Summary: Search is losing ground to AI fast. If you’re not visible to AI and included in AI-generated answers, you’re invisible. Allow AI training, deliver concise, factual content, and dominate your niche. 14 actionable guidelines for AI-friendly content are your tools to preserve your place on the Internet.
As users transition to using AI as answer engines, websites cannot rely solely on search engines to drive traffic. I already wrote an article about this problem in 2023, and back then, my recommended solution was for websites to rely on permission marketing, such as email newsletters, to drive traffic. Loyal users beat drive-by users any day. (In fact, I have my own email newsletter, and you should subscribe if you don’t already do so.)
Two years later, I stand by my recommendation to emphasize permission marketing and email newsletters as much as you can. Email may be the oldest media form on the Internet, but it remains the most effective for staying in contact with loyal fans.
However, catering to established fans can’t be the only solution. How do people become fans of a brand in the first place? Through discovery. And many services realistically will never have fans — or even a strong brand — even if they are instrumental the one time a customer needs them. Such companies also need discovery to build their business.
The direct connection with customers afforded by an email newsletter will shield you from many AI disruptions, but a newsletter can’t be your only Internet strategy. (Leonardo)
Since Yahoo! changed from being a pure set of directory listings (1994) to offering search (1995), the answer to Internet discovery has been SEO: search engine optimization. After Google became the first good search engine in 1997, it quickly dominated the web. That’s how vital search was for web users.
(Disclosure: I served on Google’s advisory board when it was a startup and received stock options in return. I exercised the options and sold that stock long ago and have no current interest in the company.)
Google famously became the web’s best search engine due to the PageRank scoring mechanism that ranked search results based on how many other websites linked to each hit. (With some subtleties, such as giving more credit to links from authoritative websites than from less important sites.)
SEO isn’t dead yet, but it will soon be supplemented and later supplanted by the need for websites to be included in AI-generated answers and recommended by AI agents. (Midjourney)
From Search to AI For Resource Discovery and Understanding
We are now seeing users turn more often to simple AI answer agents or the more elaborate “Deep Research” style AI research tools available from several AI labs, often under that exact name. (Grok’s tool is called DeepSearch, but OpenAI, Google, and Perplexity all use the same name for this feature.) For example, just today, I asked OpenAI’s Deep Research to give me an overview of pizza places in my area that “reviewers who know their pizza” think serve the best-tasting pizza. Since my area comprises three towns, researching this question through traditional search would have required many searches. Instead, Deep Research “thought” for 11 minutes, during which it consulted 29 sources, and gave me two recommended restaurants for each of 4 different pizza styles, complete with quotes from the more credible reviewers. (I ordered from one of them, and the pizza was delicious.)
Users turn to AI because it goes beyond resource discovery to deliver understanding. In other words, AI doesn’t just say, “Here’s a high-ranking website that probably talks about your problem.” Instead, AI reads that site — and many more, in the case of Deep Research — and gives you its understanding of the solution to your problem. Yes, currently, AI isn’t always right, and it doesn’t consistently deliver the “deep” understanding promised by the feature name. Sometimes, AI even makes up hallucinations, though that problem is declining monthly as AI adds more reasoning steps before it delivers its answer to the user.
AI will soon be claiming an even larger share of users’ attention with the introduction of AI agents that interact with websites and other digital services on the user’s behalf.
Search and AI are the two main ways users find solutions to their problems. (Napkin)
The Importance of Getting Listed by AI
The more people use AI instead of search to discover, understand, and interact with digital services, the more critical it becomes for your company to be well-represented in the leading AI models. We’re still in the infancy of AI as the users’ interface to the Web, but it’ll grow.
Here’s the list of the top sources people used to find my website, www.uxtigers.com, during February 2025:
Google (search)
Perplexity (AI)
ChatGPT (AI)
Bing (search)
Yandex (search)
DuckDuckGo (search)
Ecosia (search)
Yahoo (search)
Baidu (search)
Poe (AI)
AI already claims two of the top three spots.
Google still tops the medal podium for driving website traffic. But the two other podium spots are now taken by AI. (Leonardo)
A simple rank ordering underestimates the overwhelming dominance of Google. The number-one finding service is still so much bigger than anybody else that AI only accounted for 4% of the total traffic arriving at my website in February 2025 from finding services (search and AI combined). Thus, search still brought 96% of the finding-initiated visits.
That said, AI-derived traffic is growing rapidly. Comparing the four weeks ending February 26, 2025, with the four weeks ending November 17, 2024, AI traffic increased by 281%, corresponding to an annualized growth rate of 12,500%. Of course, such growth rates won’t continue indefinitely, but AI traffic is guaranteed to be dramatically higher in 2026. My main point is that if you look at your own analytics data and conclude that AI is still small, you shouldn’t expect that to continue.
While websites can do well right now by simply ranking in Google (and preferably other search engines), this will change soon. Most likely by 2026, and definitely by 2027, being included in AI results will be as crucial for digital properties as a Google ranking ever was.
AI-driven traffic is growing explosively. (Leonardo)
Be Included in AI Results
The first, and maybe obvious, point is that you should allow AI to train on your data. If you deny AI access to your content, it can’t include you when presenting results to its users.
Some companies worry that allowing AI to train on their content sets AI up to be a competitor. This is particularly true for consulting companies or other knowledge-driven organizations. However, any individual company’s information is rarely so unique and precious that the quality of AI results will be noticeably poorer by denying AI the ability to train on this data.
Remember that AI aggregates all public information, no matter how obscure. Bits of pieces from each of multiple sources are integrated into a deeper understanding. Let’s even grant that there is no other article on the entire Internet — no matter the language — that’s as insightful as your masterpiece about a particular issue. AI will still have more insights than your piece by combining, integrating, and synthesizing all the lesser articles it will find. Some guy in Uzbekistan added a small and insignificant insight over and above what you said, when writing his blog in Uzbek? AI will see it, understand it, and translate that tiny bit when synthesizing its answer to the user asking about that issue.
Denying your content to AI makes you invisible on the Internet. Your legacy will vanish from history. On the other hand, allow AI to train, and part of you will live forever in that vast knowledge base. More important, from a business perspective, having your content in AI means that there is some probability that it will be presented to customers and that some of them will visit the source (i.e., you) for more information.
You want AI to pick your content for inclusion in the answer it presents to its users. This presents a content strategy challenge because AI is prospecting in a very flowery field of many other options. (Midjourney)
Ranking in AI Results
Google famously became the best legacy search engine by being better at sorting the many Internet resources that could hypothetically address the user’s query so that the most helpful were placed on top. Google’s initial breakthrough was PageRank: a page's importance can be measured by analyzing how many other pages link to it and how important those linking pages are. The algorithm treats each link as a vote of confidence, with more authoritative pages casting more valuable votes.
Google has since improved its ranking system, emphasizing user engagement signals when scoring destination pages. This again has changed the best-practice recommendations for search engine optimization (SEO) from early gung-ho methods to current methods that stress creating pages that are better at answering users’ questions.
Engagement metrics will be important for ranking in AI, just as they already are for SEO. The difference is that AI will be better at predicting true user engagement with your services and interest in your content. (Leonardo)
Similarly, the advice I give here will likely change over the coming decade as AI becomes more advanced and more established.
My first recommendation for having your content ranked by AI results is to follow those SEO best practices I just linked. Many AI answer engines seem to rely on traditional search to generate their initial list of websites from which to expand their “deep research” continued reading. Being in the initial set doesn’t guarantee that you’ll be in the final result shown to users, but it’s a good start.
I expect AI providers to move to their own ranking model soon, especially for Deep Research products. We could call this new ranking “DeepRank” because AI can score each website and each page for how useful it is, based on its deep understanding of the content. If something has many errors or presents a shallow understanding of its topic, AI will realize this and rank that content lower.
I expect DeepRank to work on a site-wide basis, similar to the “authority” scores that Google has been accumulating for websites. This will allow AI to immediately know which sources to consult first for any given user problem. After AI has read and analyzed a page, it’ll also keep a more specific DeepRank for that page.
The only way to get a high DeepRank is to publish thoughtful and informative content consistently. It is much simpler than old-school SEO and even simpler (if more challenging to achieve) than current SEO.
Obviously, AI currently doesn’t have perfect judgment, so initial DeepRank scores may get fairly low weights compared with traditional SEO scores. However, each generation of AI gets more and more intelligent, and after 2030 AI will likely achieve superintelligence and be better than humans at judging the value of content.
Some of this may already be happening. A study in 2024 found that only 23% of sources cited by Google’s AI were found on Google’s own SERP (search engine results page) in the top 10 spots traditionally considered essential for Internet survival. In this study, 3/4 of what Google AI told users came from sources that the AI deemed to have some small bit of useful information, even if they weren’t overall substantial enough sources to rank in a traditional search.
AI-based answer engines also prioritize diversity of sources, which again tends to surface deep content from websites that are not strong enough to be top hits in traditional searches.
Finally, as AI gains users, the leading models will accumulate the type of usage data that has served Google so well in the past. Google knows which search hits people click, and those pages gradually drift toward the top of the listings. Similarly, as AI increasingly shows citations and other interactive elements, such as the ability to expand a short point, it will accumulate statistics about what content its users find interesting. Such content will also get higher DeepRank and gradually be cited more than content that users tend to ignore.
How to Create Content That AI cites
Here are 14 guidelines for website content to improve your chances of being used by AI:
Direct relevance and answer content: If a webpage contains a clear, direct answer or a crucial piece of information that the AI includes in its response, it’s likely to cite that page. For example, if a user asks, “What are the benefits of cloud computing for small businesses?” and your blog post has a concise list of those benefits, the AI may pull a point or two from it and cite your blog. Content that is formatted to answer common questions (using Q&A format, headings that match the query, or summary paragraphs) often gets picked up because the AI can easily identify the answer within the text.
Unique information or perspective: AI models prefer authoritative and original information. If your site is the origin of a fact, statistic, or quote that the AI wants to use, you stand a high chance of being cited. For instance, if a news article on your corporate site breaks a development, or your e-commerce site publishes a unique study (“X% of customers prefer…”) and an AI answer includes that statistic, it should cite you as the source. Case in point, Bing’s chatbot cited Design Ergonomics (a dental office design firm’s site) as a source when answering a question about dental office layouts. In fact, the link I just gave you is an example of this technique, because David Meerman Scott wrote an article with that rather niche dental example that Deep Research picked up when I asked it to give me examples of how to be ranked by AI.
Being the go-to source in your domain (or coining a term/concept, such as, let’s hope, “DeepRank”) will make the AI more inclined to quote you as the authoritative reference.
Authority and trustworthiness: Like search engines, AI systems know which domains are trustworthy. High-authority domains (established news outlets, well-known brands, reference sites, etc.) often get cited because the AI is more confident in their information. That said, authority is often context-specific: a small specialist blog might be the most trusted source for a specific technical question. Building your site’s authority through quality content and backlinks (and even being referenced by authoritative sources yourself) can signal AI that your content is worth citing.
Content breadth vs. depth: Because AI answers pull from multiple sources, you don’t necessarily need to answer every aspect of a broad question; you can focus intensely on one aspect and still get cited for that part. However, having a hub of content that collectively covers the breadth of a topic (in multiple pages) can make your domain a one-stop source for AI. Some early findings suggest that multiple pages from the same website can all be cited in one answer. This means if you have several strong pages on different facets of a topic, an AI might cite two or three of them at once to answer a multi-part question. This “topic cluster” approach (pillar overview page + supporting detail pages) is a promising strategy for AI visibility.
AI can pick out a small, focused part of your content where you provide a unique insight it wants to flush out its answer. (Ideogram)
Clarity and extractability: Content that is well-structured and easy for an AI to parse is more likely to be cited. This means using clear sentences, descriptive headings, and bullet-point lists and avoiding burying key info in long, complex paragraphs. If the AI can quickly “grab” the answer from your page (because you perhaps bolded it or stated it in the first sentence or in a list of key points), it will do so and credit you. On the flip side, if the information is hidden behind interactive elements or requires clicking through slides, an AI might skip it. So a plain-text, well-formatted answer is basically an invitation for the AI to cite you.
Freshness and recency: Especially for topics that evolve (news, tech, finance, etc.), the AI will look for the most up-to-date information. Content freshness can affect visibility in ChatGPT’s citations according to one study. Ensure your content is updated with current statistics, dates, and developments to remain citation-worthy for queries about the “latest” or “2025,” etc.
Query intent matching: AI models try to satisfy the specific intent of the query. If the question is comparison-oriented (“X vs Y, which is better?”), the AI might cite a review or comparison article (perhaps one that clearly compares features of X and Y). If the query is looking for “according to [source]” or “did [Person] say…”, the AI will specifically hunt for that source or person’s quote. Content that matches the user’s implied intent (e.g., a how-to article for a “How do I…?” question) has an edge. This is analogous to SEO, where aligning content with searcher intent improves ranking; here, it enhances citation chances.
Provide concise answers up front: Write brief summaries or key takeaways at the top of your articles. AI can grab a few sentences that directly answer the main question or summarize the topic. Think of this like an “abstract” or executive summary. You can then elaborate below. The AI might use the concise part verbatim and cite you while ignoring the rest, but you still get the credit, and the user may click through for details. For instance, start a blog post with a bold statement or definition of the topic in 2–3 sentences. This could become the cited snippet.
Use natural language: AI models match on meaning, not just exact keywords. Writing in a natural, conversational tone (the way a user might pose a question) can help your content align with AI queries. You can also cast a wider net for semantic matching by covering both “advantages” and “drawbacks” in your headings. However, stay focused and don’t stuff content with irrelevant keywords.
Include factual details and figures: Concrete facts (dates, statistics, specific features) in your content are catnip for Deep Research. If a user’s query asks for a stat or a specific fact, having that in your text makes your page a candidate to cite. Whenever possible, add unique data or insights from your business domain. Original research and data draw AI citations because the AI prefers to cite the original source of a fact. Just ensure any data is clearly presented with numerals: AI might even pick up “according to [Your Company]’s 2025 report, X%…” and cite accordingly.
Maintain accuracy and update content: AI models are improving at rewarding accuracy. Regularly update your high-performing pages with current information. If an AI has two possible sources to cite — one from 2019 and one from 2025 — it will often lean to the more recent for a current query. Indicate the last update dates on your page (e.g., “Updated Jan 2025”) so both users and AI know it’s fresh. For news articles, include the publication date visibly.
Multiple corroborating sources: Sometimes, an AI will mention a fact that appears in several sources. It might choose one or two of them to cite in such cases. How does it choose? Likely the most authoritative or the one that phrased it best. For content creators, if you’re writing about a widely reported fact, adding unique insight or the most straightforward explanation might make your version the one that gets picked. Also, if your page covers several aspects of a complex question, the AI might cite your page for one part and other pages for other parts. Indeed, ChatGPT can generate multiple citations in one answer. Therefore, covering a topic comprehensively (so that you could potentially answer multiple sub-questions) increases the odds of snagging at least one of those citation slots.
Content type and media: The form of content can also influence citations. AI systems primarily quote text but also display other media to enhance answers. While those aren’t “citations” in the text, they are sources being featured. So, you increase your footprint if you produce content in multiple formats (articles, infographics, videos). An AI might cite your text article while also showing a thumbnail from your infographic or mentioning that you explain a concept in a video. Particularly for “how-to” or product queries, an image from a step-by-step guide or a chart from a study might appear with credit to the site. Thus, providing helpful visuals (with proper alt text/captions describing them) can indirectly get your site featured in AI-driven results.
Tracking AI Use of Your Content
AI will likely continue evolving at lightning speed during the next decade. This means that what works this month may be a suboptimal strategy next month. I’m sure there will soon be a 15th guideline to be added to my list above, and maybe some of my guidelines will become less important in the future.
You should continuously monitor how AI treats your content since this will likely change. Two main ways of doing this:
Checking if and when AI services cite your site by asking them questions you aim to answer and see who gets cited. Use this intel to adjust content. If you see a competitor’s site is consistently cited for queries you want, analyze their content (how is it structured, what phrasing did the AI latch onto) and update your content accordingly. (Not obviously by directly copying a competitor, but by employing similar techniques to theirs.)
Watch your analytics like a hawk. Track which AI services send you how much traffic each month and which pages get the most of this traffic. Changes in these patterns can alert you to changes in the AI ecosystem that are not necessarily picked up by general influencer chatter.
Finally, stay tuned to that AI influencer chatter because new tips will likely be published occasionally.
The content strategy cycle to ensure that your company is represented well in AI results. (Napkin)
Premium Content in AI
Right now, AI models mostly cite freely accessible web content. In the future, we might see licensed content integrations. For instance, an AI might have access to paywalled articles (via partnerships) and can cite them, or use data from premium databases. This could mean, for example, an AI answer about market trends might cite a trusted industry report or a research firm (with permission).
For content creators, this means new opportunities to be cited in AI. If your content is behind a paywall, consider offering some for free or summaries that can become training data, otherwise you might be invisible to AI. Conversely, high-quality content producers might find new distribution via AI partnerships. Imagine an AI that can answer medical questions and cite articles from a medical journal because it’s licensed: that raises the bar for what content is shown.
Paywalls are not the only offenders. In the B2B sector, it’s common for companies to produce gated whitepapers that are only available to prospects who enter their email address. This has always been an odious practice that violates users’ privacy, but in the AI era, it also means that those companies won’t be cited as much by AI. Let’s hope they see the light and make whitepapers available without gatekeeping. Good riddance to any login walls!
There are now many AI labs, and they all tend to converge on about the same level of intelligence within a few months of any one lab launching an innovation. Licensed content will become a sustainable differentiator for AI. Sadly, one that will favor the biggest labs. However, as a content provider, you want to be part of these deals; the more, the merrier. You want to get customer traffic from OpenAI, Perplexity, Anthropic, xAI, Mistral, and the list goes on, so you should partner with them all.
On this point, it’s a disgrace that I had to mention “a medical journal” as an example of proprietary content. Healthcare information should be open to all — particularly since much of it is funded by the taxpayers in the first place.
In the old days, it might have been acceptable for research papers to be locked up in databases requiring login to access, since it would mainly be accessed by other researchers who tended to work at institutions with the necessary subscriptions. Today, many patients research their conditions with Deep Research to be better prepared for meeting with their doctor. Any research paper that’s not freely accessible online equates to worse health outcomes for these patients.
While medical research presents the most potent moral argument for avoiding closed databases, all research results should be freely accessible unless they are genuinely proprietary work funded by a company for use in its own product development. Certainly, from the perspective of the scientists who live by citation counts, your work will have a vastly reduced impact unless you're in AI.
Locking up content is not just unethical in the case of medical findings or taxpayer-funded research, it will also make your company invisible to AI. (Midjourney)