.png){kind=logo}
On July 10 2025 screenshots circulated on Twitter (sorry X), and LinkedIn showing a new column hiding on the far right in Ahrefs’ “Top Pages” report under the "Organic Search" menu, labelled AI Content Level. The metric gives every crawled URL a percentage score that estimates how much of its visible text was generated by large-language-model software. A couple of days earlier users had spotted the same figure tucked inside Page Inspect, AI Detector, but surfacing it in a domain-wide column turned a hidden tool into front-page data.
The move built on an earlier step. Ahrefs’ public changelog records June 4 2025 as the date the AI Detector tab first appeared in Page Inspect. That launch required page-by-page checks, while the July update scaled detection to millions of URLs in one click, handing SEO expert a new venue of insights on the automation trend.
Ahrefs says it blends statistical token analysis with neural pattern recognition and can identify specific models such as GPT-4o and Llama. When you click on the percentage number, a separate side window pops up, and color pie chart appears together with the breakdown of the content it's scanning. Green for human text, amber for AI. The percentage number itself has three colors, depending on the amount. Green for low, amber for medium, and red for heavy presence. It also suggests what type of LLM is most likely in question, and if the LLM is unknown, it just show "Unidentified LLM".
Because the detector lives inside Site Explorer, we hope that users could filter pages by AI share in the future, isolate subfolders and export the numbers. That workflow matters for competitive audits, risk reviews before link outreach and internal quality checks.
A clear illustration comes from AI companionship services, whose landing pages and character profiles tend to get generated by on the fly by large-language models. When Ahrefs crawls these domains, the AI Content Level column will probably read high percetange, confirming that the detector can quantify automation even in niches where LLM-driven text is fundamental to the product rather than a supplemental tool.
For site operators, that single metric becomes a simple steering wheel, if an AI porn chat platform decides to add expert-written safety guidelines or relationship advice, it can watch the AI percentage shift and evaluate whether a more balanced authorship mix changes session length, user trust signals, or backlink growth. Likewise, teams in finance, travel, or news can benchmark their own AI adoption against these fully automated examples and set internal thresholds with real numbers instead of guesswork.
Consultants immediately shared screenshots of pages scoring 80–90 % AI, and a lively Reddit discussion noted several false positives. Within days, agencies were already piping the new ‘AI Content Level’ column into Looker Studio dashboards (usage numbers and internal Ahrefs comments on entropy-related false positives have not been published).
With a fresh gauge available, the obvious question is whether a high score predicts ranking losses or is it there just "for fun." One week before the rollout, Ahrefs published a study of 600 000 pages that found almost no correlation between AI share and average SERP position. The gist of it is that 86,5 % of the top-20 results (not specifically the top three) showed some AI content.
Most of the SERP guidance supports that finding, where a February 2023 Search Central post notes that automation is acceptable when content is helpful, while SISTRIX’s March 2024 core-update analysis notes losses for "called content abuse" sites but no blanket drop tied solely to AI usage. Bottom line, no dataset yet isolates "high AI ratio" as a ranking risk.
No detector is perfect, and that's confirmed by multiple blind tests done by independent consultants. They roughly measured a 12% false-positive rate on well-edited human text. The current model is tuned for English and major European languages, detection for Thai or Hebrew remains less certain, while Ahrefs plans to expand its ground-truth set to two million labelled sentences and support fifty languages by late 2025.
This new metric raises the bar for transparency, so editors can no longer plead ignorance when a site jumps from 10 to 70% of hosting AI generated content. Brands may also set their own thresholds where a medical portal might allow 20% automation under clinician review, while a coupon aggregator accepts 60%. Over time, high AI ratios could influence reputation scores used in brand-safety audits or affiliate programmes.
Ahrefs is first among major suites to expose a domain-wide percentage rather than a single-page snapshot, while our guess is that rival platforms are already racing to integrate similar tools. Standardization could follow, perhaps even based on shared open-source detectors.
We predict that publishers will counter with paraphrasing layers or hybrid drafting, creating an arms race reminiscent of the early Panda era.
Benchmark from Semrush from May 2025 reached a similar conclusion - 57% of AI-heavy articles and 58 % of human-written ones appeared in the Top 10, a gap too small to be statistically meaningful. On the opposite side, a March 2024 spam-and-core update audit using Originality.ai flagged a different pattern where the hundreds of websites that some search engine manually de-indexed all showed some AI output and half were 90–100%, suggesting that abuse at scale, and not mere automation, will triggers hard penalties.
Link acquisition tells a more nuanced story. Ahrefs’ own backlink models still show that the #1 organic result carries 38x more referring domains than positions 2–10, a ratio unchanged in 2025 despite the AI surge.
Editorial surveys summarized by Adsy show that editors reject content identified as “mostly AI” roughly 24% more often than equivalent human-polished submissions. Taken together, the numbers imply that search visibility per se is not harmed by responsible AI use, but AI-sounding prose can dampen natural-link growth, an indirect ranking handicap over time.
AI Content Level is just a surface indicator because it says nothing about factual accuracy, originality or reader engagement. A human can still produce thin copy, and an AI can draft a brilliant base that an expert elevates. SEOs therefore need complementary checks such as EEAT reviews and user metrics. Mosty importantly, the score also ignores multimedia and assets attached to the articles. A recipe post might contain an AI-generated headnote, but feature original photography that boosts user's on-page time.
Without a doubt, this new "AI Content Level" column in Ahrefs will add a new axis to SEO reporting. For strategists it overlaps neatly with traffic and backlink data, while for search engines it offers raw material to test their quality benchmarks against with.
Most evidence to date shows no automatic penalty for AI text, but if writers continue to deliver clear, accurate information, automation should remain a neutral tool. What the feature will certainly do is expose reckless shortcuts, low effort and misinformation, and that alone may push our world wide web toward healthier, more accountable publishing.
