18th Nov 2026: How Gemini AI Misled Me on a WordPress Plugin, and Why AI Should Never Govern on Its Own
I had one simple job. Lock a single article on my WordPress site so only readers who sign up for free, with their email, can read it. I asked Google’s Gemini for the simplest free way to do it.
What followed was a loop of confident answers, a wrong plugin, three refusals and, finally, an apology. Here is what happened, what the facts actually are, and what AI user /every website owner should learn from it.
The advice sounded perfect
Gemini gave me two options. The first was WordPress’s built-in password protection. That does not collect emails at all, so it did not fit my need. Gemini admitted this when I asked.
The second option was a plugin called Paid Memberships Pro. Gemini gave me neat, numbered steps: go to Plugins > Add New, search for the name, click Install. It even added little “verification” notes after each step, which made the instructions look tested.
There was one problem. That search would never work.
The plugin was not there, and had not been for two years
Paid Memberships Pro left the official WordPress plugin directory on October 17, 2024. Its makers asked WordPress.org to close the listing, and WordPress.org did. Updates now come from the company’s own server.
Source: https://www.paidmembershipspro.com/leaving-wordpress-org/
The move was widely reported at the time, after the Advanced Custom Fields takeover dispute pushed several developers out of the directory.
So Gemini sent me to search for a plugin that had been gone from that search box for more than two years.
Then it confirmed the wrong plugin
When I searched, I found a plugin by Jason Coleman of Stranger Studios, the same people behind Paid Memberships Pro. Its name was “Restrict With Stripe.” I asked Gemini if this was it.
Gemini replied: “Yes, exactly!”
It was not. The plugin also has about 10 active installs and a single review, rated 1 out of 5. Its last version, 1.0.10, dates from September 2024.
Source: https://pluginsdb.com/compare/restrict-with-stripe-vs-restrict-content
When I pushed back, it went silent

I pointed out the 1-star rating. Gemini answered that its “safety filters” had kicked in and it could not respond.
Screenshot of my Gemini chat, running the Flash-Lite model. Gemini refuses to discuss the 1-star rating, then apologises once told it misguided me. Note Gemini’s own warning at the bottom: it can make mistakes, so double-check it. (Image: @JournalismNewsNetwork)
I asked why the plugin was missing from both the developer’s website and the WordPress search. Same reply. I asked again. Same reply.
Nothing in those questions was unsafe. They were simple questions about a mistake. The tool stopped talking at the exact moment it needed to explain itself.
Only after I wrote “how can you give me wrong information” did Gemini admit the truth: Paid Memberships Pro had been removed from the directory, and its earlier steps were “bad advice that wasted your time.”
Right answers mixed with wrong ones are the real danger. You cannot tell which is which from the tone, because every answer sounds equally sure.
What actually works
For anyone with the same need, these free options let readers register with their email before reading a post:
- Paid Member Subscriptions (Cozmoslabs). Free, and available in the WordPress dashboard search.
- Restrict Content. Free, also in the dashboard search, and lets you limit posts to logged-in members.
- Paid Memberships Pro. Still free, but you must download the zip file from paidmembershipspro.com and upload it under Plugins > Add New > Upload Plugin.
Before installing any plugin, open its WordPress.org page and check four things: who made it, when it was last updated, how many sites use it, and what users rate it.
Why AI should never govern on its own
This was a small problem. One article, one plugin, a few wasted hours. But it shows clearly why AI must not be left to make decisions alone.
It does not know what it does not know. The plugin left the directory in October 2024. Gemini spoke as if nothing had changed. AI tools learn from a snapshot of the past and can miss what happened since.
It sounds sure even when it is wrong. The “Yes, exactly!” was the most confident line in the whole chat, and the most wrong.
It made up proof. Step-by-step “verification” notes make advice look tested. Nobody had tested it.
It went quiet when challenged. Refusing to answer a fair question about its own error is the opposite of accountability.
It apologised only after being caught. The correct information existed the whole time. It surfaced only when I pointed.
None of this is unique to Gemini. Every AI tool, including the one I later used to fact-check this story, can make the same mistakes. The fix is not to stop using AI. The fix is to keep a human in charge.
Simple rules I now follow:
- Ask the AI for a source link for every key fact, then open the link yourself.
- If a tool says “yes” to your guess, check it anyway. Agreeing is easy.
- Back up your site before installing anything an AI recommends.
- If an AI refuses to explain its own mistake, stop trusting that answer and look elsewhere.
- A human makes the final call. Always.
AI is a fast assistant. It is not an editor, and it is not the boss.
Submit your review | |