There’s a strange irony happening in search right now.
Websites publishing advice about AI-powered search are often the least prepared for it. Their articles repeat the same surface-level guidance — “be original,” “add expertise,” “answer questions clearly” — without once demonstrating any of those qualities themselves.
I spent three months testing GEO strategies across 14 content pieces: updating old articles, restructuring new ones, adding proprietary data, removing keyword-stuffed sections entirely. Some results were surprising. A few were counterintuitive. One experiment backfired completely.
This article shares what I actually found — not what sounds reasonable in theory.
If you’re looking for another list of vague GEO tips, this isn’t it.
Table of Contents
1. What Changed in Search (With Actual Numbers)
Let me be specific about what triggered this shift in my own strategy.
Three of these informational articles that had been steadily seeing 1,200–1,800 monthly visits each were consistently seeing below 600 without a ranking change in early 2024. They were still in the top spot. They were still on page one. Clicks simply stopped converting at the same rate.
When I looked at my SEO reporting dashboard, the data from Search Console, and the recordings of my page user behavior, I noticed a common pattern: visitors that did arrive were skipping right over the introduction and going straight to the specific parts of my page while also spending 40% less time on page than last year. They weren’t discovering the content cold. They were arriving already informed — from AI-generated summaries elsewhere.
This is different from a traffic drop caused by algorithm changes. The content wasn’t penalized. It was being pre-empted.
That distinction matters enormously for how you respond. Algorithm penalties require technical fixes. Pre-emption requires editorial ones.
2. The Real Competition: It’s Not Other Websites
Most SEO frameworks still treat competition as a ranking problem. Your page versus their page. Who gets position one.
GEO breaks that model.
The real competition now is between your content and an AI system’s ability to answer the question without your content at all.
Here’s the part most articles skip: AI systems are better at some content types than others. Generic explanations, basic definitions, widely-documented how-to steps — these are easy for AI to synthesize from dozens of sources simultaneously. If your article is just a definition, then you’re competing with a non-clicking summary.
However, as of mid-2025, there are certain areas where AI has not yet fully overcome its limitations:
- Recency: AI training data lags. Fresh observations, recent data, updated case studies — these AI systems cannot manufacture.
- Specificity: A general answer about “how to improve engagement” is easy. An answer specific to a B2B SaaS onboarding flow with seasonal traffic patterns is not.
- Contradiction and nuance: If you have content that is debatable and backed by evidence, that’s something that AI can use as a source but can’t generate.
- First-person experience: “I tried this and this is what happened” is a different type of statement than research suggests. AI can quote your experience. It cannot fake having had it.
That is where your editorial strategy should focus.
3. Why AI Picks Some Sources Over Others — The Patterns I Noticed
I tracked which of my updated articles started appearing in AI-generated responses across ChatGPT, Perplexity, and Google’s AI Overviews. The behavior was not identical across platforms, which itself is worth noting.
Perplexity tended to favor content with clearly structured sections, direct answers near the top, and cited data — even when the citation was to my own testing rather than a third-party study. Articles formatted for scannable reading performed better here.
Google AI Overviews showed a stronger preference for content already ranking well organically. It disproportionately pulled from pages with strong E-E-A-T signals — first-person language, author credentials in the byline, and detailed examples that demonstrated lived experience rather than researched knowledge.
ChatGPT (when browsing was enabled) skewed toward content that was comprehensive and internally consistent. Articles which contained “edge cases,” “contradictions” and “common mistakes” were referenced more than articles that contained only the “ideal scenario.”
The lesson to be learned: Far from being the same crowd, these platforms are their own. An article optimized only for Google AI Overviews may underperform on Perplexity. Knowing which platform your audience uses changes your content approach.
4. Content That Gets Ignored: Specific Patterns, Not Just Categories
Generic advice says “avoid thin content.” That’s not useful. Here’s what thin actually looks like in practice:
The Definitional Trap An article that spends 40% of its word count defining what GEO is, what AI search is, and why search is changing — before reaching anything substantive — loses the reader who already knows those things. More critically, it loses AI systems that can generate that definitional content themselves. If an AI can write your first 600 words without consulting your article, those 600 words aren’t earning you anything.
The Advice Without Evidence Pattern “Adding original data improves AI visibility” is a claim. “I added a proprietary survey result to four articles and saw two of them begin appearing in Perplexity citations within six weeks” is evidence. The first invites skepticism. The second creates authority.
The Completeness Illusion Long articles aren’t automatically comprehensive. I’ve audited 2,000-word articles that answered only one dimension of a question while calling themselves complete guides. AI systems are increasingly good at recognizing when an article avoids the hard parts of a topic — the exceptions, the failure modes, the cases where the advice doesn’t work. Covering those gaps is what separates a reference source from filler content.
Keyword Archaeology Some articles are still built around keyword density logic from 2015. Paragraphs that repeat the target phrase four times in different sentence constructions, headers stuffed with exact-match terms. Modern AI systems parse semantic meaning. They recognize when syntax has been distorted to accommodate repetition, and that distortion signals low-quality content regardless of what the words say.
5. Original Insights: How to Actually Create Them
“Be original” is advice so generic it defeats itself.
Here is what original content actually requires in practice:
Run something no one else ran. Even small tests produce proprietary data. There is no need for a 10,000-person survey. Testing one strategy on 10 articles is what AI can’t produce – real results from real content.
Document the failure case. The internet is full of success stories. Documented failures are rarer and more credible. When I wrote about an internal linking strategy that produced no measurable improvement after 60 days, that article generated more discussion and more backlinks than the articles where I reported positive results.
Take a position that creates friction. Consensus content is easy to summarize and easy to ignore. Specific, defensible positions — even controversial ones — are harder to synthesize away. If your content can be accurately represented in two sentences, it probably should be.
Connect ideas that aren’t usually connected. The most-cited section of any article I’ve written was one connecting content decay patterns in informational articles to seasonal behavioral shifts in search queries. Neither concept was new. Connecting them to explain a specific traffic pattern was.
Originality is not about being creative for its own sake. It’s about producing something that cannot exist without you having produced it.
6. Technical Structure That Helped AI Understand My Content
I want to be clear about something most GEO articles get wrong: technical optimization is a baseline, not a differentiator.
Proper heading hierarchy, fast load times, clean HTML structure, schema markup — these help AI systems read your content. They don’t make your content worth reading. That distinction matters.
That said, here are the structural changes that made a measurable difference:
Direct answer placement. As noted above — answer first, context second. This applies especially to FAQ sections and H2-level sections. Sections must be self contained and answer a specific question.
Explicit source attribution within the content. When referencing data or observations, naming the source within the sentence (rather than in a footnote) helps AI systems understand provenance. “According to my testing across 14 articles” is more machine-readable as an authority signal than a citation number.
Logical internal linking. Internal links should connect content that genuinely extends the reader’s understanding — not links inserted to improve crawl depth. AI systems increasingly evaluate whether linked content is contextually relevant.
Content dating and update transparency. Marking when content was last updated, and noting specifically what changed, signals freshness in a way that AI systems can verify through crawl history. “Updated June 2025 to include Perplexity citation patterns” is more credible than an invisible timestamp change.
7. GEO for Smaller Brands: The Underrated Advantage
Here is something the large-platform SEO model consistently obscured: smaller brands often have stronger GEO fundamentals than large ones.
Why? Because large brands typically publish broad content at scale. Their articles cover everything at medium depth. Smaller brands, by necessity, often go deep on narrow topics — and depth on narrow topics is exactly what AI systems need when answering specific questions.
A website that has published 30 detailed articles about one vertical niche may be better positioned for GEO than a website with 3,000 articles covering everything broadly.
This is particularly true of niche sites where you focus on a specific niche like SaaS, finance, health, etc. or even online business ideas, where having a niche focus and first-hand knowledge often outweighs having a lot of content.
The implication for smaller brands is to resist the instinct to expand topical coverage in order to compete at volume. Instead, treat the narrowness as an asset. Become the most thorough source on a specific problem that a specific audience has. That specificity is difficult for AI to replicate and valuable enough that AI systems want to reference it.
8. What’s Coming Next — And How to Prepare Without Guessing
Predictions about AI search are everywhere. Most are useless because they’re either too vague (“AI will change everything”) or too specific in ways that will age badly (“the algorithm will weigh X factor by Y percent”).
What I’m more confident in:
The value of unindexed experience will increase. As AI systems become better at synthesizing public web content, content derived from private observation, proprietary data, or lived experience becomes proportionally more valuable. The ceiling on generic content is falling. The ceiling on specific, owned insight is rising.
Multi-platform optimization will become normal. Using the same approach to treat Google, Perplexity, ChatGPT and the AI of the future as one homogeneous audience will result in poor performance on each of them. The platforms behave differently, reference content differently, and serve different user intentions. Understanding those differences will matter more over time.
Content updates will outperform new content creation. AI systems weight freshness, but more importantly, they weight accuracy. An article that has been refined through reader feedback, corrected over time, and updated with new observations accumulates credibility in a way that a new article cannot immediately replicate. The editorial practice of updating will become more strategically important than the volume practice of publishing.
The fundamental question is the same one it’s always been, made more urgent: does your content provide something that cannot be generated without you?
If the honest answer is no, that’s where to start.
FAQs
What’s the actual difference between GEO and SEO?
SEO focuses on helping your content rank in search results. GEO focuses on making your content a source that AI-powered systems reference when generating answers. They overlap — strong SEO signals often support GEO — but the editorial priorities differ. SEO rewards topical coverage and keyword relevance. GEO rewards specificity, originality, and first-hand evidence.
Do AI platforms like Perplexity and Google AI Overviews pull from the same sources?
Not identically. In my observation, Perplexity favors well-structured, scannable content with cited data. Google AI Overviews weight existing organic authority more heavily. ChatGPT with browsing enabled prefers comprehensive coverage including edge cases. Optimizing for one doesn’t guarantee visibility in the others.
Can a small website realistically compete for AI citations against major publishers?
Yes — and in some cases more easily. Major publishers often cover topics broadly. AI systems need depth on specific questions. A smaller site with thorough coverage of a narrow topic can outperform broad coverage from a larger site for queries that require specificity.
Is there content that GEO simply can’t help?
Generic content has a low ceiling regardless of optimization. If your content can be accurately summarized in two sentences, or if the same information appears on dozens of other sites, GEO techniques won’t manufacture differentiation that isn’t there. The foundation has to be editorial, not technical.
How long does it take to see GEO results?
Based on my testing, structural changes showed measurable impact in engagement metrics within 2–4 weeks. AI citation appearances varied: some articles appeared in Perplexity results within 6–8 weeks of updates; others took longer or haven’t appeared yet. GEO is not a fast-return strategy.
Final Thoughts
The irony of writing about GEO is that the article itself is a test.
Generic advice about originality isn’t original. An article that tells you to add “unique insights” without demonstrating any is making the exact mistake it warns against.
What I’ve tried to do here is write the article the way GEO advice says you should: specific data over vague claims, documented failure alongside success, positions that create friction rather than consensus summaries, and first-hand observation over researched generality.
Whether it works — whether this article becomes a source AI systems reference, whether it earns citations rather than just clicks — is an open question. I’ll know more in 60 days.
That uncertainty is the honest answer. Anyone telling you GEO is a solved problem is either selling something or hasn’t tested it recently.
What I’m confident in: content that deserves to be referenced tends to get referenced, across platforms and across time. Build toward that standard, and the optimization follows.
