Ever had a company promise you a gourmet meal and then serve you a single, sad cracker? That’s the vibe right now with Perplexity AI, where asking for the steak dinner you paid for might just get you kicked out of the restaurant.

A paying Pro user noticed the much-advertised “Deep Research” feature seemed to be taking a permanent coffee break. So, they did what any good internet detective would: they used Perplexity’s own official blogs and documentation as evidence. The post blew up, with hundreds of upvotes and comments from fellow users agreeing the feature seemed severely throttled. The community’s verdict was in: the cracker was indeed just a cracker.

Here’s the hilarious part. Instead of a “thank you” or even a “we’re looking into it,” the official response was the digital equivalent of a bouncer’s hand. The user got a permanent ban from the subreddit, and the entire discussion thread vanished. It’s the ultimate “shoot the messenger” strategy, but the messenger brought the company’s own signed confession as evidence. You almost have to admire the audacity.

It creates a wonderfully ironic new customer service flowchart: Have a problem? Gather proof. Use the company’s own words as that proof. Watch the community agree. Get deleted. It’s a bold new world where the most effective way to expose a feature’s downgrade is to become a martyr for it. The real “Deep Research” was uncovering how quickly a support forum can turn into a memory hole.

In the end, this is a perfect little parable for the internet age. It reminds us that sometimes the most perilous deep research isn’t done by an AI agent, but by a customer with a receipt and a screenshot. Just be prepared for your findings to be buried faster than you can say “subscription refund.”