GEO is damaging the very SEO foundations that make AI visibility work

I know, I know, another GEO piece...but when a friend shared this piece by Lily Ray, I just couldn’t help myself. She published a brilliant piece arguing that many of the tactics being sold as Generative Engine Optimisation are actually damaging the SEO foundations that make AI visibility work in the first place. The evidence is hard to ignore.

But reading it set off something I’ve been uncomfortable about since GEO started showing up in every conference agenda and LinkedIn thread.

What is currently being sold as GEO often looks less like a distinct new discipline and more like repackaged SEO, PR, and brand authority work, wrapped in overconfident measurement claims. And I worry that communications professionals are being encouraged to treat that as a settled fact before the evidence really supports it.

The bit that keeps nagging me

ChatGPT launched in November 2022. The term “Generative Engine Optimisation” was only formally coined later, when a Princeton research team published it at the KDD conference in 2024. That left at least eighteen months in which AI tools were already pulling in content, citing sources, and surfacing brands in their answers. Without anyone having a GEO strategy. Without anyone optimizing for AI citations. Without anyone hiring a GEO consultancy.

And yet the citations happened anyway. That doesn’t mean optimization is pointless. People were being found on Google before SEO became a formal discipline too.

But it does suggest the foundations were in place before anyone gave the phenomenon a new acronym. At the very least, that should make us more cautious about claims that an entirely new strategic category suddenly appeared.

Profound analyzed 680 million AI citations and found Wikipedia accounts for nearly half of ChatGPT’s top-10 cited sources. Reddit, Forbes, and TechRadar were also prominent. Ahrefs studied 9.6 million queries and landed on a strikingly similar list: Reddit, Wikipedia, Amazon, Forbes, Business Insider. Different studies. Similar names.

Every time one of these studies drops, the same conclusion tends to follow: these are the platforms AI trusts, so you need a strategy for each of them. But that skips over the more important question. Why does AI cite these sites so often? Has it developed a wholly new way of deciding what is authoritative? Or is it often relying on the same broad authority signals that have shaped traditional search visibility for years?

That second explanation looks more plausible to me.

I think a lot of this is traditional search authority being repurposed

Let me be clear. I’m not a search engineer. I’m a communications strategist reading the evidence from my lane. But even from where I sit, the pattern looks fairly consistent, and a lot of the people closest to this space seem to be describing the same underlying dynamic.

When AI search products need current information, they often use some form of retrieval rather than relying only on model memory. In practice, that means pulling in material from external sources and assembling an answer from what is returned.

The exact plumbing varies. ChatGPT search is not Google AI Overviews. Perplexity is not Claude. Different products use different mixes of indexes, providers, partnerships, and retrieval methods. But in many current AI search experiences, visibility still appears to be shaped by the same kinds of authority signals that have long determined who gets surfaced in traditional search.

That distinction matters.

Once sources are retrieved, the model does something genuinely different from a search engine. It summarizes. It selects. It synthesizes. It chooses what to quote and what to ignore. That part is new.

What looks less new is the upstream filtering. Models may ingest vast amounts of information during training, but in live retrieval experiences, they often end up favoring content that already performs well against traditional search authority signals. Put more simply: the model may decide what to emphasize, but the retrieval layer still plays a major role in determining what gets seen in the first place. LLMs have not reinvented PageRank; they merely borrow what others have done.

Lily Ray’s article pulls together a good amount of evidence pointing in that direction. Other researchers have reached similar conclusions through controlled tests, shopping analysis, and citation studies. Search Engine Land also reported that OpenAI had been using Google results via SerpAPI before Google sued SerpAPI in late 2025. None of that proves a single universal mechanism for every AI product. But it does point to a broader reality: AI visibility is not emerging in a vacuum.

A particularly useful example came in September 2025. Google removed the num=100 search parameter. According to Semrush’s research, Reddit’s appearance in ChatGPT responses dropped from roughly 60% to around 10%. That is not, in itself, definitive proof of how every AI retrieval system works. But it is a strong signal that changes in search-engine mechanics can materially reshape what downstream AI systems surface.

And that is the part I think many people are skipping past.

When studies show that AI frequently cites Wikipedia, Reddit, Forbes, or Business Insider, that does not automatically mean AI has invented a brand-new authority model. In many cases, it may simply mean those sources already perform strongly in the systems AI retrieval is drawing from.

That is a much less glamorous conclusion. It is also, to me, the more convincing one.

Why this matters for our credibility

This is where it stops being a technical debate and becomes a communications issue.

I understand the appeal. AI search is growing. Every CMO wants to know how their brand shows up in ChatGPT because they’ve seen a LinkedIn post from someone claiming they’ve found new oil. And along comes a shiny new specialism with tools, dashboards, frameworks, and immediate actions. Of course people are leaning in. In communications, the fear of missing the next shift is real.

But following the loudest voices has never been a strategy.

Right now, many of the loudest voices on GEO are also the ones selling it. And a lot of what is being sold treats each AI engine as if it requires its own fully distinct optimization framework, complete with bespoke citation mechanics and measurement systems. Some of that work may be useful and it’s very clear that nothing is set in concrete. Debate is needed and crucial, but much of it looks suspiciously like sensible SEO, digital PR, and content strategy with a fresh label on top.

This is not to say everything labeled GEO is nonsense. Structuring content for machine readability, monitoring citation patterns, testing prompt variability, and making sure your data is clean enough for a model to extract facts from it are all sensible practices. My issue is not with those tactics. My issue is inflating them into a wholly separate strategic discipline before the evidence supports it.

Google’s Danny Sullivan told an audience at WordCamp US in August 2025:

Good SEO is good GEO.

His colleague, Gary Illyes made a similar point at a Search Central event a few weeks earlier:

...to appear in AI Overviews, all you need is normal SEO practices. You don’t need GEO, LLMO or anything else.

Google has a long history of simplifying its own advice, so I would not treat either quote as the final word on anything. But directionally, the message is hard to miss.

Rand Fishkin’s research with Datos, published earlier this year, found Google still handles 73.7% of all desktop searches across 41 major platforms. Even if you count prompts across major LLM products as search-like behavior, traditional search remains vastly larger. So if AI visibility is often downstream of search authority, then treating GEO as though it has cleanly broken away from SEO starts to look premature.

The bit that should genuinely worry us

I probably would not be writing this if GEO were merely unnecessary. A harmless rebrand. A bit of wasted money. Annoying, but not especially dangerous. What concerns me is that some of what is being sold under the GEO banner may be actively counterproductive.

Lily Ray’s article clearly sets this out. Some GEO agencies are recommending high-volume content refreshes, synthetic FAQ generation, and scaled tactics designed to trigger AI citations. The problem is that the same tactics can weaken the organic visibility that those citations often depend on. If your authority in search declines, your visibility in AI systems may decline with it.

That is not just a traffic risk. For a communications leader who has convinced a CMO to invest in this, it is a credibility risk. And those are usually much harder to recover from than a rankings dip.

So what would I actually do?

Right. Enough hole-poking. If you’re a comms person trying to work with this reality without getting burned by it, here is where I think the evidence points.

Embed with whoever owns SEO. It may not sound glamorous, but it is the most sensible place to start. Seer Interactive found a strong connection between brands that rank on page one of Google and those that appear in AI outputs. That does not mean the relationship is perfectly linear. It does mean comms teams should be spending less time looking for a GEO silver bullet and more time aligning with SEO on priority entities, themes, and proof points.

Lean into brand mentions, but make them citable. Ahrefs studied 75,000 brands and found branded web mentions matter far more for AI visibility than backlinks do. Hallam found something similar (3x more influential in fact). That should sound familiar to any decent PR team. The point is not that comms work has suddenly changed species. It is that coverage now needs to function not only as persuasion for humans, but as a clear factual source that a machine can retrieve.

Focus on depth, not the platform container. The platform obsession is one of the biggest GEO traps.

Reddit threads do not get cited simply because they are on Reddit. They get cited because the back-and-forth often contains nuance, lived experience, contradiction, and specificity.

The same is true of LinkedIn and other social platforms. That is useful material for retrieval and synthesis. The smarter question is not “do we need a Reddit strategy?” It is “are we producing anything, anywhere, that contains the kind of depth a machine would find worth citing?”

Monitor AI visibility, but be careful about ranking claims. Rand Fishkin and Patrick O’Donnell ran nearly 3,000 prompts across major AI engines.

Fewer than 1 in 100 runs returned the same list of brand recommendations. Fewer than 1 in 1,000 returned them in the same order.

That should make everyone deeply skeptical of tools claiming to give you a stable “ranking position in AI.” Track frequency of mention. Track patterns over time. Track whether you appear in the right kinds of prompts. But do not build a strategy around a measurement system that remains highly unstable.

One more thing. If your CMO asks about GEO, and they probably will, resist the urge to bluff. The honest answer is something like this:

“AI search is real and growing. The coverage we earn, the bylines we place, the credible contexts where our brand gets mentioned, and the factual depth of the material attached to those mentions are all likely feeding signals that influence whether AI systems cite us.”

There are monitoring tools worth exploring to understand how we’re showing up. But I would be cautious about major investment in standalone GEO tactics where the evidence is still immature and many of the underlying drivers still look like SEO, PR, and authority work by another name.”

That is not timidity. That is judgment.

Where I land

I’m not arguing that AI search does not matter. It does. And I’m certainly not saying comms professionals should ignore it.

What I am saying is that the most useful response currently looks a lot less like inventing a new religion and a lot more like doing the fundamentals properly. Build genuine authority. Create material worth citing. Earn mentions through credible work. Coordinate with the people who understand search. Use AI-visibility monitoring as a signal, not a fantasy dashboard.

If, over time, the evidence shows that GEO becomes a distinct discipline with its own stable mechanics, fine. I will happily update my view.

But based on the evidence we have now, the smarter stance is caution, not conversion.