Last year a client reduced their content budget by half by starting to use AI to produce blog
posts. Traffic remained steady for a few months before the March 2026 core update was
released, after which their organic visibility fell by more than half within less than three weeks.
The client’s content had not been altered in any way — it was Google that had been able to
detect the pattern of what they were publishing.
The kind of discussion that agencies should be initiating beforehand, not after the fact, is
necessary since the real policy in this case is more complex than simply saying “AI-generated
content is acceptable” or “AI-generated content is penalized”—and making an error about
this nuance would be an expensive mistake for a client.
Since February 2023 Google has maintained a consistent position on AI-generated content by issuing guidance that made it clear its emphasis is on the quality of the content rather than on the method used to produce it (Google Search Central, 2023). This same sentence remains Google’ s official policy in 2026. The change has been in the ability to enforce this policy: the March 2024 core update incorporated the Helpful Content system directly into the main ranking factors and introduced the “scaled content abuse” spam policy, so that thin AI content is not merely at risk during a regular update but is now constantly evaluated (Layer3Labs, 2026). The March 2026 spam update was completed in under 20 hours, the quickest ever recorded by Google, which shows how efficiently SpamBrain can now detect mass-production patterns regardless of whether the text was produced by a human or by an AI (theStacc, 2026).
The impact of that March 2026 update was clear: AI content farms saw their traffic drop by 60 to 80 per cent, while websites that published original data and first-hand analysis gained about 22 per cent more visibility (Kozec.ai, 2026).
E-E-A-T — which stands for Experience, Expertise, Authoritativeness and Trustworthiness — is derived from the Google Search Quality Rater Guidelines, and it’ s important to be clear about what it really is: it is not a box ticked as a ranking factor, but rather a framework that is used by human quality raters to assess content, the judgments they make in turn being used to train the algorithms that do influence rankings (OptiSEOn, 2026). The January 2025 update to these guidelines was the first time that it included specific language defining and referring
to generative AI content (theStacc, 2026).
The change in practice that agencies are observing from the data shows that 73% of the top-ranking YMYL (Your Money or Your Life) pages—those covering health, finance, legal, and safety topics—now feature detailed author credentials, compared with 58% prior to the 2026 update cycle (Kozec.ai, 2026). For any WordPress site that publishes content in a regulated or high-stakes field, author bios together with proven credentials have moved from being a optional feature to something that is now a necessary requirement.
The situation raises the stakes even more in that E-E-A-T signals are no longer merely a factor in Google’ s ranking process. About 96% of the citations found within Google AI Overviews are来自 sources that have strong E-E-A-T signals (Launchcodex, 2026), and it is the same trust signals that satisfy Google’ s human quality raters that ChatGPT and Perplexity use when choosing which sources to cite in their generated answers (OptiSEOn, 2026). If a website neglects E-E-A-T it is not only risking a drop in traffic during a core update — it is also risking being invisible in the AI-answer section of search which is now where users are more and more likely to end up.
The point that is often overlooked in discussions with clients is that Google does not impose penalties on content just because it has been generated by an AI; instead, the search engine penalizes poor-quality content that is produced in large quantities with the aim of manipulating rankings, a type of behaviour which AI simply makes much easier to carry out, not one that constitutes a new kind of violation (TheHumanizeAI, 2026). Several documented case studies have demonstrated that content assisted by AI has achieved a position on the first page for competitive search terms — provided that adequate editorial supervision is in place (TheHumanizeAI, 2026). Together with other major publishers, Bankrate has publicly admitted to using AI-assisted content creation since 2023 and has received no penalty as a result (TheHumanizeAI, 2026).
The kind of failure that is detected involves websites publishing between 50 and 500 AI-generated articles each day with the same structure and no editorial oversight, frequently increasing their output from zero to thousands of pages within weeks (theStacc, 2026).
For agencies managing client blogs or content calendars, this translates into a short, enforceable process rather than a philosophical debate about whether to use AI at all:
Include some genuine first-hand experience, for example one or two specific details that a general AI output cannot come up with on its own—such as a client outcome, a pattern from a support ticket, a screenshot, or a particular figure from my own work (MainTouch, 2026).
Get rid of generic AI language. Words such as “detailed,” “complex,” “showcasing,” and “seamlessly” are typical indicators of unedited AI output and should be manually checked for (MainTouch, 2026).
Make sure that all statistics are verified before they are published. Check that the sources being used are up-to-date and not taken from the outdated training data of an AI model, and give them proper citations.
For topics that are adjacent to YMYL, provide the name of the actual author together with their visible qualifications; rather than using a general ‘ Team’ byline, this should in particular be the case with content relating to health, finance, legal, or safety matters.
Human editorial review should be required before content is published, not after a drop in ranking—this is the one factor that distinguishes penalized “content farm” practices from AI-assisted content that performs normally.
It is not the use of AI content tools that constitutes a liability; rather, it is the publication of a large volume of content that is not managed, not edited, and not reviewed—which was also the case with human-written mass content a decade ago (Rankability, 2026). The agencies that have the most defensible stance as the core updates of 2026 continue are not those who are refusing to use AI tools, but those which have developed a real editorial process around them.
Not sureif a client’s content workflowwould survive the next core update? Build with us—we’ l auditthe E-E-A-T signals and editorial process, notjustthe words on the page.
L + P
admin@linesandpixels.team
We had a franchise client who was looking after 14 separate WordPress installations — one for each location — each with its own logins, its own versions of plugins, and its own gradually changing design. To update a single header image meant having to log in 14 times. That isn’t a website issue. That’s an […]
Last month a client asked us, “Should we be optimizing for Alexa?” That is the wrong question; the proper one to ask is whether your content is arranged so as to provide the answer, since it is precisely that which decides whether a voice assistant reads out your business or goes straight to a competitor. […]
AI Overviews Are Eating Your Clicks: What Answer Engine Optimization Means for WordPress Sites in 2026 The traffic dashboard prepared by the client told a better story than we could: the rankings were staying the same, impressions remained level, and clicks were slowly decreasing from month to month. Nothing was wrong—the fact was that Google […]
Elevate your business with a custom-designed website that captivates and converts. Let our expert team craft a digital experience perfectly tailored to your brand's needs.