Ahrefs Embraces AI, Redefining Content Creation to Combat "Slop"

Every article published by Ahrefs incorporates artificial intelligence in some capacity, a deliberate strategy to enhance, not dilute, the quality of their content. The company distinguishes between the judicious use of AI and what they term "AI slop," defining the latter as content lacking sufficient human understanding, judgment, evidence, or original contribution to warrant a reader’s attention. This distinction is crucial: AI slop fundamentally shifts the burden of effort from the creator to the reader, who is then left to discern trustworthiness, relevance, and genuine expertise.

Ahrefs’ approach to AI integration prioritizes preserving the foundational practices that made content valuable before the advent of AI, even as AI takes on a larger role in the creation process. This is achieved through a multi-faceted strategy focusing on human-led groundwork, structured decision-making, strategic time allocation, and clear ownership.

1. Prioritizing Human Work Before Drafting:

The traditional writing process, where significant time was invested in drafting, has been fundamentally altered by AI’s ability to generate prose cheaply. Ahrefs redirects this freed-up effort to the upstream stages of content creation: meticulously deciding what topics deserve coverage, how best to structure an article, and what unique contributions can be made. Before any AI is tasked with generating text, writers are expected to establish a clear premise that addresses core questions: what is the article’s objective, what is its intended audience, what is the core message, and what unique perspective or evidence will be presented?

These fundamental questions predate AI and remain essential. AI can assist in answering them by mapping search engine results pages (SERPs), challenging proposed angles, identifying counterarguments, or highlighting missing evidence. However, the ultimate decision-making authority remains with the human creator.

Crucially, AI requires high-quality input to produce unique output. Providing it with the same internet-accessible data as everyone else will inevitably lead to generic content. To overcome this, Ahrefs emphasizes the incorporation of raw materials that AI cannot generate independently. This includes insights from interviews, internal company knowledge, proprietary data, real-world demonstrations, documented failed experiments, and specific examples drawn from the team’s practical work. For instance, Mateusz, a colleague at Ahrefs, developed a "Source of Truth" application in Letaido, a searchable library designed to house the factual information his team and AI agents rely on for content creation. This repository categorizes information into facts and statistics, explanations, product details, and how-to guides.

A practical shift for many writers at Ahrefs involves a move from typing to speaking. When typing, there’s a natural tendency to edit and tidy up thoughts in real-time, often inadvertently removing the very elements that lend depth and nuance to content—uncertainty, caveats, opinions, and nascent connections. To combat this, writers are encouraged to dictate their ideas using tools like Wispr Flow. This process, akin to recording an unedited podcast episode with oneself, allows for the uninhibited offloading of thoughts, providing AI with richer, more unrefined raw material to work with. Alternatively, AI can be employed to interview the writer, identify knowledge gaps, and extract potential claims before proposing an outline. In this scenario, AI’s role shifts from invention to the organization and interrogation of human-supplied substance.

2. Establishing Decision Points and Judgment Stages:

AI’s proficiency in embedding decisions within polished prose can obscure the creative process. A single, comprehensive prompt can simultaneously dictate research, angle, structure, claims, examples, and tone. By the time the output is generated, these choices appear settled and definitive.

To counteract this, Ahrefs advocates for a pipeline approach that mirrors a human editorial workflow, rather than producing a single, final output file. Ryan’s content engineering pipeline, for example, separates research, identification of content gaps, outline generation, and drafting into distinct stages. This allows for inspection at any point, enabling corrections to flawed outputs or instructions and restarting from the last acceptable stage.

This principle can be applied even without complex pipelines. The key is to break down the work into discrete stages, inserting a human judgment point between each. These "gates" require the writer to assess the output and make decisions: requesting further research, deleting sections, altering the angle, or even abandoning the article. This structured approach is vital because the ease of AI generation can create a "sunk cost fallacy"—the tendency to want to improve what’s already produced rather than question its fundamental purpose or existence, especially when AI can generate such output rapidly.

Even when using AI for a single article without a formal pipeline, the work should be segmented. Initial brainstorming should precede any drafting. Once the angle is refined through iterative exchanges with AI, the process moves to the next stage, such as generating an outline. This outline is then meticulously reworked with AI until satisfactory, before proceeding to the subsequent stage. The goal is to facilitate better decision-making, not simply to increase word count. Without these deliberate pauses for judgment, AI can inadvertently turn a writer into an editor before they’ve had the chance to fully develop their ideas.

3. Allocating Saved Time to Enhance Content Quality:

Ahrefs differentiates itself by focusing on what can be achieved with the time saved by AI, rather than solely on increasing publication volume. While the company acknowledges that AI is used to automate tedious tasks and save time—enabling more frequent publishing and updating of content, as demonstrated by the Data Refresh Hub which automates monthly updates for 12 datasets—this efficiency is not exclusively channeled into producing more articles.

The primary constraint on content creation at Ahrefs has often been the investment required for elements that enhance an article but are resource-intensive. This includes conducting simple data analyses (which might require a data scientist), developing free tools (needing developer time), creating interactive articles (requiring design and engineering expertise), or undertaking large-scale research projects that are too manual for a single content marketer.

AI significantly lowers these barriers. Writers can now analyze datasets, prototype tools, build interactive elements, improve a post’s user interface, or automate aspects of research projects, often without needing immediate specialist intervention. While perfection and expert involvement are still necessary for high-stakes endeavors, the threshold for experimentation is considerably lower. This allows for the transformation of basic content elements, like a TablePress embed, into richer interfaces for exploring complex data, as seen in their analysis of the most-cited domains in AI Mode. It also empowers content marketers to create resources like quizzes, free tools, or data visualizations without waiting for limited developer availability, exemplified by their free LLMs.txt generator.

The core distinction lies in whether AI is used to merely reduce the effort behind each article or to undertake tasks that were previously financially or logistically unfeasible. The crucial question is not how many more articles AI enables, but rather what new, valuable elements can now be incorporated into an article that were once impossible. While the capability to scale is important, restraint is also necessary. The aim is not to transform every writer into a content factory, but to expand the scope of what each individual can achieve without compromising the standards for published content.

4. Assigning Ownership and a Second Human Review:

Even with extensive AI involvement, a specific individual must ultimately own the final output. The person managing the AI workflow must possess sufficient subject matter expertise to validate claims, correct misinformation, trace the origins of evidence, and ultimately decide whether they are comfortable attaching their name to the published work.

This is why Ryan, despite his efficient pipeline, does not publish hundreds of articles overnight. As Director of Content Marketing, he serves as the editor for all content appearing on the Ahrefs blog, personally reviewing every word. His position as an industry thought leader also equips him to critically assess articles, challenge premises, question evidence, identify generic sections, and determine if the article truly fulfills its promise to the reader.

The presence of a human in the loop is only meaningful if that human actively exercises critical judgment. They require the knowledge, authority, and willingness to reject unsatisfactory output. Therefore, even for articles that are heavily AI-generated or AI-assisted, Ahrefs ensures sufficient human hours—from both the writer and the editor—are invested to guarantee the content meets publication standards.

Conclusion:

AI-generated content is not inherently "slop." Studies, including Ahrefs’ own research on Google’s treatment of AI content, indicate that heavily AI-generated pages can achieve high rankings. However, an increase in AI usage has often correlated with a decline in search performance. This is not due to a direct Google penalty against AI content, but rather because companies frequently leverage AI to bypass essential, labor-intensive steps in the creation process, resulting in more average, less impactful content.

The fundamental issue, therefore, is not AI itself, but the abdication of responsibility. AI excels at making execution cheap, which should ideally free up human resources for the critical judgment, evidence gathering, experience sharing, and idea generation that truly make an article compelling and valuable. The pertinent question is not the quantity or origin of the words, but whether a human has genuinely understood, judged, verified, and ultimately stood behind the content that goes live. If no individual truly owns the outcome, it risks becoming "slop."