In mid-May 2026, Bogdan Krupin of SE Ranking dropped a one-line post on LinkedIn that the search community has been arguing about ever since. The opening: “How many more wake-up calls does the SEO community need? Relying on raw, unedited AI content for long-term strategy is a losing game!” He was pointing at fresh research from Lily Ray that analyzed 220 websites running scaled AI content programs and found a consistent boom-bust pattern across them (Krupin LinkedIn post, May 2026, pointing at Ray, May 13 2026).
The wake-up call is not “SEO is dead.” It is something much more specific. The shortcut that thousands of agencies, in-house teams, and bootstrapped publishers spent the last 18 months betting on, “let AI write at scale and ride the traffic up,” is producing the same boom-bust cycle that Google’s Helpful Content Update flattened in 2023 and 2024. The packaging is new but the pattern is not.
We want to write the version of this conversation that we wish more business owners had. Not “SEO is dead.” Not “SEO is fine.” Something more useful than either. A look at the actual research Bogdan was pointing to, the eight specific content patterns Ray identified as dangerous, what the broader AI search shift adds on top of it, and what to do about all of it on a Monday morning rather than at a conference in nine months.
What Krupin’s Post Actually Said
Bogdan’s full post is worth reading in his own words. The core of it:
“How many more wake-up calls does the SEO community need? Relying on raw, unedited AI content for long-term strategy is a losing game! Lily Ray just dropped some fantastic research based on data from over 220+ websites. She identified a specific list of ‘cookie-cutter’ patterns that marketers are using to try to dominate their niches, but probably they’re doing it the lazy way by leaning too hard on AI ‘expertise’.”
He then listed the eight content patterns Ray flagged in her research, which we will walk through below. The post landed with 522 reactions, 105 comments, and 40 reposts in under a week. The comment thread is a useful temperature check on the industry. Practitioners who have spent years cleaning up Helpful-Content-Update damage immediately recognized the patterns. Practitioners selling AI content scaling products mostly pushed back, with arguments that mirror the ones we saw in 2022 around AI assistance writ large.
What makes Krupin’s post a real wake-up call rather than another LinkedIn hot take is the research behind it. Ray’s analysis is grounded in 220 publicly identified case-study sites from over a dozen AI content vendors. Most of those vendors published the case studies themselves, which means the sites they listed as success stories are the ones now most visibly losing traffic.
The 220-Site Study: What the Data Actually Shows
Lily Ray’s “It Works Until It Doesn’t” report (Ray, May 13 2026) covers more than 220 domains tracked using third-party SEO data from Ahrefs, with corroboration against the Sistrix Visibility Index. Where the case study identified a specific subfolder where AI-assisted content had been published, Ray narrowed the analysis to that subfolder. Where it did not, she inferred the affected directory from the spike in new pages around the time of the case study’s publication.
The headline numbers, as of May 2026:
- 54% of the 220+ sites lost 30% or more of their peak organic traffic
- 39% lost 50% or more
- 22% lost 75% or more
Within those declines, a single trajectory shape recurs. Rapid growth in organic pages over six to twelve months. An organic traffic peak within roughly three to six months of the content peak. Then a steep decline that erases most of the gain, and frequently drops below the prior baseline, within the following year.
SEO consultant Glenn Gabe has been calling this trajectory “Mount AI” in his writeups: a near-symmetric mountain on the traffic chart, steep up and steep down. Ray’s data shows that across cybersecurity, travel, marketing, SaaS, healthcare, B2B services, crypto, and consumer goods, the shape is the same.
Two patterns inside the data deserve a closer look.
First, the case study timing. In most of the affected sites, the steepest traffic drops took place after the AI content vendor published the case study. Ray notes that this raises an obvious question about whether the case studies themselves accelerated the declines, by giving Google’s reviewers and ranking systems a direct map of where the templated content lived.
Second, the cleanup behavior. A substantial number of the brands featured in those case studies have visibly reduced their content footprints in 2025 and 2026: removing pages, redirecting subfolders, returning 410s on bulk-published AI articles. The case studies are still online. The pages they reference often are not.
The implication for a business owner is uncomfortable but clear. If a vendor pitches you AI content scaling and shows you case study sites, the most predictive thing you can do is open the case study site in Ahrefs or Sistrix and check the traffic curve since the case study was published. In a large share of cases, the curve looks like Mount AI.
The Helpful Content Update Cycle, Rerun
What is hard to overstate is just how recently the SEO industry watched a near-identical cycle play out.
In September 2023, Google launched the Helpful Content Update, which the company described as targeting content that “feels like it was created for search engines instead of people” (Google Search Central, 2023). Six months later, in March 2024, Google rolled out the longest core update in its history, with a stated goal of “reducing unhelpful, unoriginal content in search results by 45%” (Google Search Central, March 2024). Alongside that update, Google formalized a new spam policy: Scaled Content Abuse, defined as generating many pages “primarily to manipulate search rankings,” regardless of whether the production method was human, AI, or both.
Many sites affected by the 2023 and 2024 updates never recovered. We spent significant time in 2024 trying to help clients dig out from those updates, and the diagnostic pattern was always similar. Hundreds or thousands of pages built around templated structure, optimized for queries rather than for readers, often with thin or duplicative information. The fix is rarely a content tweak. It is usually a serious pruning of the page inventory plus a real investment in original content.
The new AI content vendors arrived into a market that had just been told, in the loudest possible terms, that scaled content production was a Google target. Many of them pitched the opposite: AI-driven scale as the path to dominance. Ray’s research suggests that the underlying ranking systems have not changed their stance. They have just gotten faster at recognizing the new patterns.
The Eight Risky Content Patterns
Ray identified eight recurring page templates in the affected sites. Most declining domains used at least three or four. The most aggressive used all eight. For a business owner reading this, these are the patterns to audit your own site against before the next round of updates.
We are quoting Ray’s classifications directly, with our own commentary on what we have seen in client work.
1. Comparison pages at scale
Pattern: /blog/[product-A]-vs-[product-B] published across most reasonable head-to-head matchups in a category. Sometimes including concept-vs-concept pairings that have nothing to do with the publisher’s actual business.
Why it fails: at a small scale, a genuine comparison post can rank and convert. At a scale of 50 to 500 such pages, the template becomes obvious to Google’s classifiers. Every comparison shares the same structure, the same H2s, the same conclusion paragraph that names the publisher as the winner. The variance that signals “real research” disappears.
What to do instead: pick the top five comparisons where you have actual product or service experience, including a competitor you’ve genuinely used or onboarded, and write each one as a long-form review. The five real comparisons outperform the 500 templated ones.
2. The “What is X” glossary
Pattern: /resources/what-is-[term] or /glossary/[term], one page per term, scaled across hundreds of glossary entries. Often scaled across multiple languages from a single source template.
Why it fails: these pages exist to be cited by AI engines as a single-paragraph definition. The first definition of any term is now generated by the AI itself. The hundredth glossary entry on the same topic adds no information gain, which is the property Google has been measuring for years. Ray notes that programmatic multilingual glossaries are an especially fast way to get a domain flagged.
What to do instead: if your readers genuinely need a glossary, write 10 to 20 entries that go beyond definition, with practical examples, common mistakes, and links to the underlying source documents. Skip the scaled version.
3. The “Best [X] for [Y]” listicle
Pattern: the most familiar AI-content template. Origin in the affiliate-content era, now scaled with AI across both broad-category and narrow-niche variants.
Why it fails: Google’s review-content policies, updated repeatedly since 2023, ask for evidence that the publisher has actually evaluated each product on the list. Generated listicles almost never produce that evidence.
What to do instead: limit yourself to lists where you’ve genuinely tested or used each item. If you cannot, write a different kind of article: a buying framework that helps the reader make their own decision.
4. The self-promotional listicle
A variant of the “Best [X]” listicle in which the publisher is itself a competitor in the category being ranked, and lists itself as the #1 best. Ray wrote about this pattern in February 2026 after observing extreme traffic drops on companies running dozens, hundreds, or even thousands of self-promotional listicles, beginning on the same day (approximately January 21, 2026, in what she calls an unconfirmed Google update).
Why it fails: it is biased by definition and Google’s systems are now reading it that way. The pages do not include real evidence the publisher tested all of the alternatives, which Google’s review guidance requires.
What to do instead: if you want to publish comparison content that includes your own product, be transparent about the bias and provide first-party data on how you stack up. Or skip self-promotional comparisons entirely and let third-party reviewers and integrators write them.
5. The competitor-vs-alternatives page
Pattern: /blog/[competitor-brand]-alternatives, or the more programmatic form, dedicated landing pages built for every named competitor in a category.
Why it fails: same family as patterns 1 and 4. At small scales, occasionally useful. At large scale, an obvious template. Ray observed at least one site where the majority of its top-traffic pages were dedicated to individual competitor brand names.
What to do instead: if you want to capture competitor-comparison traffic, build a single well-researched alternatives hub that links to genuine reviews and your own honest case studies.
6. Programmatic location and language scaling
Pattern: one template multiplied across every geography or language a search engine will index, with very little unique content per local landing page. Often produced even when the publisher does not have a real brick-and-mortar presence in the targeted location.
Why it fails: this is one of the oldest patterns Google has been targeting. It has been a rankings trigger for at least 10 years.
What to do instead: build location pages only for locations where you actually serve customers, and make each one specific. Real local examples, real local case studies, real service-area boundaries. Service-area businesses we work with use this approach by default. The version where you launch 1,200 thin city pages overnight is the version that gets penalized.
7. The FAQ farm
Pattern: each page answers exactly one question, with the question in the URL, the answer in the first paragraph, bullet points in the body, schema markup at the bottom. Designed for extraction by AI engines.
Why it fails: at scale, the templated tone often does not match the publisher’s actual brand context. Ray observed extensive FAQ-farming across the dataset. Google announced in November 2025 that it was deprecating FAQ Rich Results, which Ray suspects is partly a response to the new wave of FAQ schema aimed at AI citations.
What to do instead: build a single, thorough FAQ page per topic cluster, with real answers grounded in your customer service data.
8. Off-topic content at scale
Pattern: publishing high volumes of articles that have no connection to the publisher’s actual business. Funny quotes, jokes, baby names, horoscopes, meme content, biographical pieces on a B2B website. This was a huge problem during the 2023 and 2024 update cycles, and is again now.
Why it fails: Google has been explicit that topical focus is a ranking signal. Off-topic scaling dilutes the entire domain’s topical authority and is one of the fastest ways into algorithmic trouble.
What to do instead: stay in your lane. If you serve home services in the Southeast, the funny-quotes-of-2026 page is not your friend, no matter how cheaply you can produce it.
The Late January 2026 Unconfirmed Update
A secondary pattern in Ray’s data is worth its own note. Beginning around January 20, 2026, a wave of sites running self-promotional listicles and other risky GEO-optimized patterns saw organic traffic declines between 40% and 95% over the January to April 2026 window.
Google did not announce or confirm an update by name in January 2026, which is consistent with how Google increasingly handles smaller ranking adjustments. At least 40 sites in Ray’s dataset showed a negative trend starting around January 20. In many cases, the impact was isolated to the blog or subfolder where the AI content lived. In other cases, the effect carried over to the full domain.
The strategic implication is that unconfirmed updates are now meaningful enough to track. If you saw a sharp drop in late January 2026 on a site that was running AI content scaling, you are not alone, and the cause is unlikely to be temporary.
What This Means for Business Owners Specifically
The studies above are written for SEO professionals. The translation for a business owner is shorter.
If you are paying an agency or vendor for AI-driven content scaling, this is the moment to ask hard questions. Not because all AI-assisted content is bad. It isn’t. The same tools that produced the Mount AI charts can also be used well by experienced operators. But the boom-bust cycle described in Ray’s research is concentrated in the specific implementation that most vendors are still selling: high-volume, low-input, templated content where a human is barely in the loop.
A simple test. Open Ahrefs or Sistrix on your own domain. Look at organic pages and organic traffic over the past 12 to 18 months. If you see a Mount AI shape, especially with the steepest growth six to twelve months before the steepest drop, you are inside the pattern. Pruning, redirecting, and retiring the low-value pages is generally the first move. The agencies and consultants who handle Helpful Content Update recovery can be useful here. The vendor who sold you the AI content scaling, generally, is not the one to lead the cleanup.
If you are considering an AI content vendor, Ray ends her piece with five questions worth asking before publishing another page. We’ve reformatted them here.
Does this page exist because a real customer or reader needs it, or because a search engine or LLM might cite it? Could a competitor publish a near-identical version of this page tomorrow using the same prompt? Would you be comfortable if Google, a journalist, or your own customers saw the full list of URLs in this subfolder? Is the article inherently biased, and if so, is the page transparent with users about that bias? Is there any first-party data, expertise, or original perspective on this page that is not already on the first ten results ranking for the query?
If any of those answers is no, the page is probably not worth publishing. Multiply that across a 1,000-page program and the math becomes brutal.
The Wider AI Search Shift That Sits Behind All This
The Krupin and Ray wake-up call lands inside a broader shift that is worth understanding, because it changes the consequences of getting AI content wrong.
Google’s AI Overviews have become a meaningful share of the search experience. The Pew Research Center analyzed 68,879 unique Google searches from 900 U.S. adults during March 2025 and found that when an AI Overview appeared, users clicked on a traditional link 8% of the time, compared to 15% when no AI Overview appeared (Pew Research Center, 2025). In 26% of cases with an AI Overview, the user ended their browsing session without visiting another page.
Ahrefs, working from aggregated Google Search Console data on 300,000 keywords, found that the click-through rate for the top-ranking page on informational queries fell 58% between December 2023 and December 2025 (Ahrefs, 2025 update). BrightEdge’s twelve-month tracking from February 2025 to February 2026 showed AI Overview coverage up 58% year over year, with healthcare at 88% of queries, education at 83%, B2B technology at 82%, and restaurants jumping from 10% to 78% in a single year (BrightEdge, 2026). SE Ranking found AI Overviews in 82.5% of health-specific searches in their own analysis of 50,807 health prompts.
SparkToro’s 2024 zero-click study found 58.5% of U.S. Google searches ended without a click to the open web (SparkToro, 2024). The follow-up Datos and SparkToro State of Search report measured U.S. zero-click rates at 27.2% in March 2025 against 24.4% a year earlier, using a more conservative definition (Datos and SparkToro, 2025).
Stack all of this together and the picture is consistent. Fewer queries result in a click. The clicks that do happen are increasingly concentrated. And the AI surface is harsher about scaled, templated content than the regular SERP ever was, because the same retrieval-augmented generation that powers AI Overviews and ChatGPT-style answers pulls from a smaller pool of sources than the old top-100 organic results.
This is the part Ray is pointing at when she writes that what is dangerous for SEO is now also dangerous for AI search. In the old world, a thin templated page could survive at position 7 and pull a slow trickle of clicks. In the AI Overview world, position 7 with thin content is invisible to the AI and dim on the regular SERP. The downside of the templated approach has gotten steeper, not shallower.
What the Academic Research Adds
The peer-reviewed work on AI citation is small but useful. The foundational paper is Aggarwal et al., “GEO: Generative Engine Optimization,” published at ACM KDD 2024 by researchers from Princeton, IIT Delhi, Georgia Tech, and the Allen Institute for AI (Aggarwal et al., 2024).
The authors built GEO-Bench, a benchmark of user queries across multiple domains paired with the web sources used to answer them. They tested nine optimization strategies and measured two outcomes. Citation visibility (whether the source got cited at all) and impression score (how prominently it was positioned).
The findings line up directly with Ray’s pattern analysis, viewed from the other side. The three content properties that lifted citation visibility the most were citing sources within the content, including specific statistics, and adding direct quotations from authoritative voices. The three that did almost nothing were keyword stuffing, simpler prose, and authoritative-sounding language without underlying evidence.
In other words: the things AI content tools are usually optimized for are the things that don’t move citation. The things they tend to skip are the things that do.
For a business owner, the practical translation is direct. Content that uses real numbers and cites real sources is meaningfully more likely to be cited than content that reads well but sources nothing. Adding named-expert quotes on top of that helps further. We have seen this in our own client work. A client’s blog post that referenced three external studies and included a customer quote got cited in an AI Overview three months after publication. A separate post on the same topic, written in the same tone but with no citations, did not. Same domain, same publication cadence, different content properties.
The Princeton GEO paper and Ray’s pattern analysis are saying the same thing from opposite ends. Authentic, evidenced, source-grounded content gets cited and ranks. Templated, voluminous, undifferentiated content gets demoted. AI tools are neutral. The implementation determines which side of the curve you land on.
Why the Mechanics Inside AI Citation Still Look Like SEO
There is an argument inside the industry between practitioners who say “AI SEO is just SEO” and practitioners who say “you need a totally new GEO strategy.” Both positions are partly right, and both connect to the Krupin wake-up call.
Google’s SVP of Search Nick Fox stated in December 2025 that ranking in AI-generated search features works the same way traditional Google Search has always worked. Danny Sullivan reinforced the same message that month. Authorities analyzed AI Overview citations against organic ranking positions and found that the position-one organic result had a 53% chance of being cited in the AI Overview, with position 10 dropping to 36.9% (Authoritas via Search Engine Land, 2025). Ahrefs found that roughly 38% of AI Overview citations come from the top 10 organic results (Ahrefs, 2025).
This is the “just do SEO” case. The same machinery that determines organic rank largely determines AI citation. Crawlability, structured data, topical authority, and earned trust signals still matter. If you don’t rank, you don’t get cited.
The “you need GEO” case is right that the optimization targets have shifted in subtle ways. The Princeton paper isolated content properties that boost AI citation specifically. The visibility metric is now “did the brand or page appear in the answer surface, and in what position.” Brand mentions across third-party sources have become a stronger signal because LLMs aggregate across the web in ways the old algorithm did not. Schema markup that exposes structured facts is more useful than it used to be when paired with real underlying content.
The honest answer is that the work is mostly the same, the bar is higher, and the measurement layer is broader. A business that does excellent traditional SEO is doing 80% of the work that GEO requires. The remaining 20% is real and worth investing in if your category is heavily affected by AI Overviews.
The wrong move, especially in the wake of Ray’s research, is to treat GEO as a parallel discipline to be outsourced to a different vendor. We’ve seen this play out badly with several prospects. They hired a GEO specialist who recommended an llms.txt file (which no major AI provider documents as actually reading during answer generation), aggressive schema stacking, and a content rewrite that hurt their existing organic rankings (Index Lab analysis of llms.txt, 2025, and a parallel PPC Land analysis, 2025). The traffic loss from broken SEO usually exceeds the speculative gain from AI visibility. Ray has been blunt on this exact point in a separate piece: a bad GEO strategy can destroy your SEO (Ray, 2025).
The Industry Vertical Map
The risk profile for a business depends on what industry you are in and what queries lead to your front door.
Healthcare is the highest-risk informational vertical. 88% of healthcare queries trigger an AI Overview. Pew’s session-end behavior is highest on informational queries. If you run a private practice and your traffic depended on “symptoms of,” “how to treat,” and “what is” queries, that traffic is largely gone. Healthcare AI Overviews lean heavily on already-ranking authoritative sources, with roughly 24% overlap between top-10 ranked pages and cited sources (Search Engine Journal, 2025). If you were ranking, you are likely being cited. Being cited without a click does not bring a patient through the door.
Healthcare is also one of the categories where the eight patterns from Ray’s research are most dangerous. Templated condition-vs-condition pages, “what is X” glossaries written for AI citation, and FAQ farms in clinical voice are all common in this vertical and are exactly the patterns Google’s systems have grown best at flagging.
Legal sits in a different position. CPCs in legal verticals are some of the highest in Google Ads, in part because the queries are commercial, not informational. A query like “personal injury lawyer Atlanta” is far less likely to generate an AI Overview than “what is comparative negligence.” Commercial-intent queries still produce a SERP with prominent organic results, paid ads, and a local pack. Legal Google Ads spend, in our experience managing accounts in this vertical, has not seen a step-function change in efficiency. What has changed is the role of legal informational content. The educational blog that used to attract early-funnel readers and remarket them through email is producing fewer raw visits, and the templated “Best personal injury lawyer in [city]” pattern is in Google’s crosshairs.
Home services are the most mixed picture. Local intent queries like “AC repair near me” still drive heavy SERP traffic and rarely trigger AI Overviews. The Google Business Profile and the Local Services Ads slot have become more valuable, because they sit above the AI Overview. But information-stage queries like “how to troubleshoot a furnace not heating” are increasingly answered in-line. A home services business that built its content engine around hundreds of how-to guides is seeing real organic traffic decline. Service-area pages and conversion-oriented content are largely intact. The risky pattern in home services is programmatic location scaling: 1,200 city pages built from one template, often for cities where the business does not actually serve customers. That is pattern six on Ray’s list, and it has been a penalty trigger for a decade.
For our own client base of home services, legal, and small-practice healthcare businesses, the practical implication is that the strategy diverges by query type within the same client. Service pages and local content are protected and arguably more valuable. Top-of-funnel educational content is in a different game. And any client running AI content scaling on top of either category needs to audit against Ray’s eight patterns before the next round of updates.
What Business Owners Should Actually Do
Here is the part we wish more agencies would publish, because most of the GEO content online stops at the diagnosis.
Audit against the eight patterns first.
Before you do anything else, pull a top-traffic export from Ahrefs or Sistrix. Sort by traffic. Look at the URL patterns. If you see scaled “vs” pages, glossary entries, “best X for Y” listicles, competitor-alternatives pages, programmatic city pages, FAQ-per-page URLs, or off-topic content, you are inside Ray’s pattern dataset whether you intended to be or not. Decide which pages to keep, which to consolidate, and which to redirect or 410.
Audit your visibility across surfaces, not just Google rank.
A useful visibility audit now covers at least four surfaces. Google organic position. Google AI Overview citation. ChatGPT citation for relevant prompts. Perplexity citation for relevant prompts. Tools that span these are emerging quickly. Profound, AthenaHQ, Otterly, and Peec each measure brand presence across LLM answers. The free version of any of these will give you a baseline.
Run a list of 25 to 50 questions a customer would actually ask before hiring you. Run them through Google, ChatGPT, Perplexity, and Gemini. Note whether your brand is mentioned, in what position, and with what framing. That single exercise is worth more than most “GEO audits” we have seen sold for thousands of dollars.
Reorient content around evidence, not coverage.
The Princeton GEO findings, Ray’s pattern analysis, and Google’s own E-E-A-T guidance all point in the same direction. Content with real numbers and cited sources gets cited. Content without them does not. The implication for an editorial calendar is that you are better off publishing one piece of original research per quarter than ten generic posts per month.
Original research can be small. We’ve seen results from an HVAC client publishing an internal analysis of average repair times by system age across their service area. It cost nothing beyond pulling their own service data into a spreadsheet. It got cited in three different AI Overview answers within four months and avoided every single one of Ray’s eight risky patterns.
If you cannot do original research, the next-best move is to write content that pulls together credible third-party sources. Cite the studies. Quote named experts. Show your work. The format that wins is closer to a journalism article than a marketing blog post.
Invest in brand mentions and entity associations.
The mechanism by which LLMs decide what to cite is closer to a knowledge graph weighted by how often and how authoritatively your brand co-occurs with the topic across the web. This makes earned media, podcast appearances, conference talks, and review-platform presence newly important. For a service business, the practical version is local. Get into the right citation databases. Earn review volume on Google Business Profile, BBB, and category-specific platforms. Pitch local journalists. Be a source on industry-specific publications.
Optimize for the queries that still convert clicks.
Not all queries are equally damaged by AI Overviews. Local intent queries, branded queries, and transactional queries still produce clicks at near-historical rates. Pew’s data shows that when a user already knows your brand and types it in, they overwhelmingly click through. Driving brand awareness through paid social, podcasts, and local sponsorships is now a defensive moat against the AI compression of informational search.
We have shifted budgets in this direction for several home services clients over the past year. A typical reallocation looks like trimming the third-tier informational content investment and putting that budget into Local Services Ads, branded search, and a small Meta retargeting layer for past visitors. The conversion-per-dollar math has been favorable.
Build the measurement layer first.
The single most consequential change a business owner can make is improving the measurement layer. The agencies that are going to be useful to you in 2026 are the ones who can tie marketing inputs to revenue, not the ones who report clicks and impressions. Call tracking with conversation analysis to score lead quality. Form tracking with revenue attribution. GA4 properly configured with conversion events tied to actual revenue, not vanity events. A CRM that reports closed-won by source. If a vendor cannot tell you which specific marketing actions produced last month’s revenue, they cannot make good decisions in an environment where the inputs are changing this fast.
What Not to Do
A few things in the AI content and GEO conversation have turned into bad advice that we keep having to talk clients out of.
Buying into the volume-first AI content pitch. Ray’s research is now the canonical citation against this. The 220+ site data shows what happens when scale becomes the goal.
Adding an llms.txt file as a magic bullet. No major LLM provider documents reading llms.txt during answer generation. Over 844,000 sites had implemented it by October 2025 according to BuiltWith, with effectively zero confirmed adoption from OpenAI, Anthropic, Google, or Perplexity. There is no harm in adding the file. There is harm in treating it as a strategy.
Stuffing schema markup or adding fake FAQ sections to engineer AI extraction. AI surfaces do read schema, but they are reasonably good at detecting low-quality structured data. Aggressive schema without underlying content quality has not produced reliable lift in our testing, and Google’s deprecation of FAQ Rich Results in late 2025 is the clearest signal that the platform is now actively working against this pattern.
Hiring a GEO-only vendor without a clear connection to your SEO and content operations. The Princeton paper, the BrightEdge data, the Authoritas citation analysis, Google’s own statements, and now Ray’s 220-site dataset all converge on a point that the industry sometimes resists. AI visibility is mostly downstream of SEO visibility. A GEO vendor without an SEO foundation is selling a roof without a house.
Believing that everything will collapse, or that nothing will change. The traffic decline in informational verticals is real and is showing up in our clients’ analytics. The volume increase in branded and local queries is also real. The right posture is neither panic nor denial. It is a clear-eyed audit followed by a budget reallocation.
A Note on the Industry Size Argument
One last data point sometimes gets cited in this conversation. The global SEO software market was estimated at USD 74.6 billion in 2024 and is projected to reach USD 154.6 billion by 2030 according to Grand View Research, a compound annual growth rate of 13.5% (Grand View Research, 2025).
This number is real. It is also somewhat misleading as a defense of the practice. The industry is growing because the work is getting harder, not because it is getting easier. More agencies, more tools, more reporting platforms, more visibility into more surfaces. The economic activity around managing search visibility is increasing because the surface area is fragmenting. That is consistent with the Krupin wake-up-call framing, not against it.
What the industry size data does tell you is that this is not a discipline that is going to disappear. Practitioners who reframe what they do and how they measure it are doing better than ever. Practitioners who keep selling the 2019 deliverables, or the 2024 AI scaling pitch, are getting compressed out of the market.
Where This Leaves a Business Owner on Monday Morning
If you have read this far, you probably already know the work is real. The question is what to prioritize this quarter.
A reasonable 90-day plan looks like this.
Week one and two, audit your site against Ray’s eight patterns. Pull a top-pages export from Ahrefs or Sistrix. Identify any subfolder that fits one of the templates. Decide which pages to keep, consolidate, or retire.
Week three and four, run the visibility audit across surfaces. List 30 questions your customer asks before hiring. Check Google, ChatGPT, Perplexity, and Gemini for your brand presence. Note the gaps.
Month two, audit your existing top-traffic content. Sort by year-over-year traffic change. The content that has dropped sharply is mostly informational. Decide which pieces are worth rewriting with stronger evidence, citations, and structured data, and which are worth retiring. Build or fix the measurement layer in parallel. Call tracking, form attribution, GA4 conversion events tied to real revenue, CRM source reporting. If you cannot answer the question “which channel produced last month’s closed revenue” within five minutes, this is the highest-impact investment you can make.
Month three, reallocate. Move budget from low-value informational content toward original research, branded search, local presence, and the channels that still convert clicks for your specific industry. Start measuring AI-surface visibility quarterly the way you measured organic rank monthly five years ago.
None of this requires panic. It does require honesty about what is actually happening. Bogdan’s post was a wake-up call because so many marketers were still acting as if the SERP and the AI surface would reward the scaled, templated content that Google has spent the last three years explicitly working against. They won’t. The work that wins is still legible, still grounded in fundamentals, and arguably more interesting than the work that used to win. It is also no longer optional.
If you want a second set of eyes on what your business looks like across these surfaces, including a quick audit against Ray’s eight patterns, we are happy to take a look. No pitch. Just a clear read of where you show up, where you don’t, and which pages might be working against you.
Sources
- Aggarwal, P. et al. (2024). GEO: Generative Engine Optimization. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. arxiv.org/abs/2311.09735
- Ahrefs. (2025, December). Update: AI Overviews Reduce Clicks by 58%. ahrefs.com/blog/ai-overviews-reduce-clicks-update
- Ahrefs. (2025). Update: 38% of AI Overview Citations Pull From The Top 10. ahrefs.com/blog/ai-overview-citations-top-10
- BrightEdge. (2026). AI Overviews at the One-Year Mark: Presence, Size, and What They’re Citing. brightedge.com
- Datos and SparkToro. (2025). State of Search Q1 2025: Behaviors, Trends, and Clicks Across the US and Europe. datos.live/report/state-of-search-q1-2025
- Google Search Central. (2023, September). September 2023 helpful content update. developers.google.com
- Google Search Central. (2024, March). New ways we’re tackling spammy, low-quality content on Search. developers.google.com
- Grand View Research. (2025). SEO Software Market Size, Share & Growth Report, 2030. grandviewresearch.com
- Index Lab. (2025, October). LLMs.txt: Does It Actually Work? indexlab.ai
- Krupin, B. (2026, May). LinkedIn post: “How many more wake-up calls does the SEO community need?” LinkedIn
- Pew Research Center. (2025, July 22). Google users are less likely to click on links when an AI summary appears in the results. pewresearch.org
- PPC Land. (2025). LLMs.txt adoption stalls as major AI platforms ignore proposed standard. ppc.land
- Ray, L. (2026, May 13). It Works Until It Doesn’t: AI Content Strategies That Backfire. lilyraynyc.substack.com
- Ray, L. (2025). Your GEO Strategy Might Be Destroying Your SEO. lilyraynyc.substack.com
- Search Engine Journal. (2025). Google Health AI Overviews Cite YouTube More Than Any Hospital Site. searchenginejournal.com
- Search Engine Land. (2025). AI Overviews optimization guide: Ranking in Google AI search. searchengineland.com
- SparkToro. (2024). 2024 Zero-Click Search Study. sparktoro.com
The Hutch Group is a US-based digital marketing agency specializing in performance marketing for home services, legal, and small-practice healthcare businesses. We measure marketing in revenue, not clicks.
