Is Perplexity Pro Still Worth $20 a Month? A Reality Check After the Quiet Throttling

For about two years, Perplexity Pro was the low-controversy answer to a common question: which AI subscription should a research-heavy professional buy first? Real citations, access to multiple frontier models, generous limits, twenty dollars. Most reviews you will find ranking for that question still say some version of this.

Those reviews are describing a product that no longer quite exists. Something changed in the first half of 2026, it changed quietly, and the ranked reviews have not caught up. That gap is the reason for this article.

What Pro is supposed to be

The pitch is unchanged: $20 a month buys upgraded search with hundreds of enhanced queries, access to frontier models from multiple labs through one interface, Deep Research reports, file uploads, and now the Comet browser. The differentiator has always been citations: answers built from named, linked sources rather than confident free-floating text.

On the good days, that pitch delivers. Cross-checking a claim across sources with the receipts visible is genuinely faster in Perplexity than assembling the same picture from a chatbot plus a search engine. The product’s core competence is real, and none of what follows disputes it.

What happened in May

The details assembled from reporting and community documentation over May and June: users describing weekly caps on advanced models arriving days early despite normal usage. File upload limits reached after a couple of documents. Deep Research and Labs queries drawing down the same allowance as ordinary questions, so the premium features consumed the budget for everything else. Reports that using the Comet browser also deducted from the same pool. And in June, the Labs feature was removed in favour of a pay-per-use replacement that includes no credits at all.

Perplexity’s explanation, that the tightening targeted accounts on fraudulently resold promotional codes, is plausible for some of the affected accounts and is worth taking seriously. It does not address the underlying design complaint, which the community made repeatedly and which matches my own testing: the product does not clearly show you your limits or your consumption. You discover the ceiling by hitting it.

There are also unverified but persistent community claims of something worse: that when limits are reached, the product silently routes queries to cheaper models without notifying the user. I could not conclusively confirm this behaviour in my own testing, and I flag it as allegation rather than fact. But the fact that the interface makes it genuinely difficult to verify which model handled your query is itself a transparency finding.

The accuracy caveat that predates all of this

Separate from the limits story, one piece of independent research belongs in any honest Perplexity review. In March 2025, the Tow Center for Digital Journalism tested AI search tools on their ability to correctly attribute news content. Perplexity answered 37 per cent of queries incorrectly. The counterintuitive finding: Perplexity Pro, the paid tier, performed worse on this test, at 45 per cent.

The methodology tested a specific task, news attribution, and should not be stretched into a general accuracy score. But the direction of the finding matters for the buying decision: paying more did not buy more reliability, and citations that look authoritative still require the checking you were hoping to outsource. In my 90 days, the citation quality was genuinely useful and genuinely imperfect in roughly the proportions that study would predict. Treat Perplexity as a research accelerant that shortens the path to sources, not as an oracle that removes the need to read them.

Where that leaves the buying decision

The comparison landscape has also moved. ChatGPT’s search has closed much of the citation gap. Google’s AI mode is free and improving. Claude handles long-document research with fewer usage anxieties at the same price point. Perplexity’s moat, cited search as the whole product, is narrower than it was when most of its glowing reviews were written. I keep a fuller framework for choosing between AI subscriptions in my model selection guide, and Perplexity now enters that comparison without the presumption of victory it used to carry.

So, three honest recommendations.

If you are a heavy research user on the fence: wait a cycle. The limits situation is unstable, the company’s communication about it has been poor, and buying an annual plan against a moving target is how the affected users in May got hurt. Monthly billing only, if at all.

If you already subscribe and it works for you: nothing here says cancel. The core product remains good. Check which model handles your important queries, keep an eye on your consumption early in each week, and know that the free tier plus a general chatbot covers more of the use case than it did a year ago.

If you were sold by an older review: recalibrate against the current product, not the 2025 one. The tool is still useful. The generosity that anchored the old recommendations is no longer something I can verify exists, and I am not willing to promise it to you on the company’s behalf.

That is the whole Reality Check, really. Perplexity built its brand on showing sources. Its own pricing now needs the same treatment.