Feb 27,2026 
If your content strategy in 2026 relies heavily on AI-generated text, you should probably rethink that plan. In a newly published paper, Google researchers demonstrated a hybrid detection method that dramatically improves the ability to identify AI-generated content. Even in the scenarios where detection has historically been weakest.
One of the researchers specifically heads up the detection of LLM-generated content at Google DeepMind. If you were wondering if Google cares about websites using AI generated content, just look at the many instances of websites losing lots of traffic as the result of a heavy output of AI generated content these days. If Google doesn’t like (or trust) the content on your site, that could mean a drop in traffic from their search engine and reduced visibility on answer engines.
Google’s first page of organic search results is still very clearly an important part of answer engine visibility. ChatGPT, Gemini, and Perplexity all rely heavily on Google search results when constructing their answers and choosing which brands to cite. Smart marketers are thinking about optimizing for Google and answer engines in 2026.
How AI Detection Works Today
Most AI detection is done in two ways. One is watermarking, where AI providers embed invisible statistical patterns into the text their models generate. The second way uses classifiers, which are models trained to recognize the subtle stylistic and statistical patterns that distinguish machine-written text from human-written text.
Both approaches have weaknesses on their own. Watermarks depend on entropy, which is how much creative freedom the model has when generating a response. For open-ended prompts (“Write me a blog post about hiking in Yosemite”), entropy is high and watermarks are easy to detect. But for constrained, factual prompts (“What is the boiling point of water?”), the model doesn’t have many word choices, so the watermark signal gets buried. This is much harder for AI to detect.
Classifiers, on the other hand, tend to be more consistent but at times will flag human-written content incorrectly. If you’ve tools like Originality, or any other AI content detection tool, you have probably experienced this. Incorrect flagging of content is a problem Google (and other search engines) do not want.
Combining Both Approaches Works Best
What this new research shows is that combining watermark detectors with classifiers creates a hybrid system that’s far more powerful (and more accurate) than either one alone. In their experiments, a logistic regression model trained on both watermark scores and scores from a RoBERTa-based classifier boosted detection accuracy from 75% to over 95% on the hardest-to-detect content, the bottom 20% of prompts where watermarking alone struggles most. That’s a 20-percentage-point improvement on exactly the cases that used to slip through most often.
The hybrid approach also held up well when only short text snippets were available and when the text had been lightly edited or paraphrased. Both of which are things that people commonly do to try to disguise AI-generated content.
How Will Google Deal with AI Generated Content?
Expect Google to increasingly detect, devalue, and algorithmically penalize AI-generated content pages. Not all AI-generated content, but content that lacks value, originality and quality is in danger. Webpages that just “AI content spin” and say essentially the same thing as what already exists does not provide value. When AI content is generated in bulk, expect Google to target not just individual pages, but sites as a whole.
Google has a long, well-documented history of cracking down on tactics designed to manipulate search rankings. As someone who has been studying Google’s algorithms for the past 20 years, I’ve seen way too many instances of Google cracking down on tactics to try and manipulate the algo. Google has fought all of them, and often aggressively. And if Google thinks your site is trying to game their algorithm, they do not easily forgive. Penalties are often severe and long-lasting, sometimes lasting for multiple years. Some sites never fully recover.
The writing is on the wall. If your site is publishing bulk AI-generated content with minimal human oversight, you are taking on increasing risk with every Google algorithm update.
Are “Double” Penalties Coming?
Are we going to see websites get penalized in Google AND in answer engines like Google Gemini and ChatGPT? I wouldn’t be surprised. You’ll already see a drop in visibility on answer engines if you’re Google rankings get wiped out, so unofficially, yes. The answer engines are pretty weak right now when it comes to defending against LLM spam , but it won’t always be like this. Answer engines will most certainly get better at detecting manipulation tactics. We do expect to see website get penalized in both Google and answer engines in the future.
What You Should Do Now If You’re Using AI to Write Content for Your Website?
We are advising those who use AI to generate content to have human editors to evaluate, clean up, QC and improve the content generated by LLMs. Here’s what smart brands and agencies should be doing:
Audit your existing content: If you’ve published a significant amount of AI-generated content in the past, you should review that asap. Is it adding genuine value? Does it reflect real expertise? Or could it be flagged as low-quality AI filler by an increasingly capable detection system?
Treat AI as a starting point, not the finished product: Don’t copy and paste straight from an LLM. Use AI to accelerate your workflow, absolutely, but make sure a knowledgeable human is editing the final content. Remove content that reads like AI, since not only is that bad for online visibility, but it’s often bad for converting website visitors.
Choose the Right Tools: Most AI generated content tools sound fine, but many right now have too many obvious signs that it was AI generated. Especially ones that don’t have a deep understanding on Google’s algorithms and how Google determines quality. Ultimately, content generated from these tools required more editing work to get them into an acceptable final form.
Focus on building ‘citation-worthy” content. Citation-worthy content is structured, machine-legible, extractable, and semantically unambiguous. Answer engines do not want essays. They want extractable blocks. Statistics with sources, dates, references, tables, pricing, percentages, etc.
Use “Atomic Answers” in your content: Atomic Answers are a great way to improve visibility on answer engines by increasing your chance of being cited, since they are a high-density, self-contained block of information engineered to be “mathematically irresistible” to an AI model. Read more about it here.
Check your Google analytics: Has your traffic been going up or down? Often times, sites using AI generated content in bulk see a jump in traffic…but then see that traffic fall off a cliff. Don’t just check your overall traffic, check individual pages as well to dive deeper in to the data.
Monitor your visibility across both Google and answer engines. You can’t manage what you can’t measure. Track how your brand shows up on ChatGPT, Gemini, Perplexity, Grok, and Google AI Overviews and understand which citation sources the models are pulling from. If your visibility is declining, it could be a sign that your content strategy needs fixing. You also should be tracking your rankings on Google. Rankings are not nearly as important as traffic, but they certainly are an indicator if Google likes (or does not like) your content.
TL;DR
AI content detection is advancing fast. This research by members of Google’s DeepMind shows that hybrid detection methods can already catch AI-generated text with high accuracy. As these techniques get adopted, the risk of relying on unedited AI output keeps climbing.
For brands and agencies focused on search visibility, don’t generate AI content in bulk. When you do use AI to help you generate content, make sure you edit it before you post it. Your Google rankings are directly correlated with answer engine visibility, so what Google thinks about your content matters. Anything that threatens your Google rankings threatens your AEO/GEO performance too. The brands that invest in authentic, expert-driven content now will be the ones that maintain visibility across both traditional search and the answer engine landscape as it continues to evolve.
At Hoverboard, we help brands and agencies track and improve their visibility across all the major answer engines. Knowing where you stand today is the first step to making sure you stay visible tomorrow.
Want to see how your brand shows up on answer engines? Request a demo and find out.