Comment by Aurornis | Hacker News Reader

Comment by Aurornis | Hacker News Reader

I used one of the 12-month free Perplexity offers when they were everywhere. It felt slightly useful at first for simple queries where I didn’t want to go through the top 10 Google results manually. If I was looking for a specific recipe I remembered or a help page or user manual it would usually find it quickly.

Then they started optimizing for speed of responses over quality of results. I can enter a query and see my results appear in a second, but they’re garbage. The links and references it gives frequently don’t match the text right next to them. It feels like someone had a KPI to make responses as fast as possible and they optimized for that above all else.

They added a “Computer” option that’s supposed to do research for you. Half the time I can’t get it to trigger through the UI. Pressing the submit button doesn’t work. When I can get it to trigger, most of those sessions will work for a while and then just stop before an answer comes back.

The only reason I keep using it is to keep observing a company that has been heavily marketed and hyped, which should have had a market leading position for something. Even non-technical people I know who listen to Joe Rogan (where Perlexity is advertising heavily, I’m told) are asking me about it.

Now there are reports of people being billed at the end of their trial period without warning, despite them saying that they will warn before this happens. There are some alarmingly bad customer support screenshots where the customer support agent (AI? Probably) acknowledges that they didn’t send the email they promised but refuse to help anyway. It takes escalating it on Twitter to get it corrected.

If I want to do actual research or AI assisted web searching I have Claude or ChatGPT do it. The results are so much higher quality and it does exactly what I ask. It may take 45 seconds instead of the instant response from Perplexity but I save time overall because the response and links are more likely to be correct

I paid for perplexity pro for 3 years. I genuinely enjoyed using it and felt it was better overall than ChatGPT etc due to the way it showed sources etc. I liked being able to use different models depending on what I was looking for, and the deep research was helpful.

I think they probably damaged themselves by going for a land grab of user base through freebies. It meant the users weren’t ever going to convert to paid customers, so it was more to show investors that they had a user base. But, with an increased base of users who weren’t paying, it then meant they needed to find either new revenue streams or cheaper ways to provide the service. Unfortunately, it seems they went with the new revenue streams whilst also decreasing the functions paying members were able to access (something I find quite abhorrent- I paid a service level, but then they change what I receive mid-subscription). And then computer - rammed down my throat. One reason I pay for pro is to stop the nagging noise of paid tiers. And instead, they actually created a way of logging in and continually seeing gated functions.

So, after paying them upwards of $400-$500 and being a loyal customer, I walked.

What do you use now? I have used Kimi, but the spaces of perplexity are keeping me to stay.

More or less the same experience, I used to recommend it to others, and observed a steady decrease in performance over long periods of time, smooth enough that it took a while for it to register.

I am just about to end my free trial, it lasted 12 months for the educational version which does not have any credits for Computer, I just canceled it on the webpage and it says it is going to be canceled at the day my renewal would happen. I’m observing closely.

Perplexity promised me this plan would be 5 USD once it finishes, the interface say they will bill me for 20, supports tells me not to worry, but I’m afraid it’s just a bot. I trust my instincts.

Regardless of price, I decided to cancel. Assuming 5 USD was a real promise, that looked like a great deal on paper, but I had a hunch that something is not good once they locked me out for generating images after a couple of tries, but I’ve been able to do so in the past, so I compared it with the free tier of Gemini and had more luck with image generation on the latter. The webpage is also really slow on Safari after a couple of exchanges.

What killed my interest was that I compared the search results and deep research on a couple of queries with ChatGPT Go, and I found that the search results were worse on Perplexity, while the deep research reports looked better superficially but, on close inspection, collected older information from fewer sources.

No comments on Computer, I’ve heard good things but didn’t receive credits for trying it, it’s probably outside of what I want to pay.

If they had some sort of tier that was like $5 to $10 and the main thing it had on it was their custom model for research, no other model, I might re-subscribe, but they were definitely burning way too much compute trying to give it away in the hopes others kept their subscription. I was using it to trim down on my direct Claude Compute usage since they ran me unmetered for a while.

I would draft a development plan with Claude on there, then feed it to Claude Code. This isn't sustainable, but given that I had x number of months pre-paid for, I just used it.

I suggest comparing the search results with other providers for your own usage, I’m not convinced myself that even at 5-10 USD it would be worth it, and I was doing the comparisons using the better models on Perplexity.

Perplexity really fell off for me, especially the free version. Their syntax highlighting stopped working, and even the fonts seemed messed up. You could tell they were mixing and matching whatever the cheapest model was because the sources would be messed up and printed as plane text <grok source 1 http:// ... />

I guess it makes sense though, unless you've got the lowest pricing on your own model how can you compete.

I got a year of free Perplexity Pro with my Revolut sub. At first I really, really liked it. It felt faster, better, and smarter than (free) ChatGPT. But instead of getting better (or at worst staying at the same level), it seems like it gets worse and worse instead.

For a few months now I have it that it randomly switches to another language for like 1–2 words (most often Japanese) mid sentence. It feels like it got worse at referencing earlier parts of the same convo. For example, I asked it to compare a 9‑year‑old flagship phone chip to the newest model of the smartwatch the same company produces. I used the short version of the phone name (left away the company name since I thought the relation would be obvious enough).

It then asked what I meant by the short version, if I meant the watch of a competitor, if I meant some random vacuum robot, or if I meant something entirely different.

I then said the full phone name and it gave me back a summarization of “Is Phone Name still usable as a daily driver in 2026?” Something I never asked.

> I got a year of free Perplexity Pro with my Revolut sub. At first I really, really liked it. It felt faster, better, and smarter than (free) ChatGPT. But instead of getting better (or at worst staying at the same level), it seems like it gets worse and worse instead.

Same experience here. I think I started using Perplexity for researching stuff before ChatGPT had search and particularly before I had access to Pro mode. Then I experienced the very same thing as you, the Perplexity experience got faster, but also quality degraded a lot, until eventually ChatGPT + search tool + Pro just completely overtook it. Some queries might take 20-30 minutes to fully complete, but then the results are also really good and extensive.

I’ve personally found that Google’s AI mode is surprisingly capable and almost absurdly fast, although the hallucination rate is somewhat proportionally higher too to match its seeming over-responsiveness. Which means Perplexity in effect doesn’t have anything to differentiate it.

I don't notice it's hallucinations (in the sense of making up answers) so much as where it's simply wrong. For example, earlier in the week, I was working with product X, and since the original vendor of X doesn't really support it any more, I asked "who else sells product X under their own brand", and Google AI quickly told me that noone else sells product X, that it was full of proprietary tech and quickly devolved into replaying product X marketing spiel. Except...I knew that at least one other vendor did sell product X, only difference is paint job. After too much "no, you're wrong..." and "Yes, you're right...", it finally coughed up that there were 2 other vendors that sold it. And if I didn't already know about another vendor, I probably would still be putzing around with poor support and little documentation.

I think one of the big problems is that the AI Mode and the AI Overview thingy rely on the found web sources too much. They seem to be the ground truth which often conflicts with Gemini's internal knowledge rather than "enriching" it.

EDIT: Like this comment said: https://news.ycombinator.com/item?id=49538500

I was one of the people that used the free Pro tier for a year and it is still my go to app when I want to quickly check something and have some sources linked, but I agree that over that year the quality of linked sources dropped significantly. But at least I got noticed when my trial period was near the end, so they either learned, or I was lucky.

I use perplexity simply to avoid underlying vendor lock in.

> Then they started optimizing for speed of responses over quality of results

Imagine all of the blast radius to society if this type of incentive is reproduced everywhere. Police searches through Flock databases, matches for hiring candidate resumes, organizing targets in a war with Iran, etc.

Reply on news.ycombinator.com