Jarrett Sidaway
CEO & Co-Founder, FetchRight
Over the last two years, we have become accustomed to a strange kind of answer: one that sounds confident, reads smoothly, and is almost correct. It might miss a key nuance, blur a time frame, oversimplify a causal link, or blend two related but distinct facts. To a casual reader, it feels plausible. To someone who knows the subject, it feels off.
This is the signature of an AI system operating without a solid anchor. It has seen vast amounts of text, learned patterns, and become incredibly good at synthesizing language. But it has not been given a reliable way to distinguish what is definitively true from what is merely likely, or to prioritize the judgment of domain experts over the noise of the open web.
For consumers, this ambiguity translates into eroded trust. For publishers, it creates a reputational and strategic problem: their work is increasingly woven into AI answers, but not always in ways that preserve accuracy, context, or attribution. For AI platforms, it undermines the very promise they are making to users: that complex questions can be met with reliable answers.
At the center of all three concerns is the same issue: accuracy cannot be treated as an emergent property of scale. It has to be grounded in authoritative sources, and those sources are overwhelmingly publishers.
One of the great achievements of modern AI is fluency. Models can generate paragraph after paragraph of coherent, well-structured text. They can adapt tone, mimic genres, and respond conversationally. For product teams, this has been a breakthrough: users now feel like they are interacting with something that understands them.
But fluency is not the same as correctness. A model that has been trained on a mixture of high- and low-quality sources can easily learn to sound authoritative without actually being anchored in authoritative content. When it does not have enough signal, it interpolates. It fills in gaps. It infers.
From the system's perspective, this is rational behavior. It is designed to produce the "most likely" continuation based on its training. From a user's perspective, however, it can be dangerous. A health answer that is 80% correct is not "good enough." A misinterpreted policy detail can meaningfully change understanding. A misattributed quote can damage reputations.
This is where publishers matter most. Their value lies not just in the information they publish, but in the discipline behind it: the editorial standards, verification processes, corrections, and accountability mechanisms that distinguish professional journalism, expert analysis, and curated reference from unvetted content.
If AI does not know how to recognize, prioritize, and preserve that discipline, it will continue to produce answers that sound right while missing what actually is right.
The problem is not that AI systems reject expertise. It is that, in the current environment, they are not given consistent, structured ways to identify and rely on it.
Publisher content often reaches models through generic crawling and large-scale scraping. The model sees the text, but not the metadata that would tell it "this comes from a brand that has a decades-long track record of accurate reporting in this domain," or "this explanation has been carefully peer-reviewed." It sees the words, but not the editorial intent.
Even when content is highly reliable, the lack of structure can cause it to be misused. A complex article might contain a speculative paragraph, a historical comparison, and a clear, current fact. To a human editor, the distinction is obvious. To a model ingesting text without guidance, those lines can blur. Later, when the system assembles an answer, it may grab the wrong sentence, or fail to bring along the context that makes a claim accurate rather than misleading.
Add in the fact that much of what models see is outdated or disconnected from current conditions, and it becomes clear why accuracy is fragile: the system is assembling answers from pieces it cannot fully evaluate.
What is missing is not more data. It is clearer signals, stronger structure, and explicit guidance from the people who actually know what is true and what is not.
Publishers are uniquely positioned to supply what AI systems lack: grounded, curated, context-aware knowledge about the topics they cover. The value they create is not simply the accumulation of articles, but the ongoing maintenance of a living corpus of expertise.
Think about a major publisher in any domain — news, finance, health, technology, or consumer reviews. Behind every piece of content is a set of practices: sourcing, fact-checking, editorial oversight, subject-matter consultation, and updates when facts change. These practices collectively produce something models cannot fabricate: ground truth.
In an answer-driven world, that ground truth is more important than ever. When users ask AI systems questions about elections, medical treatments, financial decisions, or safety issues, they are implicitly relying on that truth. They assume the system has access to the best available information and knows how to distinguish it from everything else.
That assumption is only justified if AI systems are able to recognize, trust, and prioritize publisher content — not in an abstract way, but in the mechanics of how they retrieve and construct answers. And that, in turn, requires publishers to present their content in ways that make its authority visible and actionable.
Today, most publisher content is optimized for human readers and for traditional search engines. Pages are laid out for comprehension and engagement. Metadata helps search engines understand topics, authors, and relationships. But AI answer systems need more.
They need to know which sentences represent verified facts, which sections offer interpretation, which paragraphs are summarizing prior events, and which pieces of context are essential to avoid distortion. They need signals about recency, levels of certainty, and whether a publisher considers a piece to be a canonical reference on a topic or a snapshot of a moment.
This is what it means to structure content for AI: not to rewrite articles in jargon, but to expose the editorial logic behind them in machine-readable ways. It means defining, explicitly, the parts of your content that should anchor answers, and the conditions under which they should be used.
If publishers do not supply such structure, AI systems will continue to infer. They will guess which parts matter. Sometimes they will get it right; often they will not. The result is a steady stream of "almost correct" answers that normalize approximation as acceptable.
If publishers do supply it, AI systems can move from guessing to grounding. They can build answers atop the same facts, distinctions, and caveats that the publisher uses in its own work.
The role of FetchRight in this context is straightforward: it creates the infrastructure that allows AI platforms to learn from publishers on the publishers' terms, in ways that preserve accuracy, context, and rights.
Rather than letting crawlers scrape unstructured pages indiscriminately, FetchRight enables publishers to expose structured, rights-cleared representations of their content. These representations can be tailored to different AI use cases — real-time question answering, summarization, citation, or model training — and they carry with them the signals AI systems need to rely on them confidently.
This is not just about licensing, although licensing is important. It is about ensuring that when an AI platform claims to be using authoritative sources, that claim is technically true. It is about putting publisher-defined insights at the center of the answer construction process instead of at the periphery.
By operating at the edge of the publisher's domain, FetchRight can enforce these rules uniformly. It can ensure that legitimate AI agents receive the structured content they need, while unmanaged scraping is constrained or refused. It can provide observability into how content is being accessed, and by whom, so that accuracy is not just a hope, but a design choice.
In effect, FetchRight gives AI systems a reliable way to connect to the people who actually know what they are talking about.
For publishers, the stakes are not limited to philosophical questions about truth. Accuracy is quickly becoming a business issue.
If AI answers are routinely wrong or misleading, audiences lose trust in the entire information environment. That erosion of trust does not stop at the model. It eventually touches the brands whose names are associated with topics, whether or not they were responsible for the error. Publishers cannot afford a world where their expertise is diluted into a pool of approximations, and they certainly cannot afford a world where their content is widely used to support answers they would never stand behind.
For AI platforms, accuracy is also a strategic differentiator. As users become more sophisticated, they will gravitate toward systems they believe are grounded in reliable sources. Regulatory and legal pressure will increase around demonstrable harm from incorrect answers. The platforms that can credibly say "we are anchored in verified publisher content" will enjoy a tangible advantage.
The only way to get there is through collaboration built on structure: systems that let publishers define and enforce how their content is used, and let AI products reliably ingest and rely on that content.
We are entering a phase where AI will mediate an increasing share of everyday questions, decisions, and opinions. In such a world, "close enough" is not good enough. The entire ecosystem — consumers, publishers, and AI platforms — depends on a shift from approximation to accuracy.
Publishers are the natural anchors of that shift. They already produce the most reliable content; what has been missing is the infrastructure to ensure AI systems can recognize, respect, and correctly use it. FetchRight exists to provide that infrastructure: to connect AI systems directly to structured, rights-cleared publisher expertise, so that what users see in an answer reflects the best of what publishers know.
When accuracy becomes the standard, not the exception, trust follows. And in a future defined by AI-mediated information, trust will be the most valuable asset any publisher or platform can possess.
Artificial intelligence is reshaping how people seek and consume information. But AI runs on trusted data, and trusted data originates with publishers.
FetchRight and Peek-Then-Pay let LLMs ask publishers directly for pre-filtered, attribution-ready context — cutting duplicate embedding work and lowering costs for both sides.
The audience has migrated to AI-driven discovery channels. Publishers who structure and govern what they contribute can become essential partners in the AI economy.