Disclosure: Yolando gave AiNews.com access to its platform in exchange for promotional coverage, including this article. Yolando supplied the initial analysis and an early draft, but AiNews independently reviewed and verified the findings, substantially rewrote the article and retained full editorial control. Yolando reviewed the article before publication, while AiNews made all final editorial decisions.

Claude recommends AiNews.com more than any other AI model we measured. There’s just one problem: it never once cited AiNews.com.

We discovered that through Yolando, an AI visibility platform that tracks how often brands appear in AI-generated answers, what the models say about them and which sources they use.

Across the periods Yolando measured, Claude mentioned AiNews in 12.9% of relevant answers—the highest rate among the AI systems included in the analysis. But it did not directly cite our website once.

At first, I wasn’t sure what to make of that. Claude clearly knew enough about AiNews to recommend us, so why wasn’t it using our reporting as a source?

That question opened a much larger one. We tend to talk about AI visibility as though being known, being recommended and being cited are all versions of the same thing. Yolando’s data showed us that they are not. An AI system can speak positively about a company when it appears, yet rarely think to include it in the first place. Yolando recorded tens of thousands of references to AiNews through third-party websites, while our own site was cited far less often. And one AI platform can see the same company very differently from another.

That matters because research suggests AI-generated search summaries are already reducing clicks to traditional results. Pew Research Center found that users clicked a traditional result in 8% of visits to Google search pages with an AI summary, compared with 15% of visits without one. Ahrefs separately found that the presence of an AI Overview correlated with a 58% lower average click-through rate for the top-ranking page, compared with the 34.5% reduction it measured in an earlier study. As discovery shifts from lists of links toward AI-generated answers, whether an AI system names a publication or business is no longer a vanity metric. It can affect the top of the discovery funnel.

AiNews is the case study for a much larger business question: What happens when customers ask AI which companies they should trust, buy from or pay attention to? Whether a business appears in that answer may depend on far more than what it publishes on its own website.

There is no proven formula that guarantees an AI system will choose one company over another. But businesses are not powerless. They can make it easier for AI systems to understand what they do, answer the questions real customers ask, build legitimate signals beyond their own websites and measure whether those efforts appear to change anything.

If an AI system knows your company but doesn’t use your website as a source, what determines whether it recommends you—and what can a business actually do about it?

AI discoverability measures whether and how often a business appears in AI-generated answers, while citation share measures how often an AI system uses the business’s own website as a source.

I met Martin Cheung, Yolando’s Founding Director of GTM, at HumanX in April 2026. AiNews had started paying much closer attention to AI discoverability, and I took the meeting because I wanted to understand how businesses could tell whether they were appearing in AI-generated answers.

During the meeting, Martin pulled up AiNews and showed me something I had never seen before: systematic data about how AI models viewed our publication. I could see the questions Yolando was asking, which competitors appeared in the answers, which sources the models cited and how often AiNews showed up.

At the time, AiNews appeared in roughly 10% of the relevant answers and ranked ninth among the publications being compared. I think Martin expected me to be disappointed.

I was thrilled.

AiNews.com is a small, independent publication competing against companies with much larger audiences, longer histories and far more resources. I didn’t look at that 10% and think, We’re only at 10%. I thought, We’re already on the board.

A month or two later, I tried to measure our AI visibility myself. I created 50 questions and tested them manually across different AI systems. It took forever—and this time, AiNews essentially disappeared from the results.

The difference showed me why measuring AI visibility is so difficult. The questions you ask matter. The way you phrase them matters. The platform you use matters. Even asking the same question again can produce a different answer. To find a meaningful pattern, I would need to repeat enough questions across enough AI systems and over enough time to separate a trend from a random result.

That was what made Yolando useful to me. It turned something I had found extremely difficult to measure manually into a process that could be repeated, compared and tracked over time.

A traditional search rank tracker tells a business where one of its webpages appears in a list of search results. Yolando asks a different question: When someone asks an AI system for information or a recommendation related to your industry, does your company appear in the answer? 

Yolando begins with the audience rather than keywords. It builds questions that people in a company’s category might actually ask an AI assistant, uses search-demand data to identify the phrasings they use and groups those questions into the subject areas where the company competes. For AiNews, that process produced 25 questions. Yolando also tags each question according to where it sits in the reader’s journey. The four questions at the top were broad discovery questions that never mentioned a brand, including questions about where to find reliable AI news. The 19 middle-of-journey questions were comparisons, including questions about resources on specific AI topics or recommendations for podcasts and newsletters. The two bottom-of-journey questions asked about AiNews directly.

Those 25 questions were designed as a representative sample of the kinds of questions AiNews readers might ask; they were not intended to capture every possible question about AI publications.

The 25 questions covered five topics:

The initial setup took approximately 15 minutes. I then spent another 30 minutes with Yolando refining the question set and deciding which publications belonged in the competitor comparison. That second step was more consequential because it helped determine whether the measurement reflected questions our readers might genuinely ask rather than simply producing a workspace filled with numbers.

Martin said Yolando’s content team uses the follow-up conversation and finalized question set to identify content gaps: questions or subject areas important to the audience that the company’s existing website does not address or does not cover in enough depth. Inside the platform, those gaps appear as detailed recommendations explaining what content the company could create, expand or update. The initial setup creates the workspace, the refinement call shapes what Yolando measures and the resulting data informs the content recommendations that follow.

Yolando then ran those questions repeatedly across ChatGPT, Gemini, Perplexity and Claude. Across two sampling periods covering approximately six weeks in total, those repeated runs produced roughly 2,700 question-and-answer observations. The analysis compared AiNews with 10 other AI newsletters and publications: The Rundown AI, The Algorithm, Superhuman AI, The Batch, VentureBeat, TLDR, Import AI, TechCrunch, Ben’s Bites and AI Weekly.

Those two sampling periods provide a picture of how the four AI systems responded during the measured question set, but they are not enough to establish a trend over time.

AI discoverability isn’t a set-it-and-forget-it optimization. Generative AI systems can give different answers even when someone asks the exact same question. A publication may appear one day and disappear the next. A competitor that ranks first in one response may rank fourth in another.

Research illustrates how much those answers can vary. BrightEdge analyzed tens of thousands of identical prompts across ChatGPT, Google AI Overviews and Google AI Mode and found that the platforms produced different brand recommendations for 61.9% of queries. Only 17% produced the same brands across all three. In a separate study reported by Search Engine Land, 600 volunteers ran 12 identical prompts through ChatGPT, Claude and Google’s AI nearly 3,000 times. Across the platforms and prompts, the odds of receiving the same brand or product list twice were less than one in 100.

Asking once gives you one answer. Asking repeatedly begins to show you a pattern.

Yolando uses those answers to calculate four main measurements:

The overall scores provide a starting point, but the more useful information often sits underneath them. Yolando lets users open an individual question, see the AI-generated responses, identify which competitors appeared and examine the websites and pages used as sources. That helps turn a low score into something a business can investigate.

The questions also need to keep up with the business they are measuring. Some of the original AiNews prompts asked for the best “daily” AI newsletter because that reflected our positioning when the account was created. We have since changed our publishing schedule to three times a week.

Questions about the best daily AI newsletter can still show us whether AI systems continue to associate AiNews with that type of coverage. But those questions no longer fully represent how we describe the publication or the audience we are trying to reach today.

Models change. Competitors change. The information available online changes. A company’s positioning, products and customers can change too. Businesses need to monitor their AI visibility over time and update the questions they measure as those changes happen so they can catch shifts in how AI systems understand, describe and recommend them—and decide when action may be needed.

Yolando gives us a baseline we can return to, update and measure again. That makes it possible to see whether our visibility is changing, identify where we may need to take specific action and then measure whether those changes appear to help.

Once we had enough data to establish a current benchmark, Yolando placed AiNews at 9.1% discoverability. That means we appeared in fewer than one out of every 10 relevant AI-generated answers.

The four discoverability leaders and AiNews’s position were:

We ranked tenth out of the 11 publications being compared. Our share of voice—which measures how much of the overall conversation belonged to AiNews—was 2.3%, also placing us tenth. By comparison, The Rundown AI, which led the group in discoverability, had a 10.8% share of voice.

Those numbers sound terrible without context.

AiNews was being compared with publications such as TechCrunch and The Rundown AI. Other competitors included The Algorithm, MIT Technology Review’s weekly AI newsletter; The Batch, published by DeepLearning.AI, which was founded by Andrew Ng; and VentureBeat, which launched in 2006 and focuses on enterprise technology. Several of these competitors reach much larger audiences, have been publishing longer and operate with larger editorial teams. By contrast, AiNews launched in October 2023 and operates with a two-person team and a much smaller budget.

That is why I did not see 9.1% as evidence that we had failed. We were already appearing alongside publications with significantly more history, authority and resources behind them. The result gave us a baseline—and showed us that AI systems already considered AiNews relevant enough to include in some of those answers.

The ranking also changed depending on what Yolando measured. AiNews ranked tenth in discoverability but eighth in citation share, placing us ahead of TLDR, Ben’s Bites and AI Weekly on that particular measurement. Simply put, we were not performing equally across every part of AI visibility.

Martin gave me another piece of context that helped make sense of the small percentages. AiNews.com’s direct citation share was approximately 0.3%. That placed us eighth among the 11 publications in our competitor group and 54th among all websites cited in the measured answers. Martin said reaching roughly 1% would move AiNews.com into approximately the top 10 citation sources overall.

And yes, the benchmark is only 1%. That number can sound almost absurdly low if you read it like a grade. Citation share does not measure how well a website scored out of 100. It measures how much of the total pool of citations went to that one domain. A citation share of 90% or 100% would mean that nearly every citation across the measured answers pointed to the same website. When those citations are spread across a wide range of sources, reaching 1% can be enough to make a website one of the most frequently cited in the entire dataset.

A fraction of a percentage point can look meaningless until you understand the scale around it. The 0.3% showed us where we were starting. The 1% gave us a meaningful benchmark to work toward: becoming one of the sources AI systems return to more often.

These numbers become useful only when a business understands what is being measured, who it is being compared with and what a strong result actually looks like. Without that context, even a meaningful result can look like failure.

Yolando’s reputation data told a much more encouraging story than our discoverability ranking.

AiNews received a reputation score of 81.2. Across the measured answers, AI systems described us positively 1,981 times and negatively 458 times. When AiNews appeared in a list, our average position was 2.2—meaning we were usually placed near the top.

Yolando also grouped those descriptions into 611 recurring reputation themes: 423 positive and 188 negative. The three most common positive themes were broad AI coverage, in-depth analysis and expert-driven insights. The three most common negative themes were research coverage depth, policy coverage focus and enterprise AI coverage. In this analysis, “negative” did not mean readers had criticized AiNews, nor did it necessarily mean the AI response was unfavorable overall. It meant that the model framed an attribute as a limitation for the particular question being asked. AiNews’s narrower publishing scope, for example, can count against it when someone wants comprehensive, high-volume investment news—but it can work in its favor when someone wants selective, deeper analysis.

That distinction also explains why depth appeared on both lists. “In-depth analysis” described whether AiNews explained what happened and why it mattered rather than simply directing readers to someone else’s account. “Research coverage depth,” by contrast, referred to how directly a publication engaged with primary technical work, including reading and analyzing research papers.

The comparison set helped determine which description appeared. Against roundup-style publications such as TLDR and Ben’s Bites, AiNews’s explanatory depth was a differentiator. Against technical specialists such as Import AI and The Batch, which publish primary technical analysis, research coverage depth became a limitation for us. The same publication and editorial approach could therefore be assessed differently depending on which competitor appeared beside AiNews in the answer. Together, those comparisons placed AiNews in the middle of the spectrum: deeper analysis than the roundup-style publications, but less specialized in primary technical research than the research-focused publications.

Put those numbers beside our 9.1% discoverability score, and the problem becomes clearer. The models seem to like the publication when they mention it. They simply do not think to include us often enough.

That distinction matters because a company with a reputation problem faces a different challenge from one with a visibility problem. If AI systems regularly mention a business and describe it negatively, the company needs to understand what is shaping that perception. If the descriptions are positive but the company rarely appears, the more immediate question is why competitors are being included more often.

Those negative themes still gave us areas worth investigating. They are useful signals because they point to topics where other publications may be seen as stronger or more specialized—and may therefore be recommended more often for those questions.

But reputation data also comes with an important complication: AI systems do not always separate similarly named businesses correctly.

During my earlier manual testing and again while reviewing Yolando’s results, I found examples where an AI system appeared to confuse AiNews.com with the similarly named publication Artificial Intelligence News. One response described AiNews.com as reputable and knowledgeable, then criticized us for having too many advertisements on our homepage.

At the time that characterization was measured, AiNews.com did not have a single advertisement on its homepage.

That example was only one reason entity confusion became my biggest concern when interpreting the reputation data. Even Claude mixed up AiNews.com with Artificial Intelligence News when I asked directly about our publication.

Yolando lets me open and read the individual characterizations, which is important because the total alone cannot tell me how many were actually describing AiNews.com. The examples of entity confusion mean each criticism needs to be checked before we use it to make editorial decisions.

For businesses that see weak results and have no idea what to do next, Yolando’s content recommendations can identify questions where the company is not appearing, topics where competitors are appearing, existing content that may need improvement and gaps worth investigating.

That gives businesses a specific place to start: a visibility gap and a recommendation that says, in effect, Start here. This may help. Human judgment still determines whether that recommendation fits the company’s customers, expertise and larger strategy.

For AiNews, that means using the findings to inform our editorial decisions without allowing an AI visibility score to make those decisions for us. If AI research coverage is a real gap that matters to our readers, we can address it. If a recommendation would pull us away from the audience we serve, we can leave it alone. The recommendations also provide evidence to review before we write or update content intended to close a gap.

Those choices may need to differ by platform, because Yolando’s results showed that each AI system was seeing AiNews differently.

We often talk about “AI search” as though it were one channel. Yolando’s results showed us several different versions of AiNews depending on which AI system answered the question.

In Google’s results, AiNews received a reputation score of 98.3. Perplexity gave us a reputation score of 57.7—a difference of more than 40 points for the same publication during the same measured periods.

Those scores summarize how Yolando classified the way each AI system framed AiNews—as positive or negative—rather than objective facts about our reputation. Some of the 40-point difference may reflect genuine differences in how the two systems described us, while some may reflect how Yolando’s scoring system interpreted those descriptions and converted them into scores. The underlying characterizations therefore need to be reviewed alongside the totals.

Perplexity was also where AiNews had its lowest discoverability, at 4.2%. Claude produced the opposite kind of problem: it mentioned AiNews in 12.9% of relevant answers, more often than any other AI system measured, while directly citing AiNews.com zero times.

That absence stood out because AiNews published several Anthropic-related articles during the sampling windows. We covered the Claude Opus 5 launch, the Fable 5 shutdown and what it meant for U.S. AI oversight, and Anthropic and AISI’s finding that evaluation agents reached real systems. Claude cited none of those articles.

A business looking only at an overall score could miss those differences. One platform may describe the company positively but rarely include it. Another may recommend it frequently while relying on other websites as sources. A third may see the company less favorably than the others.

For decades, businesses largely built their search strategies around Google. Customers can now ask ChatGPT, Claude, Gemini, Perplexity and other AI systems the same question and receive different companies, descriptions and sources in response. Performing well on one platform gives a business only one part of the picture.

That makes platform-level measurement important. Yolando lets businesses see which systems are including them, where they are falling behind competitors and which responses or sources may be contributing to the pattern. For AiNews, Perplexity clearly emerged as an area that deserves closer investigation because it gave us both our lowest discoverability and our lowest reputation score. Any changes we test will therefore need to be measured separately across all four platforms, because an improvement in how one system includes or describes AiNews is not evidence that the others changed their minds.

Right now, we can only observe those differences. We still cannot say with certainty what caused each AI system to recommend or cite one business over another. The explanation could involve the sources it retrieved, how it interpreted the question, differences between the models or some combination we cannot see.

The platform differences raised another question: Which sources were AI systems using to decide what counted as relevant or authoritative? The source at the top of Yolando’s list was one I did not expect: Reddit.

Reddit accounted for 11.7% of the citations Yolando recorded in our category. That was more than twice Wikipedia’s 4.9% and nearly four times arXiv’s 3%.

As a Gen X-er, I admit this feels a little backward. We were taught that Wikipedia was not an authoritative source because humans could write and edit it. Then AI came along and apparently asked: Where are all the humans talking?

Reddit.

In Yolando’s measured answers, a discussion forum, a collaboratively edited encyclopedia and a preprint server occupied the top three positions. Axios was the only conventional newsroom among the top 10. In this dataset, those three source types appeared more often than the trade press covering AI every day.

Yolando’s result was not isolated. A March 2026 Peec AI analysis of 30 million sources across five AI platforms ranked Reddit as the most-cited domain overall and placed it first or second on every platform examined. A separate Semrush study of more than 100 million citations across 230,000 prompts also placed Reddit among the top five domains on ChatGPT, Google AI Mode and Perplexity—while showing that citation patterns can change sharply over time.

The larger lesson is that authority in AI discovery may look broader than many businesses expect. Company websites, journalism and research still appear in the source data. So do community discussions, social platforms, independent articles and other places where people share experiences and opinions about companies and products.

For businesses accustomed to SEO, that changes the information environment they need to consider. Publishing useful content on the company website remains important. Reviews, industry directories, media coverage, LinkedIn articles, expert references, podcasts and legitimate discussions elsewhere on the web may also help AI systems understand how the business fits into its industry.

That does not mean companies should rush to Reddit and start manufacturing conversations about themselves. Artificial promotion can damage trust with the very communities a business is trying to reach.

The more useful takeaway is that human perspectives and independent third-party sources appear to play a larger role in AI discovery than many businesses may realize.

For AiNews, the results point to a more specific hypothesis. We have invested heavily in clear article structure, direct answers, key takeaways and explanations of why each development matters. We have also updated our SEO metadata, added an llms.txt file, reviewed our robots.txt settings and rewritten key website language to explain more clearly who we are, what we publish and who we serve. In other words, we had already completed a substantial amount of first-party optimization. AI systems generally describe us positively when we appear. Yet our discoverability remains low, and only a small number of independent third-party sources currently establish AiNews as a publication worth knowing and recommending.

All of that suggests first-party optimization alone has not been enough. AiNews likely needs more legitimate third-party references—such as media coverage, industry directories, expert mentions, partner websites, podcasts and social articles—to give AI systems broader evidence that the publication is known and trusted.

Yolando’s data shows that third-party sources appeared frequently in the measured answers, but it cannot tell us exactly how much influence additional third-party references to AiNews would have or guarantee that they would improve our visibility. That is a hypothesis we can test by strengthening those signals and measuring whether our discoverability and recommendations change.

That uncertainty leads to another important distinction. An AI system may be able to read and understand everything a company says on its own website while still lacking enough evidence to recommend that company to someone else.

Understanding your content and recommending your brand are two different jobs.

Traditional search encouraged a relatively straightforward strategy: publish strong content, optimize it and work toward being found. AI discovery adds another layer. Great content can shape an answer only if it becomes part of the information an AI system draws on when deciding what to recommend or cite. If it never enters the information pool used for that answer, its quality cannot affect the result. Yolando cannot show whether that explains why AiNews was omitted from any particular recommendation, but it gives us one possibility to investigate.

Consider two questions someone might ask an AI system:

The first question asks the AI system to understand the company. A clear website may give it everything it needs to explain what the business offers, who it serves and how its products or services work.

The second question asks the AI system to make a recommendation. Now it has to compare companies and decide which ones deserve to appear in the answer. The information on a company’s own website may be only one part of that decision.

That distinction helps explain what we saw with AiNews. The models know we exist, generally describe us positively and sometimes recommend us. Yet our overall discoverability remains low, and Yolando recorded far more references to AiNews through third-party domains than through our own website.

Understanding your content does not automatically give an AI system enough evidence to recommend your brand.

The company website remains the foundation. It should clearly explain who the business is, what it offers, who it serves and why it is different. The wider web adds another layer through reviews, media coverage, expert references, directories, partnerships and other independent discussions of the company.

That changes the question businesses need to ask about AI visibility. “How do I get my webpage cited?” is still important. Businesses should also ask: “What evidence exists across the web that gives an AI system a reason to know, understand and recommend my company?”

We do not yet know exactly how AI systems weigh first-party content against third-party information. Yolando’s results and our conversation with Martin suggest that independent signals may play a larger role than many businesses realize.

Yolando’s data gives us a baseline for examining that hypothesis: its AiNews dataset contained 52,104 source references through third-party domains and just 146 through AiNews.com—a difference of roughly 357 to 1.

Before deciding whether that ratio was a problem, we needed to know whether it was unusual.

The 357-to-1 ratio initially looked like one of the worst findings in the entire analysis.

But first, the numbers need to be understood correctly. Yolando recorded 52,104 source references through third-party domains and 146 through AiNews.com across the repeated answers in our dataset. These figures count how many times sources appeared across those answers, rather than the number of unique websites. If the same website appeared as a source in 100 different answers, it would contribute 100 references to the total.

Martin explained that AI systems normally use multiple sources to build an answer, so most of the source references surrounding a company often come from outside its own website. The highest direct-domain share he had seen from one of Yolando’s clients was approximately 10%, which he considered exceptionally strong.

Even at 10%, that means that roughly 90% of the source references would still come from somewhere else.

That context changed how I looked at our 357-to-1 ratio. A large share of what AI systems use to understand AiNews comes from websites we do not own or control. The ratio shows that AI systems used third-party domains far more often than AiNews.com in the measured answers. It may mean our reporting is being underused as a direct source. It could also reflect the types of questions in the dataset: AI systems may have needed information about AiNews rather than reporting published by AiNews, preferred other sources for particular questions, failed to retrieve our pages at all, or made choices for reasons we cannot observe. The data shows the pattern, but it does not identify the cause.

My next reaction was: Wait. What does that actually mean?

If another website gives an AI system accurate information about AiNews, the model recommends us and someone subscribes, how much should I care which domain supplied the information?

As a publisher, I do care. A direct citation gives AiNews attribution, establishes our reporting as a source and may send a reader to our website. That traffic also matters to advertising revenue. Publishers need website visits, impressions, clicks and other performance data to show advertisers whether their campaigns reached readers and performed as expected. Direct citations also give us more control over whether the information being used is accurate and current.

But a third-party reference can still create business value. If an AI system recommends AiNews because it found us in an industry directory, a podcast page, a partner biography or an independent article, that recommendation may still lead someone to discover and subscribe to the publication.

The ratio therefore raises two different questions. How do we increase the number of times AI systems use AiNews.com as a direct source? And how do we strengthen the legitimate information about AiNews that exists across the rest of the web?

Both matter, but they may serve different goals. Direct citations can provide attribution, authority, potential referral traffic and measurable advertising value. Third-party references can provide independent evidence that helps AI systems decide where a company belongs among the businesses being recommended—and whether it should be considered at all.

That distinction forced me to reconsider another assumption I brought into this analysis: that citations always matter more than mentions.

I went into my conversation with Martin believing that citations mattered more than mentions. Much of the AEO advice I had been reading also focused on getting AI systems to cite a company’s website as a source.

A citation felt like the harder—and therefore more valuable—result. The AI system had not only recognized the company; it had chosen to use something from the company’s website to build its answer.

Martin challenged that assumption.

He explained that citation rates and mention rates often move together, but from a business perspective, mentions may be more important. The goal is usually for the AI system to recommend the brand when someone is deciding which company, product or publication to choose.

His example made the difference clear. An AI system could cite reporting from AiNews.com while recommending that the user subscribe to TechCrunch. We would receive the citation, but TechCrunch would receive the potential subscriber.

The opposite outcome could also happen. An AI system might recommend AiNews as a publication worth reading without directly citing one of our articles. We would lose the attribution and possible website visit, but the mention could still lead to a new subscriber.

AiNews is free to read and funded through advertising and sponsorship. That model depends on readership, and readership depends on discovery. Historically, discovery meant search, social media and word of mouth. Increasingly, it can also mean someone asking an AI assistant which newsletters are worth subscribing to and acting on the few names that appear.

For AiNews, both outcomes have value. Mentions can create awareness, consideration and subscriptions. Citations can provide attribution, authority, referral traffic and the website activity we need to demonstrate value to advertisers.

I had been treating the citation as the finish line. Martin’s explanation showed me that recommendation is another finish line—and for many businesses, it may be the one closer to revenue.

For AiNews, every answer in which someone asks which AI newsletters are worth subscribing to and hears other names instead represents a potential subscriber-acquisition opportunity happening without us.

The click still matters. So does being named at the moment someone is deciding what to trust, buy or subscribe to. AI systems can influence that decision before a customer ever visits the company’s website, and sometimes without sending a traditional referral click at all.

That becomes even more important when we consider how much more information customers reveal when they ask an AI system for a recommendation than they typically include in a short search query.

Traditional search taught businesses to think in keywords. Someone looking for a product might type:

best running shoes women

That tells a search engine the general product category, but very little about the person behind the search.

The same customer might ask an AI system:

I’m training for a half marathon and need women’s running shoes for plantar fasciitis in a size 8.

In one sentence, the customer has revealed what she is training for, the type and size of shoe she needs, a physical condition that may affect her choice and enough purchasing intent to suggest that she is actively looking for a product.

That creates an opportunity for businesses because an AI system has far more information it can use to match the customer with a specific product or company. The catch is that the AI still needs enough information to make that match.

A retailer may sell the perfect shoe for that customer. But if its website describes the product only as a “premium performance running shoe” and never explains that it is designed for longer distances, offers support for runners with plantar fasciitis or comes in women’s size 8, the AI system may have no clear reason to include it. The product can fit the customer’s needs perfectly and still disappear from the AI answer because the available information does not make that fit visible.

That is why businesses need content that clearly explains who a product or service is for, what problems it solves, how it is used, which needs or limitations it addresses and what makes it different. Product pages, service pages and FAQs should answer the questions customers actually ask in the language they naturally use.

The questions being measured can also provide a different kind of business intelligence. Traditional website analytics tell companies about the people who reached their websites. AI visibility measurement can expose some of the questions and decision-making situations happening before a potential customer ever arrives.

If a company repeatedly disappears from answers involving a particular need, customer group or use case, that may point to a missing piece of content, a gap in the company’s offering or a lack of independent information supporting its claims. The measurement cannot determine which explanation is responsible on its own. It gives the business a specific place to investigate.

Once a company can see where it is being left out of those decisions, the next question is what it can realistically do to improve its chances of being included.

AI discoverability can be frustrating and difficult to figure out, but businesses can take several steps to improve the information AI systems have available when deciding which companies belong in an answer.

The company’s own website is the place to start. It should make clear who the business is, what it offers, who it serves, which problems it solves and what makes it different. If that information is vague, scattered across several pages or buried under marketing language, an AI system may struggle to connect the company with a specific customer need.

Useful FAQs can help close those gaps because they allow businesses to answer complete questions in the same language customers naturally use. A potential customer may ask whether a service is available in a particular city, whether a product works for a specific condition or whether a company serves businesses of a certain size. Answering those questions clearly gives an AI system more information it can use when deciding whether the company fits the request.

Business information also needs to remain consistent across the web. That includes the company’s name, services, leadership details and contact information, such as its phone number, address and email.

If the company website lists one address or description while LinkedIn or an industry directory lists another, an AI system may repeat outdated information or struggle to determine whether the listings refer to the same business. Clear, matching details make it easier to connect those references to the correct company.

The next step is strengthening legitimate evidence outside the company’s own domain. Reviews can show that customers have experience with the business. Industry directories can confirm what the company does and where it fits. Media coverage, podcasts, partner websites, expert references and leadership or contributor biographies can provide additional context from sources the company does not completely control.

That does not mean creating artificial reviews or manufacturing conversations. Those tactics violate the rules of many review and community platforms, and the content may be removed if it appears fake. Customers who recognize the manipulation may also trust the company less.

Businesses do not need to place themselves in every directory available either. Inaccurate or outdated listings can spread conflicting information about the company, while low-quality placements may make an attempt to establish credibility look less credible. The goal is to build an accurate presence in places that make sense for the business, its customers and its industry.

Businesses should also think about distribution. Publishing useful content on a company website is important, but that content should not exist in isolation. Sharing real expertise through social posts, LinkedIn articles, interviews, partnerships and relevant industry communities gives potential customers more ways to encounter it. It also contributes to the broader information environment AI systems may use to understand where the company belongs.

Then comes measurement. Establish a baseline, identify a specific weakness, make a deliberate change and measure again. If a company is missing from questions about one service, it can improve the relevant pages and supporting information, strengthen credible third-party references and track whether its visibility changes over time.

Any movement should be treated as evidence to investigate rather than automatic proof that one tactic caused the result. AI-generated answers vary, and several things may change at once. The useful pattern is simple: measure, form a hypothesis, make a change and measure again.

Repeating enough questions across several AI systems often enough to identify a meaningful change can quickly become difficult to manage manually. That is where Yolando became most useful for AiNews.

Yolando gives us a repeatable baseline we can return to after making changes. It shows where AiNews appears, which competitors appear instead, how each AI system describes us and which sources it uses to build the answer. We can examine the overall scores, then open individual questions and responses to investigate what may be contributing to them.

The value comes from making more of AI discovery observable and measurable. Yolando cannot see inside an AI system and explain exactly why it chose one company over another. It can show us the pattern, give us competitive context and help us identify specific places worth investigating.

Yolando has published customer case studies describing improvements recorded during two client engagements lasting roughly three months. The company reported that Phoenix, a Canadian telehealth provider, increased its unbranded AI visibility for weight-loss treatment from 2.2% to 25.4% in 95 days. Yolando also reported that Coinflow, a B2B payment infrastructure company, earned 671 AI citations across four assistants during a 90-day engagement in which LLM traffic emerged as its third-highest inbound lead channel. These are vendor-reported results. AiNews did not audit the underlying methodology or independently verify that Yolando’s work alone caused the changes, so the examples should not be treated as proof of what another company can expect.

Those case studies reflect broader client engagements that included strategy and content support. Businesses evaluating Yolando have several ways to start, from a free page-level assessment to paid tracking and strategy services.

Yolando offers a free Chrome extension that, after the user creates a free account, grades a webpage for AI readiness and provides a checklist and recommended improvements tailored to the type of page. The extension evaluates features Yolando associates with citation readiness; it does not show whether an AI system actually retrieved or cited the page or recommended the business. That requires repeated visibility tracking.

Yolando’s Self-Serve plan currently costs $399 per month. For an organization as small as AiNews, that is a meaningful expense. It would have to produce enough useful information and measurable improvement to justify nearly $4,800 a year against everything else competing for the same limited budget.

The calculation may look very different for a medium or large company that already pays an employee, marketer, SEO specialist or outside agency to monitor search performance and develop content strategy. For that company, $399 per month could be affordable compared with paying someone to repeat questions across several AI systems, analyze competitors, investigate the results and determine what to do next.

Yolando’s Growth plan starts at $3,499 per month and is aimed at companies with larger budgets, teams and resources. Those companies may already have marketers, writers and SEO specialists capable of carrying out the work while still having no clear idea which AI visibility findings matter, where to focus or what their data is telling them.

The Growth plan adds biweekly strategy and review meetings with the Yolando team. They help interpret the data, identify priority questions, develop the content strategy and plan what the company should do next. For an organization with employees ready to execute, that personal guidance may be the most valuable part. The company receives analysis from specialists who work directly with AI visibility data, then its own staff can turn that guidance into action.

The price alone cannot answer whether either plan is worth it. That depends on the size of the company, the value of each customer and whether the business has the time and resources to act on what the platform finds. A dashboard identifying opportunities has limited value if no one can investigate or pursue them.

For AiNews, the partnership gives us an opportunity to find out what happens when we do act on the findings. That is enough to turn this article into an experiment.

Over the next 90 days, AiNews will test three main hypotheses.

Our first hypothesis is that AiNews may need stronger third-party authority. An AI system may understand our content and still lack enough independent evidence to recommend the publication consistently. We will work to strengthen legitimate references to AiNews through industry directories, media coverage, expert mentions, podcast appearances, partner websites, contributor biographies and other relevant sources. Then we will measure whether our discoverability and mentions change.

Our second hypothesis is that stronger first-party content may increase the number of times AI systems use AiNews.com directly. We will continue improving the clarity, depth, sourcing, accessibility and quotability of our own pages without changing our editorial strategy simply to satisfy an AI visibility score.

AiNews.com currently receives approximately 0.3% of the direct citations in Yolando’s dataset. Martin said reaching roughly 1% would place our website among approximately the top 10 citation sources in the measured answers. We will watch whether our direct citation share moves toward that benchmark.

Our third hypothesis concerns the questions themselves. Some of the existing prompts still reflect an earlier version of AiNews, including questions that describe us as a daily newsletter. We will update the question set to reflect our current schedule, audience and positioning, then examine the results at the individual-question and AI-platform levels.

At the question level, we will identify where AiNews rarely or never appears, which competitors appear instead and which sources or pages the AI systems used to build their answers. That will not tell us why the models selected those publications, but it will narrow the gaps worth investigating before we decide whether competing for them would make AiNews more useful to our readers.

Changing the questions also changes the measurement. Any new or substantially revised prompts will need their own baseline. We cannot replace an outdated question, receive a better score and claim that our visibility improved as though we measured the same thing twice.

Yolando will provide the measurement, competitive context, content recommendations and repeated tracking. AiNews still has to supply the business context and judgment. We have to decide which gaps matter to our readers, which recommendations fit our strategy and which changes are worth testing. A platform can show us that a competitor appears where we do not. It cannot decide whether competing for that particular question would make AiNews more useful to its audience.

After approximately 90 days, we will compare the results and report what changed, including if nothing changed at all. If a metric rises after we make a change, we will treat that as a result worth investigating rather than proof that one tactic caused the improvement. Models, sources, competitors and the wider information environment may all change during the same period.

Before those results can mean anything, we also need to be clear about what this measurement cannot tell us.

The most important limitation in AI discoverability I found is not specific to Yolando. It is a limitation of AI visibility measurement itself: the data can show what happened far more confidently than it can explain why.

Yolando can show that AiNews appeared in one answer and disappeared from another. It can show that Claude mentioned us without citing AiNews.com, that a competitor appeared instead or that Reddit supplied information used in the response. It can also show which pages were cited and how our reputation differed across AI platforms.

Those observations give us a pattern. They do not reveal exactly what caused it.

An AI system may choose one company over another because of stronger content, greater name recognition, better reviews, more third-party references, the sources retrieved for that particular answer or the way the question was worded. Training data and differences between the models may also play a role. The decision could involve several of those factors or something we have not identified.

If a business guesses incorrectly about what caused a weak result, it could spend months publishing the wrong content, pursuing references that do not help or trying to solve a reputation problem when the real issue lies somewhere else.

The measurement showed us that Claude did not cite AiNews.com during the sampled periods. It did not tell us why. Therefore, a measurement should not be mistaken for a diagnosis.

Martin deserves credit for being clear about that boundary. When I asked what determined whether an AI system recommended one company over another, he did not pretend Yolando had uncovered a formula that no one else could see. The platform can help businesses investigate the possibilities, track patterns and test changes. Human judgment is still required to decide what those patterns might mean and which response makes sense.

That honesty matters in an industry where businesses are already being offered confident AEO formulas and guaranteed results. Answer engine optimization, or AEO, is still developing. The systems behave differently, and the information they rely on continues to change. Businesses should be skeptical of anyone claiming that a fixed series of steps will guarantee a recommendation across ChatGPT, Claude, Gemini and Perplexity.

Still, measurement has a clear role. It gives businesses a baseline, narrows the questions worth investigating and allows them to see whether the results change after they act. Platforms like Yolando give businesses a place to start by showing how AI systems currently mention, cite, describe and recommend them. The data provides enough visibility to form hypotheses, test changes intended to improve mentions and citations and see whether the results move.

After 90 days, we will report whatever the data shows. Any explanation for those results will remain a hypothesis until we have stronger evidence.

I began this audit trying to understand how Claude could recommend AiNews more often than any other AI system we measured while never citing AiNews.com. That question remains unanswered.

The analysis did reveal why businesses cannot approach AI discovery with the same expectations they bring to traditional search. SEO has had decades to develop increasingly mature practices, tools and benchmarks. AI discovery is spread across several systems that can interpret the same question differently, rely on different sources and change their answers from one attempt to the next.

A company can make its website clear, publish useful content, build legitimate third-party authority and follow every sensible AEO practice available today. A competitor may still receive the recommendation.

That uncertainty is frustrating, especially for businesses accustomed to experts telling them which changes should improve their rankings. Customers are already asking AI systems which companies, products and sources they should trust. Businesses need some way to see whether they are participating in those decisions.

That is the role Yolando serves. It makes more of this new discovery environment visible by showing businesses where they appear, how they are described, which competitors appear alongside them and which sources contribute to the answers. The platform gives businesses a baseline they can investigate and return to as they make changes.

Yolando is offering the first 20 AiNews readers the same audit used for this article at no cost. The audit includes their own questions and competitors across ChatGPT, Gemini, Perplexity and Claude, using the same four measurements: discoverability, share of voice, citation share and reputation. It requires approximately 15 minutes of setup and a 30-minute call with Yolando. The offer will remain available until all 20 places are claimed or AiNews publishes its 90-day follow-up. AiNews will not be paid if readers participate.

AiNews readers can claim one of the 20 places through Yolando’s dedicated audit page: Run this same audit on your own domain

When no one has a definitive playbook, measurement provides a place to begin. Businesses can form hypotheses, decide which actions make sense for their customers and strategy and watch whether the results change over time.

For businesses accustomed to SEO's increasingly mature playbook, AI discoverability requires getting comfortable with something much messier: measurement without certainty.

Q: What is AI discoverability?
A: AI discoverability measures how often a business, publication or product appears in AI-generated answers to relevant questions. It is one part of broader AI visibility, which can also include share of voice, direct citation share and how positively or negatively AI systems describe the business.

Q: What does Yolando measure?
A: Yolando measures whether and how businesses appear in answers from AI platforms such as ChatGPT, Gemini, Perplexity and Claude. Its four main measurements are discoverability, share of voice, citation share and reputation. Businesses can also examine individual responses to see which competitors appeared and which websites the AI system used as sources.

Q: What did Yolando find about AiNews.com?
A: Yolando ran 25 audience-based questions repeatedly across ChatGPT, Gemini, Perplexity and Claude, producing approximately 2,700 question-and-answer observations across two sampling periods covering about six weeks. AiNews appeared in 9.1% of relevant answers overall and ranked tenth among 11 AI publications. Claude mentioned AiNews in 12.9% of relevant answers—the highest rate among the four AI systems measured—but directly cited AiNews.com zero times.

Q: How is AI discoverability different from SEO?
A: Traditional SEO generally focuses on whether a webpage appears and ranks in a list of search results. AI discoverability asks whether a company appears inside an AI-generated answer when someone requests information or a recommendation. Different AI platforms can select different companies and sources for the same question, and their answers can change when the question is asked again.

Q: Is an AI recommendation more valuable than a citation?
A: The more valuable result depends on the business and its goals. A recommendation may influence which company, product or publication a customer chooses, even when the company’s website is not cited. A direct citation can provide attribution, authority, potential referral traffic and measurable website activity, so businesses should measure both outcomes rather than treating either one as the only goal.

Q: Why do third-party websites matter in AI search?
A: AI systems may use reviews, directories, media coverage, community discussions, podcasts, social articles and other independent sources to understand where a business belongs in its industry. Yolando recorded 52,104 source references through third-party domains and 146 through AiNews.com in the measured dataset. That pattern does not prove that gaining more third-party references will improve a company’s AI visibility, but it provides a hypothesis businesses can test.

Q: How can a business improve its AI discoverability?
A: A business can clearly explain who it is, what it offers, who it serves and what problems it solves on its own website. It can answer intent-rich customer questions through useful pages and FAQs, maintain consistent business information across the web, distribute genuine expertise and strengthen legitimate third-party signals through relevant reviews, directories, media coverage, partnerships and expert references. The business should then establish a baseline, make deliberate changes and measure whether its visibility appears to change.

Q: How should a business measure AI visibility?
A: A business should use a representative set of customer questions, run them repeatedly across multiple AI platforms and establish a baseline before making changes. One answer provides an example rather than a reliable pattern because generative AI systems can produce different responses to the same question. If the questions are substantially changed, the business should establish a new baseline rather than comparing the new results directly with the old ones.

Q: Can AI visibility software explain why one company was recommended over another?
A: AI visibility software can show which businesses appeared, how they were described, which competitors appeared and which sources contributed to an answer. It cannot see inside an AI system or establish exactly why one company was selected over another. The data can reveal patterns and guide hypotheses, but it should not be mistaken for a diagnosis or proof that one optimization tactic caused a result.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing support, AEO/GEO/SEO optimization, image concept development, and editorial structuring support from ChatGPT, an AI assistant. All final editorial decisions, perspectives, and publishing choices were made by Alicia Shapiro.