her site the only source. In late September, Google's AI answer on her SEOktoberfest title showed her own consultancy's site as the visible source for the award sentences, although the event's host had confirmed the win publicly three days earlier and further sources may have sat behind collapsed links. The first claim was false and the second true. In both, the visible citation led back to the claimant, and the display did not show whether independent confirmation counted for anything - the same opacity that self-ranking pages and bought mentions rely on. Whether a spam or core update can remove it is a different question from whether Google demotes the sites using it.

Then there is the forecast itself. Google released core updates in March and May 2026 and confirms several broad updates in a typical year, so some update will follow the October edits. The measurable claim in Ray's essay is narrower: that the next confirmed release will fall hardest on scaled AI pages, invented authors and AI-facing self-promotion. That part, unlike the timing, can be tested once an update arrives.

Timeline

Summary

Who: Lily Ray, founder of the consultancy Algorythmic and vice-president of SEO and AI search at Amsive, with supporting comments from Glenn Gabe, Marie Haynes and Cyrus Shepard. Google is the subject, through its Search Central documentation and staff including Gary Illyes and Duy Nguyen.

What: An essay predicting a large Google ranking update aimed at scaled AI content, invented authors and tactics built to influence AI answers. It cites four spam updates in 2026 with intervals shrinking from 92 days to 37, documentation rewritten on October 1 and 2, Ray's measurement that brands ranking themselves first were left out of 83% of AI Overview recommendations in 100 B2B software queries (up from 69% in June), and a 38% drop in such pages being cited. Other datasets, from Kevin Indig and Scrunch, point in different directions, and one of Ray's documentation claims is contradicted by version comparisons.

When: Published on Sunday, October 4, 2026, while the September 2026 spam update, begun on September 24, was still rolling out. The documentation changes it cites date from October 1 and October 2, 2026.

Where: On Ray's Substack, concerning Google Search worldwide, including AI Overviews and AI Mode, with supporting material from Search Central Live Deep Dive Europe in Barcelona.

Why: Ray argues that Google has historically put its intentions in writing before major crackdowns, and that the current wording on effort, accuracy and fabricated authors, together with the accelerating spam cycle, follows that pattern. Google has named no update and drawn no link between the October edits and any release.