Seventy three percent. That's how often people in a UC San Diego study chatted with a real person and an AI side by side, and decided the AI was the human.
Not sometimes. Most of the time. Significantly more often than they picked the actual person.
Now picture that same AI drafting your client's next op-ed. It sounds confident. It sounds polished. It sounds exactly like someone who knows what they're talking about. And somewhere in paragraph three, there's a statistic it made up.
Would you catch it?
Google is betting a lot of people won't. On October 1, it updated its official guidance to say that having a human manually fact-checking AI content before publishing is "critical."
Here's the uncomfortable truth we need to reckon with: AI has become extraordinarily good at sounding right. It has not become equally good at being right. For anyone whose name, or whose client's name, sits at the top of a piece of content, that gap is where the risk lives.
What Is the Turing Test? Alan Turing's 1950 Imitation Game Explained
In 1950, Alan Turing (often called the father of computer science) proposed a thought experiment he called "the imitation game." Today we know it as the Turing test.
The setup is simple. A person (the interrogator) has text conversations with two hidden parties at the same time. One is a human. One is a machine. If the interrogator can't reliably tell which is which, the machine passes.
Notice what the test does not ask. It doesn't ask whether the machine is conscious. It doesn't ask whether it understands what it's saying. And it doesn't ask whether anything it says is true.
The Turing test measures one thing: whether a machine can convincingly imitate human conversation.
This nuance matters, because a lot of the public conversation about AI blurs the reality. When people hear "AI passed the Turing test," many assume it means AI is now thinking, feeling, or knowing. It means none of those things. It means AI got very, very good at seeming human.
How AI Passed the Turing Test: Inside the UC San Diego GPT-4.5 Study
In 2025, researchers Cameron Jones and Benjamin Bergen at UC San Diego ran their own version of Turing's game, assessing 4 different AI systems. Nearly 500 people took part. Each interrogator chatted with a human and an AI at the same time, then picked which one was the person.
When GPT-4.5 was told to adopt a humanlike persona, interrogators picked it as the human 73% of the time. That's significantly more often than they picked the actual human. The study, first released as a preprint in March 2025, was published in the Proceedings of the National Academy of Sciences in 2026. The authors call it the first empirical evidence that an AI system passes a standard three-party Turing test.
But the most revealing part is how it won.
The AI didn't win by being smarter. According to Bergen, the models succeeded partly because they made mistakes the way a human would. They nailed tone, humor, and casual fallibility. Bergen said the test is now increasingly measuring human likeness rather than intelligence.
Jones put it even more bluntly: for the models, the Turing test is "a game about lying."
Communications professionals should read that again. The skill that let AI pass was the ability to sound convincingly human, not the ability to be accurate. Those are two completely different skills. And the first one can easily mask the absence of the second.
Public Trust in AI Is Falling: What Pew Research Data Shows
Remember when AI felt like magic? Ask it anything, get a polished answer in seconds. For a while, the main reaction was awe.
That sentiment has changed, and the data shows it.
In 2021, 37% of Americans told Pew Research Center they were more concerned than excited about AI in daily life. By June 2025, that number had climbed to 50%. Only 10% now say they're more excited than concerned (Pew Research Center).
Another Pew Pew survey found that 76% of Americans say it's extremely or very important to be able to tell whether content was made by AI or by a person. Yet 53% aren't confident they could actually spot the difference (MarketingProfs summary of Pew data).
Put those numbers next to the Turing test result and the picture comes into focus. People want to know what's human and what's machine. The machines are now better at hiding it than ever. And audiences increasingly assume that if something might be AI, it is probably wrong.
In a nutshell, it isn't really about technology. It's about trust. Nobody likes finding out that the confident voice they believed was confidently wrong.
Google's New AI Content Guidelines: Fact-Check Before You Publish
On October 1, 2026, Google updated its official guidance on using generative AI content. They added three sentences. It doesn't sound like much, but it makes Google’s stance crystal clear.
First, Google explains that generative models don't retrieve facts. They predict a likely sequence of words based on their training data. Second, because of that, AI outputs can contain inaccuracies, also known as hallucinations. Third, and this is the line that got everyone's attention, it is "critical to manually factcheck and review all AI-generated content" for accuracy and trustworthiness before publishing.
Google doesn't use the word "critical" lightly in its documentation. And it didn't stop at article text. The review also covers titles, meta descriptions, structured data, and image alt text.
One day later, Google updated its guidance on creating helpful content, naming four qualities its raters look for: effort, originality, talent or skill, and accuracy.
To be clear, Google isn't anti-AI. The same guidance says AI can be genuinely useful for researching a topic and adding structure to original content. The warning is about publishing content nobody checked.
Now connect the dots. Google's explanation of why AI gets things wrong is the Turing test story in a single sentence. These systems are built to produce the most plausible next word. Plausible is what fooled 73% of interrogators. Plausible is also a publishing risk.
What Google's AI Guidance Means for PR and Communications Professionals
If you work in communications or PR, you're probably already using AI somewhere in your workflow. A first draft of a press release. Background research for an executive op-ed. A starting point for a ghostwritten LinkedIn article. That's smart. It saves hours.
But here's the catch. The better AI gets at sounding human, the harder its mistakes are to spot. A clunky sentence waves a red flag. A fluent, confident, beautifully structured sentence with a made-up statistic in the middle doesn't throw a red flag. It just sits there, looking credible, until someone else finds it. And when they do, nobody blames the AI.
That's why the human layer isn't a nice-to-have anymore. It's the whole job. Google has now made that official. The question for every communicator is simple: what does your review process actually look like, and would it catch an error that sounds completely right?
A few habits make a real difference:
- Check every number against its original source. Not the article that quoted it. The study, the survey, the report itself.
- Treat fluency as a warning sign, not a green light. The smoother a claim reads, the more it deserves a second look.
- Review the invisible parts. Headlines, meta descriptions, and image alt text show up in search results too. Google says they need the same scrutiny.
- Make verification repeatable. A process that depends on remembering to double-check will eventually fail on a busy day.
If you’re not sure where to start to get your content review workflow into shape, our content review guide is for you.
Credibility, Not Imitation, Is the Next Test for AI Content
Alan Turing asked whether a machine could imitate us. Seventy-five years later, we have our answer. It can, sometimes better than we can imitate ourselves.
But imitation was never the standard that mattered for anyone who publishes. The standard is truth, credibility, reputation. Can you stand behind every claim? Can you show where it came from? Would it hold up if a journalist, a competitor, or a skeptical reader went looking? If someone asked to see your AI content policy, could you show them?
AI passed the Turing test. Your content still has to pass a much tougher test: the credibility test.
That's exactly what we built AmICredible to help with. Run your next AI-assisted draft through AmICredible’s credibility review app before it goes live. You'll get a credibility score, a clear explanation of why, and the credible, original sources behind the analysis, so you can check our work too. Because, being able to show it's right is the real advantage.
Sources
- Google Search Central, Google Search's guidance on using generative AI content on your website, updated October 1, 2026
- Google Search Central, Creating helpful, reliable, people-first content, updated October 2, 2026
- UC San Diego Today, AI Can Seem More Human Than Real Humans in a Classic Turing Test, Study Finds, May 19, 2026
- Jones, C. R. and Bergen, B. K., Large Language Models Pass a Standard Three-Party Turing Test, Proceedings of the National Academy of Sciences, 2026 (preprint: arXiv, March 2025)
- Pew Research Center, Key findings about how Americans view artificial intelligence, 2026
- Pew Research Center, How Americans View AI and Its Impact on People and Society, September 2025 (survey conducted June 2025)
- University of Canterbury, The Turing Test, July 24, 2024
