I Called GEO a Scam. Here's What I Got Wrong (and What I Didn't).

A while back I called GEO a scam on LinkedIn.

I've changed my mind about half of that.

What changed is my respect for the discipline. Buyers now use AI assistants to research what they buy, and a brand that doesn't show up in those answers is missing from conversations it used to be part of. Getting a brand into those answers takes serious work, and it deserves a real strategy.

What didn't change is how I feel about the hacks.

If you're reverse-engineering formats and phrases to match whatever ChatGPT favors this month, throwing up llms.txt files, duplicating your content as markdown, publishing hundreds of "best," "top," and "X vs. Y" pages because they showed up in a citation study, or adding "official" to every title because ChatGPT started searching for it, your marketing's running on snake oil.

Sorry not sorry.

1 model update, and the playbook was gone

In August 2026, a ChatGPT update changed which kinds of pages it cites. Peec AI data shared by Tomek Rudzki and Lily Ray showed:

• Listicles fell from 15.77% of citations to 7.80%
• Comparison pages fell from 9.08% to 6.17%
• Searches for "official" pages and "site:" searches both went up

Think about the agencies that spent the first half of the year cranking out listicles for their clients. One update cut the value of that work in half.

These shifts don't stop. Retrieval changes every few months, and if your strategy depends on keeping up with each one, you'll be rebuilding it every few months too.

The SEO industry has been playing that cat-and-mouse game with Google for 20 years. I've been in it for 16 of them, and the house usually wins.

A comparison page is fine when a buyer with intent is looking for one. We build them for clients. It becomes a hack when you publish 400 of them because a citation study told you to.

So what holds up when the formats keep moving?

What other people say about you

Google figured out a long time ago that what other sites say about you tells it more than what you say about yourself. That's what links were always for.

AI assistants took that idea a lot further. ChatGPT, Google's AI Mode, and Perplexity pull from news coverage, reviews, Reddit threads, YouTube videos, industry publications, analyst commentary, and comparison articles, and none of those sources need to link to you to shape the answer.

A few things shift when that happens:

• Owned claims → corroborated claims. Calling yourself the best doesn't count for much. Credible independent sources saying it counts for a lot.
• Backlinks → mentions and context. A Reddit thread or a podcast can shape how AI describes you without ever linking to your site.
• Rankings → representation. The better question to ask is how ChatGPT describes your brand when someone asks about your category.
• Keywords → associations. If the web keeps connecting your brand with a concept, like affordability or great service, the models pick that up.
• Your domain → everyone else's. More of your credibility gets decided on sites you don't own.

The research lines up with this. Ahrefs looked at 75,000 brands and found branded web mentions correlated with AI visibility at roughly 0.66 to 0.71 across ChatGPT, AI Mode, and AI Overviews. YouTube mentions correlated even more strongly, at about 0.74, the highest of any factor they tested.

That's a correlation, so it doesn't prove cause. But it points the same direction as everything else here.

Andrew Holland, Director of SEO at JBH, makes the case that GEO works a lot like brand marketing. As he puts it, "Brands with a strong marketing presence are naturally referenced in relevant buying contexts."

I agree.

Chunking copy or chasing citation formats might change how a page gets retrieved. It does nothing to give the rest of the web a reason to vouch for you, and that vouching seems to be what AI leans on when it decides who to recommend.

Brand wins ties. Proof breaks them.

One study from this summer makes the case well. Xi Chu (Trine University) and Yupeng Hou (Texas A&M) showed GPT-4o-mini, Claude Sonnet, and Gemini 3 Flash skincare products with identical specs, some from established brands and one from a brand they made up.

The established brands got 100% of the recommendations.

That sounds bleak if you're the challenger. But when they gave the made-up brand a rating just 0.075 stars higher, it won 64% to 80% of the time.

They also tested the kind of tactic GEO vendors sell, like adding authority-style language to product copy. It helped the first brand that used it. Once every brand used it, the payoff dropped from +0.802 to +0.007 by their measure, which is basically nothing.

So brand wins ties, a little real proof beats brand, and any tactic everyone can copy stops working once everyone copies it.

What it looks like with a client

One of our clients is a real estate brokerage. When we started, their blog brought in zero leads.

In under a year, that blog went from 0 to 51 leads a month, and 29 of them converted in July 2026 alone. The page that ended up doing the most work explained a program they already offered.

We got there by writing useful pages for people about to make one of the biggest financial decisions of their lives.

That would've worked in 2015, too.

Back to basics

My take today is that GEO is just good marketing for the AI era.

That goes double for the companies we work with. If your sales cycle is long, your product is complicated, and legal has to sign off on everything, shortcuts cost more. A model update can wipe out a hack faster than legal can approve the replacement page.

What holds up through the updates is the same list it's always been:

• Clear, differentiated positioning
• Content that proves you know what you're talking about
• Presence where your market pays attention, like YouTube, Reddit, LinkedIn, industry publications, podcasts, and review sites
• Independent reviews that back up your claims
• Original research and named ideas other people want to reference
• Content-led PR that gets that research in front of the journalists and publications your buyers read
• A technical foundation that lets search engines and AI assistants crawl and read your site

The technical piece still matters. We audit it for every client. It just won't carry a brand nobody's heard of.

Research earns its keep over time. A good content-led PR report gets a bump in the press when it launches, then keeps earning links, mentions, and citations for years, as long as the data stays relevant. Compare that to a listicle format that lost half its citations in 1 update.

Share of voice in AI answers, how the models describe you, and which topics they tie you to are all worth tracking. We track them for clients every month. But they only count if they lead to pipeline.

Nothing matters if you're not growing the business.

Fame is expensive and slow. Anyone selling you a shortcut is betting your brand on the next update.

2 things to do this week

1. Run a 10-minute check. Ask ChatGPT and Google's AI Mode to explain your category and recommend a few providers. Write down which brands come up, how they describe you (if they mention you at all), and which third-party sites they cite. Then check whether you show up on those sites. That list is your starting point.

2. Pressure-test your GEO vendor. Ask them:

• What happens to our program if ChatGPT stops citing listicles tomorrow?
• How do you connect AI visibility to pipeline and revenue?
• Where does third-party coverage come from in your plan?
• How do you track prompts when every buyer's conversation is different?

If every answer is a new format, walk. If they tell you not much changes because they never built for the format, keep talking.

A good GEO strategy gives the world something worth finding, something worth citing, something worth talking about, and a reputation worth recommending.

Which looks an awful lot like good marketing.

If you want more field notes like this from our client work, subscribe to The Narrative. And if you want us to run that check on your brand, book a free 20-minute teardown.

References

1. Lily Ray and Tomek Rudzki, ChatGPT Is Changing Which Page Types It Cites in AI Responses (New Peec AI Data), 2026
2. Louise Linehan and Xibeijia Guan, AI Brand Visibility Correlations, Ahrefs, December 2025
3. Andrew Holland, Fame Engineering: The Key to Generative Engine Optimization, Search Engine Land
4. Xi Chu and Yupeng Hou, Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems, arXiv, June 2026