AI Search and the End of Traditional SEO
Search Engine Optimization (SEO) is giving way to Generative Engine Optimization (GEO)
How Search Worked Before AI
To appreciate how AI search is different, it’s important to cover how traditional search engines like Google and Bing operated before the recent AI revolution. Conventional search engines are incredibly sophisticated in their own right by relying on a fundamental approach—crawling, indexing, and ranking web pages based on relevance to a user’s query and the authority of the pages.
When a user types in a query, the search engine breaks it down into keywords and attempts to match those keywords to text/metadata on the countless web pages in its index. Search algorithms look at factors like:
Keyword relevance: Does the page contain the query terms (or synonyms) in important places (title, headings, body text)?
Page authority: How many other websites link to this page and how reputable are those linking sites? (This is the essence of Google’s PageRank algorithm where backlinks are seen as votes of confidence).
Content quality and freshness: Is the content comprehensive and well-written? Is it up to date? (Google introduced numerous updates over the years, like Panda and Hummingbird, to favor higher-quality content).
User experience signals: Does the site load quickly? Is it mobile-friendly? Do users engage with it or bounce back shortly after arriving?
The result of this process is the search engine results pages (SERPs) we all know so well. Ten blue links, supported by short descriptions (plus ads and other features), ranked in order.
Users scan titles and snippets, click one, and then possibly refine their search if they don’t find a satisfactory answer. In essence, traditional search matches search terms to relevant webpage content and uses link-based authority signals to sort those webpages.
As an example, if someone searches for “best project management software,” Google looks for pages that have those words (and related terms) and ranks them partly by which pages have garnered the most inbound links (among other signals). A SaaS company in that space, say a project management platform, might invest in SEO by creating a comparison or “Top 10” article and building backlinks to it, hoping to appear near the top of those results.
But traditional search has its limitations. For one, it often struggles with understanding natural language beyond simple keywords. A query phrased as a question (“What software do project managers use for agile teams?”) might be broken into keywords (e.g., software, project managers, agile), and the engine might miss the nuanced intent. Today, Google has incorporated AI elements like RankBrain and BERT (machine-learning models for query understanding) to better grasp the meaning behind search terms. But even with these advances, the output is still a list of links for the user to investigate.
Another limitation is that search engines do not consolidate content. They only retrieve what exists on different sites. If a clear, concise answer is present on a webpage, Google might show a featured snippet at the top of the results. However, for more complex or multi-faceted questions, the search engine still just provides a list of links to click through, forcing the user to piece together information from different sources.
Before the advent of generative AI, SEO best practices revolved around making your content visible and authoritative to search engine algorithms. That included doing keyword research, optimizing on-page elements, ensuring technical crawlability, and building authority through backlinks and useful content. The playing field was well understood by SEO professionals, even as Google made hundreds of minor algorithm tweaks each year.
The Emergence of AI Search
Over the last couple of years, generative AI and Large Language Models (LLMs) have arrived in a big way. When ChatGPT debuted in late 2022, it became apparent that the web as we know it was about to change.
Unlike a traditional search engine, AI doesn’t just find relevant web pages—it generates an answer by synthesizing information from multiple sources. Let’s briefly highlight a few milestones in the AI search era.
ChatGPT and conversational search
When ChatGPT was released, users could suddenly have a dialogue with AI that would answer questions, explain concepts, and solve problems in natural language. People quickly realized they could use ChatGPT instead of a search engine for certain tasks. For example, asking for a summary of a topic or advice on how to do something.
At launch, ChatGPT didn’t have web search capabilities and its knowledge wasn’t up-to-date, but it demonstrated what was possible. The key innovation was that it could produce a direct answer in a conversational manner, often in paragraph or bullet point form, without the user needing to click any external links.
Bing’s integration of ChatGPT (Microsoft Copilot)
Microsoft moved quickly to integrate ChatGPT with Bing, launching Microsoft Copilot in early 2023 (initially named “Bing Chat”). This was a watershed moment. A traditional search engine was augmented with generative AI and could have a dialogue about search results.
When a user asks Bing a complex query (e.g., “Create a Maui trip itinerary”), Copilot compiles information from multiple web pages and presents a written answer as if a human assistant had researched and summarized the topic.
Even more, Copilot provides citations in its responses (superscript numbers that link to sources), combining AI answers with underlying web content. For example, instead of showing ten separate links for the trip itinerary query, Copilot in Bing might produce a day-by-day travel plan by pulling information from travel blogs, government tourism sites, and forums into a single answer supported by citations.
Google’s response with Gemini and Search Generative Experience (SGE)
Google, the search market leader, also developed a generative AI model (Google Bard, later rebranded as Google Gemini). In 2023, Google began testing Search Generative Experience (SGE)—an experimental version of Google Search that provides AI-generated summaries at the top of the results. With SGE, a user’s query might trigger a synthesized answer consisting of key points, with traditional results shown below.
Google SGE uses generative AI to provide concise information, address complex queries, and even assist with shopping by summarizing product details. For instance, a query like “best DSLR camera for wildlife photography under $1000” might yield an AI summary comparing a few cameras and drawing on data from different review sites—before listing the usual link results.
Google’s approach has been cautious. They clearly label that SGE results are AI-generated and note it’s experimental. But it’s still a significant step toward AI-powered results on the world’s most popular search engine.
Other AI search engines and assistants
Apart from big names, several new players have emerged purely focused on AI Q&A. Perplexity.ai is one such example. It’s an AI search engine that takes your query, searches the web, and gives a concise answer with footnoted citations for each sentence. Users have likened Perplexity to a hybrid of Google and Wikipedia—you get a direct answer but also see exactly which sources support each part of that answer.
Other examples include You.com’s AI chat and DuckDuckGo’s instant answers powered by AI. Even Amazon has rolled out an AI search assistant for shopping and numerous SaaS applications (like Notion and Slack) integrated AI helpers to search within siloed data.
In all these cases, the common thread is that AI is changing the search experience from “find me a webpage that might have the answer” to “give me the answer directly.” The AI models can interpret the nuance of queries better than before, thanks to advanced natural language processing.
To put it more succinctly, they can handle conversational or long-tail questions gracefully. Instead of a user trying different keywords to get the information they need, they can ask a question in plain English (or any language the model supports) and get a useful answer in one go. This has been described as moving from keyword search to semantic and conversational search.
It’s important to note that these AI systems are trained on vast amounts of data, which include the contents of millions of websites. So in a way, they carry an internalized version of the web’s knowledge. This is why ChatGPT could previously answer questions without live access to the internet—it “learned” from web content that existed before 2021 (as of March 2023, ChatGPT now accesses the live internet).
For SEO practitioners, the emergence of AI search represents both an opportunity and a challenge. The opportunity is that, in theory, the best content—or more broadly, the most informative, relevant, and user-friendly content—will be what AI chooses to incorporate into its answers. If your site provides exactly what the user is looking for, an AI summarizer will hopefully pull from your content and potentially give you credit with a citation or mention.
The challenge, however, is that the user might never visit your site if AI provides everything they need in the answer box.
Why Optimizing for AI Search Is Different
Optimizing for AI models requires a mindset shift. While many foundational SEO principles still apply, your tactics and priorities will differ. This is because the “audience” now includes AI algorithms that read and reinterpret your content, not just index and rank it. Let’s explore the key differences that make GEO a new ballgame.
From ranking to referencing
Until recently, your goal was to rank as high as possible on SERPs—ideally #1 for your target keywords. This was obviously because a higher ranking translated to more visibility and clicks.
In AI search, there isn’t a traditional ranking to climb. Instead, the AI might scan dozens, hundreds, or thousands of sources and then choose a few to reference (explicitly or implicitly) in its answer. Your content could be used even if it’s not the top result for a keyword, as long as AI finds it relevant and trustworthy.
For example, a well-written post on a niche blog might be summarized by AI for a specific query, even if that blog post would never be #1 on Google due to a lack of backlinks or a low domain authority. This means being one of the sources the AI “likes” is the new win, even if you’re not at the top of traditional results.
Contextual and semantic understanding vs. keywords
AI models operate on semantic understanding. They don’t look for exact keyword matches the way a conventional search algorithm might. Instead, they’re considering the context and meaning.
Because of this, optimizing content for AI is less about repeating a keyword phrase and more about comprehensively answering the implied question. If a user asks, “How can I improve my website’s SEO for voice search?” a traditional approach might be to have a page optimized for “improve SEO for voice search” and related terms.
A GEO-oriented approach is to ensure your content clearly answers that question and provides context (e.g., covering why voice search is different, listing specific tactics for voice search optimization). The AI will parse the content deeply—possibly looking at entire paragraphs or sections to see if they contain the answer. It’s akin to writing for a very attentive reader who will consume every bit of what you publish.
Multi-source answers
Traditional search ranks individual pages. But AI search often combines multiple sources. A single AI-generated answer to a complex query might draw from three different websites—one for a definition, one for a statistic, and another for an example or anecdote.
As an optimizer, you might not win with just one perfect page that covers everything. Instead, having specific pieces of content that answer specific sub-questions can make you part of the answer set.
Let’s take the question, “What are the benefits and drawbacks of microservices architecture for a SaaS product?” AI might take the “benefits” from one blog (perhaps your company’s engineering blog) and “drawbacks” from another source (maybe one of your competitor’s blog posts).
Again, traditional SEO factors alone won’t guarantee AI picks your content. It will pick whatever snippet best answers the user’s question. In fact, AI models sometimes pull text from sites with lower authority if they directly answer the query. However, they tend to lean toward high-authority sites for very sensitive queries.
Less emphasis on backlinks
Backlinks have long been the foundation of SEO authority. While they remain important for traditional search (and indirectly help AI find your content via search indexes), AI-generated answers don’t care about backlinks per se. Ultimately, the LLM isn’t tallying links when formulating an answer. It’s focusing on content quality and relevance.
A page with zero backlinks that contains a clear, well-structured explanation could be favored over a top-ranked page that has more fluff. For example, Perplexity AI will readily quote a lesser-known blog if it has a pertinent sentence, whereas Google might have buried that blog on page three of results due to low authority.
This isn’t to say authority is meaningless. AI systems trained on the open web are likely influenced by reputable sources. Even more, Microsoft Copilot and Google Gemini use their search indexes as a starting point, both of which weigh backlinks.
However, once the AI is reading the content, a golden nugget of information on a small site can take precedence over a generic statement on a big site. One study noted that affiliate-heavy content (often produced by niche sites that Google tends to downrank) can still surface in AI answers on platforms like Perplexity. This is because AI is evaluating only the content and doesn’t care about the affiliate links.
The role of structured data and metadata
In traditional SEO, structured data or schema markup helps you get rich snippets or appear in specific search result features. In GEO, structured data plays a slightly different but still important role.
AI doesn't directly use schema to rank content but it can feed knowledge graphs and provide context. For example, Google Gemini and Microsoft Copilot have been known to leverage structured data for certain facts.
Google SGE uses structured data to identify key product attributes like brand, price, and reviews to ensure AI summaries of products are accurate.
Microsoft Copilot uses Schema.org markup to enhance its knowledge with reliable facts about companies or people.
What this means for optimization is that including structured data on your site (for products, FAQs, how-tos, etc.) can make it easier for AI systems to pull precise information.
Let’s say a SaaS company offers different price points. AI could directly extract “starting price: $X/month” to answer a user’s question about cost, rather than scraping outdated or inaccurate text. We’ll delve into this more in a later article but the key takeaway is that structured data supports AI’s contextual understanding.
AI’s emphasis on factual accuracy and consensus
Generative AI has a well-known issue with “hallucination,” meaning it makes up answers that sound plausible but are false. To mitigate this, AI search often prefers content that is factual, specific, and corroborated by multiple sources.
For GEO, this underscores the importance of accuracy. If your content contains a dubious claim or an outdated statistic, AI might omit it. But if your content aligns with what other high-quality sources are saying on the topic, AI is more likely to trust and use it. Being seen as a credible source through consistent content, expert authorship, and brand reputation matters.
This is challenging as “accuracy” is not a quantifiable metric like Moz Domain Authority but rather determined by AI on a case-by-case basis. That being said, content that reads as knowledgeable and unbiased tends to be favored.
User interaction and follow-up queries
Another difference is the interactive nature of AI search. Users can ask AI follow-up questions, whereas with Google, they’d have to type an entirely new query. This means content may be surfaced in a multi-turn context.
For instance, a user might ask AI, “What’s a good CRM for a mid-sized SaaS company?” The AI gives an answer, citing a few options, including your company. The user then asks, “How does [Your Company’s CRM] integrate with marketing automation?” If you have content (say a support article) that provides that information, AI can pull it in.
Optimizing for AI requires anticipating not just the first question but also related follow-up questions. It’s similar to topic clustering when creating a comprehensive content strategy. By providing content that covers various angles of a topic, your site and brand will keep showing up as the user drills deeper.
All in all, optimizing for AI search is different because the target is no longer an algorithmic ranking system—it’s a reading, synthesizing intelligence. You have to consider what that intelligence values: clear answers, relevant context, credible sources, and structured information it can easily parse.