SEO Fundamentals Are the Base for GEO, and Gimmicks Are Being Punished

For about a year, one tactic showed up in nearly every GEO playbook: publish a “best companies in our category” list on your own blog, and put yourself at number one. It worked, too. Brands that did it started appearing in Google’s AI answers, and plenty of teams treated it as a shortcut to AI visibility.

Then, in late January 2026, something shifted. Lily Ray, one of the more careful watchers of Google updates, noticed a cluster of large B2B and SaaS blogs losing visibility at almost the same moment. One of them, a well-known brand she describes as an $8 billion company, lost 49% of its organic visibility in about two weeks. Its blog was carrying 191 listicles that ranked the company first.

The lesson isn’t really about listicles. It’s about what happens when a tactic built for AI answers collides with the search systems those answers rely on. Google now says in its own documentation that optimizing for generative AI search is still SEO, and the data we have so far points the same way: when visibility drops in Google, AI citations tend to follow.

This piece covers what the evidence shows, what is actually being punished, and what B2B SaaS teams should do with their content over the next 90 days. It’s written for marketing leaders who’ve been told to “do GEO” and want to know what is safe to build on.

What the data actually shows

Start with the listicle story, because it has the clearest fingerprints. In a February 2026 analysis, Lily Ray showed several large B2B and SaaS sites that lost between 29% and 49% of their Google visibility, starting around January 19 to 21. In most cases the blog or resource folder took the hit. The hardest-hit sites she examined shared a common thread: they had all published self-promotional listicles, meaning “best” articles where the publisher ranks itself first. Counts ran from 10 to 340 such pages per site.

Two caveats come straight from her own write-up. Google never confirmed an update, so nobody can say for certain what triggered the drops. And listicles weren’t the only issue. The same sites often had rapidly scaled, templated content, artificially refreshed dates and misused schema. That 191-listicle blog, for instance, had roughly 30,000 articles indexed, and a few hit sites had no self-promotional listicles at all.

The second piece of evidence is about AI. In a follow-up study, she looked at 11 content folders that lost Google traffic between January 20 and February 16, 2026, with declines ranging from 5.7% to 53.1%. All 11 also lost AI search citations, averaging 22.5%. ChatGPT citations fell 27.8% on average and Google’s AI Mode 23.8%, against a 26.7% average drop in organic traffic. Gemini fell for 10 of the 11. Perplexity was the outlier: only 4 of the 11 declined, and it averaged just 2.9% lower, which fits the idea that it retrieves from a different index.

The third piece is a boom-and-bust pattern. In May 2026, Ray tracked more than 220 sites that had been publicly named as customers of AI content tools. Of those, 54% lost at least 30% of their peak organic traffic, 39% lost at least half, and 22% lost three quarters or more. The typical shape was fast growth in pages over six to 12 months, a traffic peak a few months later, then a steep fall that erased most of the gain within a year.

All three analyses rely on third-party traffic estimates, small or hand-picked samples, and observed correlation. Ray says so herself. Treat them as strong signals, not proof of cause.

Why AI search follows Google

It helps to remember how these answers get built. Google’s own guide to optimizing for generative AI features, published in May and last updated in July, says AI Overviews and AI Mode are rooted in its core ranking and quality systems. They use retrieval, which pulls pages from the Search index, and query fan-out, which runs related searches behind the scenes. A page has to be indexed and eligible to appear with a snippet before it can show up in those features at all.

From that, Google draws a blunt conclusion: from Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and so it’s still SEO. The guide also lists things you can ignore for Google: llms.txt files and other special markup, chopping content into tiny chunks, rewriting copy only for AI, chasing inauthentic mentions, and fixating on structured data.

This doesn’t necessarily mean the debate is closed. Google is describing its own product, and the guide speaks to Google Search, not to ChatGPT or Perplexity. That’s where Ray’s data adds something. In her study, ChatGPT’s citations moved with Google visibility more closely than Gemini’s did, which she reads as a sign that ChatGPT leans on Google results. She is clear that this has not been officially confirmed. Perplexity behaved differently.

The practical point is still simple. For most B2B SaaS companies, Google sits upstream of a large share of the discovery that happens in AI answers. Ray argues that the quickest route to losing visibility in AI search may be losing it in Google first, and the evidence so far fits that reading.

What’s actually being punished

The common thread isn’t AI, and it isn’t listicles on their own. It’s content that exists for a search engine rather than for a reader. Four patterns come up again and again.

• Self-promotional listicles. A page that says “best” implies an independent evaluation. When the publisher ranks itself first, with no disclosed method and no evidence it tested the others, it reads as biased. Ray points to Google’s guidance on review content, which asks for real evidence of first-hand testing. These pages rarely include any.
• Content at scale. Google’s guide says that creating separate pages for every variation of a query, including fan-out queries, mainly to manipulate rankings or AI responses violates its scaled content abuse policy. Ray’s 220-site analysis names eight templates that keep appearing on declining sites, including mass comparison pages, “what is X” glossaries, competitor “alternatives” pages, one-question FAQ pages and programmatic location pages. Declining sites typically used three or four of them, and the most aggressive used all eight.
• Fake authorship. In early October 2026, Google updated its helpful-content documentation to say that fabricating creator profiles, such as AI headshots, made-up names or false credentials, is deceptive and a signal of a low-quality page. Search Engine Land covered the change.
• Artificial freshness and markup tricks. Ray also saw recent dates added to articles that hadn’t been substantially updated, “2026” stuffed into titles early in the year, and ratings schema used on pages that weren’t eligible for it.

Then there’s the September spam update. It ran from September 24 to October 8, close to 14 days, and Search Engine Journal notes that was longer than any of the three earlier spam updates this year, which each finished in under three days. Google hasn’t said what it targeted. Early data from Diib, covering about 32,600 sites, showed the median share of visits from Google search slipping from 34% to 31%, with 10% of sites losing half or more of their share, compared with 5% in the earlier comparison period. That is an early read, and rank-tracking data was distorted during the rollout, so check Search Console before blaming the update for a dip.

Ray has also predicted that another major Google update is close. That’s her read of the signals, not an announcement, and we haven’t verified the reasoning behind it.

What still works

The good news is that the fundamentals aren’t exotic. Google’s guide says unique, compelling and useful content will likely influence your presence in generative AI search more than any other suggestion in it, and most of the rest covers familiar ground.

• Non-commodity content. Google contrasts generic tips with content built on first-hand experience. For a B2B SaaS company, that means implementation lessons, real customer data, original benchmarks, pricing and ROI breakdowns, and teardowns written by people who did the work.
• A clean technical foundation. Pages need to be crawlable, indexed and eligible to show with a snippet. Reduce duplicate content, keep the site fast and readable, and use semantic structure where it helps people. Structured data is still worth using for rich results, but Google says no special markup is needed for AI features.
• Real, verifiable authorship. Named experts with genuine credentials, and an author page that a buyer could check. Given Google’s October change, invented bylines are now a liability.
• Honest comparison content. Buyers do want comparisons. The version that holds up discloses the publisher’s relationship to the products, explains how they were evaluated, shows evidence of actual testing and treats alternatives fairly.
• Fewer, better pages. Ray suggests asking whether each page exists because a reader needs it or because a search engine might cite it, whether a competitor could publish a near-identical version tomorrow with the same prompt, and whether it holds anything the top ten results don’t already have.
• AI as an assistant, not a factory. Ray is clear that the tools aren’t the problem. Research, briefs and data synthesis with a human expert in the loop are fine. Publishing volume without review is where sites got into trouble.

One practical habit is worth adding. Use the generative AI performance report in Search Console to see where your pages appear in AI features on Google, and be wary of third-party tools that claim to use Google’s internal metrics. Google’s guide says no outside tool has access to those systems.

A 90-day content audit

Before you publish anything new, audit what is already live. Sort every page in your blog or resource center into one of four buckets.

To find the riskiest pages quickly, Ray suggests searching Google for your own domain with a query like site:yourdomain.com/blog/ intitle:best "1. yourcompany". It’s a rough check, but it surfaces self-ranking listicles fast.

A realistic sequence looks like this:

1. Days 1 to 30: inventory and score. Export every indexed blog and resource URL. Tag each page by template type, author, last real update and traffic. Mark the pages that match Ray’s eight risky templates.
2. Days 31 to 60: fix and consolidate. Start with your highest-traffic and highest-intent pages. Add evidence, authors and methodology, and merge near-duplicates.
3. Days 61 to 90: remove, redirect and monitor. Retire pages that can’t be saved, track Search Console impressions and clicks, and only then restart publishing, with original data and expertise at the center.

One caution on timing. If your traffic is moving during an update rollout, resist panic edits. For core updates, Google’s advice is to wait at least a week after the rollout ends before reviewing Search Console data. Fix quality problems because they are problems, not because a chart wobbled.

What we don’t know yet

There’s another side to this, and it’s worth being straight about it.

• Correlation isn’t cause. Every dataset here relies on third-party estimates and shows patterns, not proven mechanisms. Ray says so in her own methodology notes.
• The samples are selective. The 220-site study covers sites publicly named as customers of AI content tools, and the 11-site study was chosen as a clean group of sites hit by the same drop. Neither represents every B2B blog.
• Google hasn’t confirmed the January shift, and hasn’t said what the September spam update targeted. Anyone claiming certainty is guessing.
• Self-promotional listicles can still work for some sites. Ray notes this herself. The argument is that the tactic is getting riskier, not that it fails everywhere.
• The ChatGPT and Google link is an inference. The pattern is strong in her data, but the mechanism hasn’t been confirmed, and Perplexity moved differently.
• Google’s guide is Google’s view of Google. It tells you how Google Search works. It says nothing about how other assistants pick sources.

This doesn’t necessarily mean AI-specific work is pointless. Checking how AI describes your company, correcting inaccurate descriptions and earning genuine third-party mentions are all still worth doing. The claim is narrower: don’t pay for AI visibility with tactics that put your organic rankings at risk.

BuzzIQ’s take

The cleanest way to read all of this is as a sequencing rule. Get the SEO foundation right first, then layer AI-specific work on top of it. The reverse order is where teams got hurt. A tactic that boosts AI answers for a few months but puts your organic rankings at risk is borrowing visibility from the system that feeds it.

For B2B SaaS, the stakes are higher than for most. The content that ranks and gets cited is often the same content a prospective buyer reads while building a shortlist. If that content is thin, biased or written by invented experts, you lose twice: once in the algorithm and once in front of the buyer.

Comparison content is the clearest test case. Buyers genuinely want comparisons, so the honest version, with a disclosed relationship, a real method and evidence of testing, serves them well. The shortcut version risks the whole blog. That’s the bar we’d hold our own comparison content to.

We’d also steer clear of the opposite mistake. This isn’t a case for ignoring AI search. It’s a case for building visibility on assets that survive scrutiny: original data, named expertise and pages that exist for readers.

Where this applies at BuzzIQ. Organic and AI visibility work (content, SEO, AEO and GEO) is where this sits. A content audit using the keep, fix, consolidate and remove framework above is a sensible starting point, because it shows which existing pages are assets and which are liabilities before any new content is commissioned.

Sources

• Lily Ray: Is Google finally cracking down on self-promotional listicles? (Search Engine Journal, Feb 5, 2026)
• Lily Ray: Are citations in AI search affected by Google organic visibility changes? (Substack, Feb 17, 2026)
• Lily Ray: It works until it doesn’t, AI content strategies that backfire (Search Engine Journal, May 18, 2026)
• Google Search Central: Optimizing your website for generative AI features on Google Search (last updated Jul 10, 2026)
• Search Engine Land: Google warns against using fake author information (Oct 6, 2026)
• Search Engine Journal: Google completes September 2026 spam update after nearly 14 days (Oct 8, 2026)

Counts of self-promotional listicles and visibility percentages are Lily Ray’s own, from third-party tools (Sistrix, Ahrefs), and have not been independently checked. The Diib and SEOmonitor spam-update figures are as reported by Search Engine Journal.