SEO vs. AI Search: The 101 Questions Every Marketer Needs Answered

I've been in SEO for over a decade. I lived and breathed it. I had the playbook down: keyword research, link building, technical audits... the whole nine yards. I thought I'd seen every update Google could possibly throw at us. Then ChatGPT happened. Then Perplexity. Then Google's AI Overviews. And just like that, my entire rulebook felt like a history textbook.

It's not just a feeling. Last month, I did an experiment. I spent three hours—three hours—plugging the exact same query into ChatGPT, five times in a row. I got five completely different sets of sources. Not just a different order. Different. Sources. That's when the panic really set in. This isn't just another algorithm update. This is a whole new ballpark. We're not playing the same sport anymore.

So I did what any slightly-panicked-but-trying-to-be-productive person does. I started making a list. Every time I had a "wait, that makes no sense" moment. Every "but what about..." that hit me in the shower. Every time I felt like I was flying completely blind. Before I knew it, that list wasn't so little. It was 101 questions long.

I'm putting this list out here because I know I'm not the only one. If you're a marketer, a founder, or an SEO specialist, you're probably staring at the same abyss, feeling the same pit in your stomach. Maybe you'll find an answer here. More likely, you'll just feel a little less like you're going crazy. Either way, here they are—the 101 questions that are keeping me up at night.

Understanding How AI Search Works

We have to start at square one. Because, let's be honest, I'm not convinced anyone really knows what "square one" is anymore.

Ranking and Retrieval Mechanisms

For years, my brain was wired for Google. CTR matters. Backlinks matter. We knew the levers. But does OpenAI care about CTR? I seriously doubt it. I've personally seen sites with awful engagement get cited over and over in ChatGPT. So, are they blind to it, or is it just a completely different equation? Which is it?

Do AI search engines use click-through rates (CTR) to rank citations, similar to Google?
How do AI models retrieve and rank pages differently from traditional search engines?
What's the difference between retrieval-rerank-generate and Google's crawl-index-serve process?
How do cross-encoder rerankers evaluate query-document pairs compared to PageRank?
How many reranking layers occur before an AI model selects its final citations?
Which part of the AI system actually chooses the final citations shown to users?
Is reranking deterministic or stochastic (random) in AI systems?
Why can a site with zero backlinks outrank authority sites in LLM responses?

That last one is what really gets me. I've watched a brand-new, no-authority site get a top citation in Perplexity while a household-name brand with thousands of backlinks was nowhere to be found. In the old world, that's impossible. In this one, it's just... Tuesday.

Content Processing and Understanding

This one keeps me guessing. When an AI crawler hits my site, what is it actually looking at? Googlebot got smart; it started to understand layout, UX signals, the whole visual experience. But these AI models? Are they just scraping text? Are they blind to everything we've learned about good design?

Do AI systems read page layout the way Google does, or only extract text content?
Can AI systems read and evaluate images in webpages, or only the surrounding text?
Should we write shorter paragraphs to help AI chunk and process content better?
How do vector embeddings determine semantic distance compared to keyword matching?
How does the semantic relevance between content and a prompt affect ranking?
What makes a passage "high-confidence" during AI reranking?
Can two very similar pages compete within the same embedding cluster?

I've been playing around with this. I've noticed shorter paragraphs seem to get pulled as citations more often. But is that a real signal, or am I just seeing patterns in the clouds? That's the maddening part. We're all back to just... guessing. It feels like 2005 all over again, but infinitely more complex.

Model Architecture and Technical Details

Look, I'll be the first to admit I'm no data scientist. I'm not building neural networks in my garage. But I've been forcing myself to read the whitepapers and the technical blogs because you can't win the game if you don't even know the rules. And right now, the rules are written in a language I barely understand.

Do we need to understand the "temperature" value in LLMs for SEO purposes?
How does temperature=0.7 create non-reproducible rankings in AI responses?
Why is a tokenizer important for understanding how AI processes content?
How do token limits create boundaries that don't exist in traditional search?
Are Google and LLMs using the same embedding model? If so, what's the corpus difference?
How does Knowledge Graph entity recognition differ from LLM token embeddings?

That "temperature" setting is a perfect example. It's designed to be random. To give different answers. That's a feature, not a bug. But for a marketer who needs repeatable, trackable results? It's pure chaos. How in the world do you optimize for random?

Citation and Visibility Mechanics

Okay, deep breath. This is the part that really makes me want to throw my laptop. In this new world, citations are the new "rank #1." They are everything. And they are also completely, utterly unpredictable.

How Citations Work

I'm not exaggerating. I tested this just yesterday. I asked ChatGPT the exact same question at 9 AM, 2 PM, and 8 PM. I got three totally different sets of citations. Same day, same user, same question. Different reality. Why?

Why are citations continuously changing in AI responses?
Why does the answer structure change even when asking the same question within a day?
How do AI models decide when to search again mid-answer?
Why do we see multiple automatic LLM searches during a single chat window?
Can a single sentence from a blog post be quoted by an AI model?
Is there a way to track when our content is quoted but not linked?
Why do LLMs sometimes fabricate citations while Google only links to existing URLs?
Why doesn't Google show 404 links in results, but LLMs include them in answers?

The hallucinated citations are terrifying from a tracking perspective. I've seen ChatGPT literally make up a URL. That would be a major bug in Google. In AI? It's just a Tuesday. How do we track our visibility when the "links" don't even exist?

Brand Visibility and Mentions

This is the question that haunts my client calls. How do I prove this is working? In the old world, I had rank trackers, Google Analytics, conversion funnels. With AI search, I feel like I'm flying blind and just hoping for the best.

Is there a way to track how many times our brand is mentioned in AI answers?
Does being cited once make it more likely for our brand to be cited again?
Can frequent citations raise a domain's retrieval priority automatically?
How can we determine if AI tools cite us following a change in our content?
Can we track which prompts or topics bring us more citations and what's the volume?
Why do some LLMs cite us while others completely ignore us?
How can a small website appear inside ChatGPT or Perplexity answers?

I've seen tiny, brand-new blogs get cited constantly. Meanwhile, established industry leaders are invisible. There's a pattern, there has to be, but I can't nail it down. Not yet.

Signals and Ranking Factors

This is where my old SEO brain really short-circuits. We knew the signals. Backlinks. CTR. Dwell time. E-A-T. But in the AI-verse? It feels like all bets are off.

User Behavior Signals

Does user behavior even exist for these models? Google tracks every scroll, every click, every second you spend on a page. But an LLM? It serves an answer. Does it know or care what you do next?

Can scroll depth or mouse movement affect AI ranking signals?
How do low bounce rates impact our chances of being cited by AI?
Does post-click dwell time on our site improve future inclusion in AI responses?
Can AI models use session patterns (like reading order) to rerank pages?
Does session memory bias citations toward earlier sources in a conversation?
Are user clicks on cited links stored as part of feedback signals?
Does past click behavior influence future LLM recommendations?

I've been running tests, trying to see if I can "teach" the AI that my site is good by clicking the citations. So far, the results are... inconclusive. But that doesn't mean it's not working. Or maybe it just means I'm wasting my time.

Content and Freshness Signals

Freshness is another-level-of-weird. Sometimes I'll publish a new post and it gets cited within an hour. Other times, weeks go by and it's ignored, while the AI cites a competitor's two-year-old (and outdated) post. What is the pattern?!

Does AI favor fresh pages over stable, older sources?
Is the freshness signal sitewide or page-level for LLMs?
Does freshness outrank trust when signals conflict in AI systems?
How often do AI systems refresh their understanding of our site?
Do AI systems also have search algorithm updates like Google?
How does knowledge cutoff create blind spots that real-time crawling doesn't have?

The "knowledge cutoff" thing is a trip. ChatGPT will tell you it "doesn't know anything after 2023," but then it'll do a web search and cite something from this morning. So which one is it? And which one matters more for my content strategy?

Trust and Authority Signals

Ah, E-E-A-T. Our old friend. But here's the kicker: I've seen sites with zero author bios, no 'About Us' page, and a design from 1998 get cited as an authority. Meanwhile, meticulously-crafted expert content gets ignored. What is going on?

Can AI build a trust score for our domain over time?
Can a heavily cited paragraph lift the rest of the site's trust score?
Why is EEAT (Experience, Expertise, Authoritativeness, Trustworthiness) easily manipulated in LLMs but not in Google's traditional search?
Do model updates reset past re-ranking preferences, or do they retain some memory?
How can we make sure LLMs assert our facts as facts?

My gut feeling? I think these models are looking for confidence, not authority in the human sense. They're looking for clear, declarative statements. But how do you prove your fact is better than someone else's? That's the billion-dollar question.

Platform Differences

Just to make life more fun, it's not one new system. It's a dozen. ChatGPT acts differently than Perplexity. Perplexity is different from Google's AI Overviews. And they all work differently than classic search. My job just went from optimizing for one (mostly) stable system to 3-4 chaotic ones. It's exhausting.

Comparing AI Platforms

I run the same queries across all platforms. The results are a total mess. Sometimes we're cited on ChatGPT but invisible on Perplexity. Sometimes it's the reverse. There's no consistency.

Are ChatGPT and Perplexity using the same web data sources?
Do OpenAI and Anthropic rank trust and freshness the same way?
Will Google AI Overviews and ChatGPT web answers use the same signals?
Are per-source limits (max citations per answer) different across LLMs?
Is OpenAI also using CTR for citation rankings?
Are LLMs using the same reranking process as each other?

AI vs. Traditional Search

Here's the really weird part: sometimes the "old way" is just plain better. I'll search for something on Google, get a perfect 10 blue links. Then I'll ask ChatGPT the same thing, and it will confidently hallucinate a wrong answer or cite sources that are just plain bad. Why?

Why can we find better results with 10 blue links without hallucination (mostly)?
Why do some pages show up in Perplexity or ChatGPT, but not in Google?
Are Google and LLMs using the same deduplication process?
How do retrieval and reasoning jointly decide which citation deserves attribution?
Why do LLMs retrieve 38-65 results per search while Google indexes billions?

That last point is blowing my mind. Google is sorting through billions of pages. These AI systems seem to be picking from a tiny pool of 50-60 results. How do they choose that first pool? And why do they so often seem to pick the wrong ones?

Optimization Strategies

So, we get to the big one. The "what do we do?" question. If everything I knew about SEO is on shaky ground, how do I actually optimize for this new world?

Content Optimization

I've been in the trenches, testing. Writing content specifically for AI. Using different structures, different tones. Some of it works. A lot of it doesn't. But I'm starting to see... something.

How can we ensure that AI understands what our company does?
What happens if we optimize our entire website solely for LLMs?
How do you optimize a web/product page for a probabilistic system?
Can we train LLMs to remember our brand voice in their answers?
Does linking a video to the same topic page strengthen multi-format grounding?
Do internal links help strengthen a page's ranking signals for AI?
Are internal links making it easier for AI bots to move through our sites?

I'm pretty sure internal links still matter. A lot. I've seen it work. Pages with a strong internal linking structure seem to get cited more. But again... is that causation? Or am I just seeing what I want to see?

Technical Optimization

Do schema changes result in measurable differences in AI mentions?
Can we use Cloudflare logs to see if AI bots are visiting our site?
Does OpenAI allocate a crawl budget for websites?
How can we track whether AI tools use our content?

Local and Industry-Specific

How can we make a local business with a map result more visible in LLMs?
Why do we still see very outdated information in some languages, even when asking current questions?
Can form submissions or downloads act as quality signals for AI?

Tracking and Measurement

Visibility Tracking

What's the easiest way to track prompt-level visibility over time?
How can we know which prompts or topics bring us more citations?
Can citation velocity (growth speed) be measured like link velocity in SEO?
Do we track drift after model updates?
Which pages are most requested by LLMs and most visited by humans?

Testing and Analysis

Should we run multiple tests to see the variance in AI responses?
Can we use long-form questions with the "blue links" on Google to find the exact answer?
Is web_search a switch or a chance to trigger in AI systems?

Strategic Questions

Business Impact

Are we chasing ranks or citations in the AI search era?
What would happen if we renamed monthly client SEO reports to "AI Visibility AEO/GEO Report"?
How many of us drove at least 10x traffic increases after Google's algorithm leak?
Can the same question suggest different brands to different users?

Long-Term Strategy

Will LLMs eventually build a permanent "citation graph" like Google's link graph?
Do LLMs connect brands that appear in similar topics or question clusters?
How long does it take for repeated exposure to become persistent brand memory in LLMs?
Will LLMs remember previous interactions with our brand?
Will AI agents remember our brand after their first visit?

Recovery and Adaptation

Do LLMs retraining cycles give us a reset chance after losing visibility?
How do we build a recovery plan when AI models misinterpret information about us?
How can a new brand be included in offline training data and become visible?

Advanced Technical Questions

Model Behavior

Do human feedback loops change how LLMs rank sources over time?
Why are LLMs more biased than Google in their responses?
Does offering a downloadable dataset make a claim more citeable?
Why do we need to be visible in query fanouts? For multiple queries at the same time?
Why is there synthetic answer generation by AI models even when users are only asking a question?

System Architecture

How does crawl-index-serve differ from retrieve-rerank-generate?
Is there any way to make AI summaries link directly to our pages?
How does AI re-rank pages once it has already fetched them?

So What Now?

So... what now? If you came here looking for a neat "10-step guide to AEO," I'm sorry to disappoint you. I don't have all the answers. I probably don't even have most of them.

Honestly, I'm not sure anyone has all the answers right now, no matter what they're shouting on social media. But I'm asking the questions. And that's the first step.

Here's the one thing I know for sure: This is not just "SEO 2.0." This isn't a new-and-improved version of the old game. It's a brand new one. It has different rules, a different playing field, and a different way to score. Everything has changed.

The companies and marketers who realize this today are the ones who will win. The ones who keep trying to jam their old SEO tactics into this new machine are going to get left behind. Period.

But here's the good news. Or maybe it's just comforting news. We are all in the same boat. There is no playbook. There are no "best practices" yet. We are all explorers in this new, weird territory. We're all just testing, failing, learning, and testing again.

So if you're reading this and you've found an answer to any of these questions—or if you have 10 new ones to add—please reach out. Seriously. We have to figure this out together.

The future of search is here. It's messy. It's confusing. It's unpredictable. But... I have to admit, it's also the most exciting (and terrifying) thing to happen to our industry in a decade. Right?