Skip to content

Are Product Hunt's featured products still online today?

A featured launch once signalled that a product had stood out from a crowded field. This analysis follows more than 90,000 of those products from December 2013 to January 2022 and asks a simpler question: how many original websites still respond today, and what patterns explain the ones that do not?

How the dataset was built

The evidence comes from the Product Hunt API, covering featured products from December 2013 through the end of January 2022. Two API versions were available at the time: an older REST interface, described as deprecated, and a newer GraphQL version. The older interface was chosen because each query returned 50 posts with complete data, compared with 20 posts with limited data from the newer version, allowing the full dataset to be collected within a few hours.

Collecting usable URLs took three stages. First, paginated product records were fetched as JSON, using the largest practical page size and an authorization bearer token. Second, because the API exposed only redirect URLs, a HEAD request was made to each redirect to recover the product's original address. Third, another HEAD request to that original address recorded its current status code and other header information.

Two practical details matter when repeating this kind of work. Rate-limit response headers indicate how many requests remain and when the allowance resets, so a crawl should pause until the reset point rather than retrying immediately. Scale is also a design issue: more than 90,000 products required over 180,000 additional HEAD requests. Asynchronous HTTP requests, here with Python and HTTPX, make that volume far more manageable than issuing every check one after another.

The headline result: 22% no longer respond

Monthly featured-product volume rose steadily through 2016, then fell sharply in 2017. A note from Product Hunt founder Ryan Hoover explains that podcasts, games and books were deprecated as separate categories, reducing the scope of what counted as featured. Volume rose again during the pandemic period of 2020 and 2021.

When each original URL was visited, outcomes were grouped by HTTP status code or by another exception such as a DNS or SSL failure. Responses in the 2XX and 3XX ranges counted as successful. Across the full dataset, 20,970 products, or 22%, now produced an error of some kind. That headline figure is the starting point for every other comparison: the question is not simply whether a product failed, but whether failure tracks popularity, topic or age.

Votes did not predict survival

Ordering products by votes puts Startup Stash first, with 21,839 votes in 2015, followed by Slack with 17,432 in 2013, Remove.bg with 16,951 in 2018, Attach.io startup pitch decks with 11,101 in 2016, and Pexels with 10,247 in 2015. The lowest-voted products were not listed, but every product in both the top and bottom groups still returned a successful HTTP response.

Votes turn out to be a weak predictor of survival. Most products received between 10 and 50 votes, the average was 179, and just 31 products passed 5,000 votes. Expressed as percentages, failure rates fluctuated within a band of roughly 14% across vote groups, with the largest share of failures appearing among products that received 1,500 to 5,000 votes. A launch-day surge in attention, in other words, did not guarantee a longer-lived website.

Topic differences mostly reflect age

By total votes, the generic tech topic led by a wide margin, attracting over 10 million votes, more than double any other topic. Topic comparisons become more informative when failure rates are examined instead of raw counts, and when small topics with fewer than 100 products are excluded.

On that basis, no-code recorded the lowest percentage of errors, while crowdfunding recorded the highest. The gap is largely a timing effect rather than proof that one kind of product is inherently more durable. No-code entries appear mainly in the final couple of years of the dataset, so their websites have had little time to disappear. Crowdfunding entries recur throughout the full history, giving them many more years in which to fail. Any topic comparison should therefore be read alongside the age of the products in that topic.

Age is the strongest signal

Older featured cohorts produced more errors in absolute numbers, with the peak aligning with 2016, the period when the greatest number of products were being posted. The 4XX and other error types moved closely together across the timeline.

Proportions tell a slightly different story. The busiest posting period did not have the highest failure rate; the highest was 32% for products featured in July 2014. Grouping results by year shows the overall error proportion steadily decreasing toward 2022. Age remains the strongest correlate: the longer a product has existed, the more opportunities its website has had to lapse, move or shut down.

What to check before trusting a product list

Use this checklist before treating any historical product list as a live directory:

  • Resolve redirect URLs to original addresses before judging whether a product still exists.
  • Count 2XX and 3XX responses as reachable, and record DNS, SSL and other exceptions separately from HTTP errors.
  • Respect rate-limit headers and pause until the stated reset rather than hammering the API.
  • Plan for asynchronous requests when the dataset runs into tens of thousands of URLs.
  • Compare failure rates within age bands before attributing differences to votes or topics.
  • Exclude very small topic groups before ranking topics by failure percentage.

The overall lesson is measured rather than bleak. Most featured products from 2013 to 2022 still respond, but 20,970, or 22%, no longer do, and age explains more of that loss than launch-day popularity. The Product Hunt API remains a rich source for further experiments on how products, topics and attention evolve over time.

A four-step survival check for historical product lists
CheckWhat to recordWhy it matters
Resolve the redirect URLOriginal product addressThe API exposes redirects, not final URLs
Send a HEAD request to the original addressStatus code and headers, or the exception typeSeparates HTTP errors from DNS and SSL failures
Classify the outcome2XX or 3XX as reachable; everything else as an errorMatches the method used for the 22% headline figure
Compare within age bandsFailure rate by year or cohortPrevents votes and topics from being credited for what is really age

Derived from the captured method and findings; not a regenerated source schema.

Frequently asked questions

How many featured products were checked, and how many were no longer reachable?

The captured dataset covers featured products from December 2013 through the end of January 2022, totaling over 90,000 products. Of those, 20,970, or 22%, produced some form of error when their original URLs were checked, while 2XX and 3XX responses were counted as successful.

Did higher vote counts predict whether a product stayed online?

Votes alone did not show an obvious relationship with survival. Most products received between 10 and 50 votes, the average was 179 votes, and only 31 products exceeded 5,000 votes. Failure percentages fluctuated by roughly 14% across vote groups, with the highest share of failures in the 1,500 to 5,000 vote range.

What best explained whether a product had disappeared?

Age was the strongest pattern. Older cohorts produced more errors in absolute terms, peaking around 2016 when posting volume was highest. As a proportion, the highest failure rate was 32% for products featured in July 2014, and the overall error proportion steadily declined toward 2022.

Which topics had the lowest and highest failure rates?

After excluding topics with fewer than 100 products, no-code had the lowest recorded error percentage and crowdfunding the highest. That contrast largely reflects timing: no-code products are mostly recent, while crowdfunding products span the full history and therefore had more time to fail.