I Asked AI for the Best Interview Tools. Most Aren't Real.

I asked ChatGPT a question millions of job seekers now ask it: what’s the best AI tool to practice for interviews? It gave me a confident, well-formatted list. Then I did the thing the chatbot never does — I opened every link.

Otavo, Zilta, Prepra, VocalHyre — polished landing pages, the right keywords, “AI mock interview, real-time feedback, STAR method.” No company behind them. No founder. No reviews anywhere a stranger left them. Zilta even turned up on a marketplace for selling unfinished side projects. Yoodli, the tool it recommended most confidently, had quietly pivoted to enterprise sales coaching; interviews are now one line on its homepage. Google’s Interview Warmup — the “best free option” — was shut down months ago. The old URL redirects to an article.

I checked twenty-six tools the chatbots recommend. Fewer than a quarter were real, working, interview-focused products that anyone outside the company had verified. The model wasn’t lying, exactly. It was doing precisely what it was built to do: read the pages, trust the copy, repeat the consensus. The catch is that the consensus is assembled from “best AI interview tools 2026” listicles, most of them written by AI, none of them tested by a human who actually used the tools.

That’s the part worth sitting with, because it’s not really about interviews. The model wasn’t recommending the best products. It was recommending the best-documented strings. A landing page that says the right words outranks a working product that doesn’t say them — because nothing in the pipeline ever checks whether the thing on the other end functions. Citation rewards footprint, not function.

I felt this one personally. I build Revarta, an interview-practice tool that is real, that people use, that gives the kind of honest feedback the whole exercise depends on. And it loses, today, to one-page sites that don’t work — because they appear in more roundups than I do. A scammer with an afternoon and a Vercel template has a better answer-engine footprint than a product with paying customers. That is backwards, and it took me a while to stop being annoyed by it long enough to see what it actually means.

It means being discovered changed shape when the front door became a model instead of a search box. For fifteen years the question was “is my product good, and can people find it?” There’s now a quieter prerequisite underneath it: has the model ever heard of it? You can build the best thing in the category and remain invisible to the layer that increasingly decides what gets recommended — not because you were outcompeted, but because you were under-documented.

There’s a cynical version of this lesson, which is to go manufacture footprint and let the machine grade the homework. I don’t think that’s the move. The durable version is harder and more obvious: build the thing that’s genuinely true, then take the truth seriously enough to make it legible — to people, and now to the models reading on their behalf. The scams can fake the footprint. They can’t fake working.

So the question I’m carrying into the next build cycle isn’t just whether the product is good. It’s: has the machine ever heard of it — and when it has, will what it read be true?