TL;DR
- Ranking first on Google no longer guarantees traffic. According to Ahrefs, when an AI Overview appears, the top result gets 58% fewer clicks.
- People no longer search in three words. They describe their whole situation, and AI breaks that question into sub-questions.
- You’re not competing for a position anymore. You’re competing to be one of the few sources an AI answer cites.
- AI reads passages, not pages. Every section of your content has to make sense on its own.
- Keywords aren’t dead. Keyword stuffing is. Use your customer’s words, and add the knowledge only you have.
These are interesting times for digital marketing. Ranking number one on Google no longer means traffic. And this isn’t a forecast. It’s happening right now.
In the reports I look at every week, in Search Console and Analytics, I keep seeing the same picture. The site ranks first. Impressions are growing. But traffic is falling, clicks are gone, and sales drop with them.
The data confirms it. Ahrefs analyzed 300,000 keywords and found that when an AI Overview appears, the top-ranking page gets 58% fewer clicks. Pew Research found that people clicked a regular search result in 8% of visits when an AI summary appeared, compared with 15% when it didn’t.
The reason: we no longer compete for a position. We compete to be cited. And who gets cited follows a clear logic that has nothing to do with keywords and how often you repeat them.
If you still write “SEO copy” stuffed with keywords, this could save you years of wasted work.
From keywords to questions
Old search was fragmented. People typed one to three words, like “air conditioner repair Sofia”. The machine looked for documents where those words appeared often, close together, in the right places: the title, the URL, the first paragraph. That was search by word matching. For almost 30 years, the entire SEO industry optimized for exactly that.
Today, people don’t type three words. They type a whole paragraph.
Instead of “CRM for small business”, they write: “I have a company with five salespeople. We use Excel for our clients and it’s falling apart. Which CRM should I choose that works with local accounting software, without paying for features I won’t use, under a certain budget?”
That’s not a keyword. That’s a situation.
How AI answers a question like that
No single page can answer that question. So the machine doesn’t look for one page. It breaks the question into sub-questions and finds an answer for each of them.
Google describes this directly. AI Mode uses what it calls query fan-out: “breaking down your question into subtopics and issuing a multitude of queries simultaneously on your behalf.”
For the CRM question, the sub-questions might be: How much does a CRM cost? Which CRMs integrate with local accounting software? Which ones suit a small team? Which have an interface in the local language? What do real users say?
For each sub-question, the system pulls different sources. Then it assembles one answer.
The result: the ten blue links on the first page shrink to one answer, with three, four, sometimes eight sources cited below it. So you’re not fighting for first place anymore. You’re fighting to be one of those few sources. And most importantly, you’re competing for sub-questions you’ve probably never targeted, and that don’t exist in any keyword spreadsheet.
The old world was built on word frequency. The new world is built on meaning. The machine turns both your text and the person’s question into numbers, vectors, and measures the distance between their meanings. Not the match between letters. That’s why a text that repeats the key phrase twelve times brings no extra value today. It only makes the text worse.
AI reads passages, not pages
The model doesn’t read your site. It reads a piece of your page.
The system takes your article, cuts it into passages and organizes them by meaning. When a sub-question comes up, it pulls the passage that answers it, not the whole page.
So if your strongest answer sits in the middle of the text and starts with “As we said above…”, that passage doesn’t work on its own. Taken out of context, it makes no sense. The machine skips it.
The practical rule is simple: every subheading plus the first two or three sentences below it must make sense to someone who hasn’t read the rest of the page.
I give people who work on optimization this exercise. Imagine your content is read by someone on their phone, scrolling on the bus, distracted, barely paying attention. If they can’t skim it without getting lost, it’s not written well enough for search engines or for today’s readers.
Or, as experts put it: if a passage can’t stand on its own, it doesn’t exist for the model.
Context gives meaning to your words
The attention mechanisms in these models look at which words appear with which other words.
A page that says “GEO” with nothing around it, no definition, no example, no source, means nothing. A page where the topic is surrounded by specific statistics, numbers, names and related concepts becomes recognizable as a source on that topic. And more credible.
Your brand has to exist as an entity
These systems rely on knowledge graphs. Your brand may be well known among your customers. But for search engines and bots, you’re either a clearly defined entity, or you’re nobody.
If your company name, what you do and the people behind it are described the same way everywhere, on your site, on LinkedIn, on industry sites, in interviews, podcasts and press mentions, you become a separate, recognizable entity. The machine can be confident that all those mentions refer to the same company.
If there’s nothing out there, you’re nobody. If every place says something different about you, you’re an uncertain source. And uncertain sources don’t get cited.
I had a client that changed its name but forgot to announce the rebrand anywhere. The old and new names lived side by side, with nothing connecting them. It became a real SEO problem. It’s a quiet, easily missed reason why consistent mentions matter, and why they’re starting to matter as much as links. PR and media coverage are hugely underrated in SEO.
Three technical checks everyone skips
These can take a single afternoon to fix. But they decide whether you’re in AI answers at all.
1. Access. Check whether AI crawlers are allowed to read your site. Every AI company has its own bots, such as GPTBot, ClaudeBot and PerplexityBot, and many sites block them without knowing it. It might be a line in robots.txt, a firewall rule, or a setting someone added two years ago and forgot. If you’re blocked, nothing else in this article matters.
2. Structured data. Schema markup is the map that tells the machine what’s what: this is a product, this is a price, this is the author, this is a question and this is its answer. It won’t give you an advantage on its own, but it removes ambiguity. And ambiguity is the most common reason a system picks another source instead of you. Google’s own guidance for AI search asks that structured data match what’s visible on the page.
3. The author. A name, a face, a bio, a history, proof of expertise and experience. Anonymous content is in the weakest position right now. When the machine cites something, it needs someone to stand behind it. This matters even more for voice and wearable devices, where there’s one answer, not ten. If it isn’t you, you don’t exist in that session.
New metrics: citations instead of positions
The way we measure has changed completely. Monthly search volume isn’t enough anymore. It tells you how many people ask. It doesn’t tell you who gets chosen.
Start with a simple exercise: write down 20 real questions your customers ask before they buy. Run them in three different AI models, record everything in a table, and track whether you’re mentioned, whether you’re cited with a link, and who gets recommended instead of you. Repeat it every month. I explain how to do this properly, and the traps to avoid, in my guide to AI prompt research.
One more thing: if your traffic drops but your sales don’t, that’s not a problem. It’s the new normal. Fewer people may come, but they come better informed. Always measure in the context of revenue, not just sessions and impressions.
Keywords aren’t dead. Keyword stuffing is.
To be precise: what’s dead is keyword density, the list of phrases sorted by volume and poured into a text. Words still matter, but differently. Search systems are hybrid. There’s semantic search, and there’s classic search by words.
If people search for something using a certain term, and you call it something else, the system won’t connect you with them.
One of my clients offers a modern building method that’s new to the local market. It took me months to convince him that people search for it under a different name: passive houses. It’s not the most precise term. But it’s how people know it. When he started talking about passive houses, the algorithm connected him with people looking for that information, and he could offer them his alternative.
So when your customers describe what you do with a different but related term that meets the same need, listen.
That’s the real rule: speak your customer’s language, but write the knowledge only you have. Share that expertise and build authority around it. SEO and GEO help you get found. Knowledge is what gets you cited. One doesn’t work without the other.
Do this this week
Want to see how your brand looks in AI answers right now? Do the 20-question exercise this week. It will take about an hour, and the results may surprise you. If you need help, get in touch.
About The Author
