The Product Hunt Graveyard: 1 in 4 top launches from 2023–2025 is already dead
Original research by RightAIChoice · every website re-probed, verdicts as of July 18, 2026 · every verdict carries downloadable evidence
TL;DR — as of July 18, 2026
- We visited the live website of 2,291 productsthat made Product Hunt’s daily leaderboards between January 2023 and April 2025.
- 24.4% are hard-dead — the domain no longer resolves, the page is parked for sale, the security certificate expired, or the homepage returns 404.
- Another 4.5% now redirect to a different company (acquired or quietly folded). Total gone as independent products: 28.9%.
- AI tools die at the same rate as everything else — 24.5% vs 24.4%. The AI boom changed what gets built, not what survives.
- Launch-day success is real protection: finish in your day’s top 5 and the death rate drops to 9%, vs 25.9% for everyone else. But it’s no guarantee — several 600-upvote launches are gone without a trace.
- Death comes fast: the 2025 cohort is already 20% dead after roughly a year.
The headline finding
Every day, Product Hunt features roughly twenty new products — the launches that beat hundreds of others to the front page. These are not random side projects; they are the products that won attention, upvotes, and often press coverage on day one.
Plenty of curated “graveyard” lists exist — from Killed by AI to CB Insights’ startup post-mortems — but we couldn’t find anyone who had measured the actual rate: of the launches that win attention, how many still exist? So we took 59complete daily leaderboards spread across January 2023 – April 2025, found each product’s real website, and visited all 2,291 of them — as of July 18, 2026.
The answer: 24.4% are hard-dead (505 of 2,066products with a definitive verdict). Their domains fail DNS, sit parked with a “buy this domain” page, serve an expired security certificate, or 404 on their own homepage. Add the 4.5% whose domains now forward to some other company, and 28.9% of the products that won Product Hunt’s attention in the last three years no longer exist as independent products.
And this is the floor, not the ceiling: 9.8% of products gave ambiguous answers (anti-bot walls, timeouts, vanished app-store listings) and were excluded rather than counted — plus a site that still loads isn’t necessarily a business that still operates.
Death rate by launch year
Each row is the cohort of products that hit a sampled daily leaderboard in that year. “Dead” includes parked domains; “absorbed” (redirects to another company) is shown separately.
| Cohort | Verified | Dead | Absorbed | Dead % (AI only) |
|---|---|---|---|---|
| 2023 | 797 | 26.5% | 5.5% | 23.1% |
| 2024 | 860 | 24.7% | 3.7% | 26.5% |
| 2025 | 409 | 20% | 4.4% | 23.7% |
Read the curve bottom-up: the 2025 cohort — barely a year old — is already 20% dead. The climb to 26.5% for the 2023 cohort means most products that die do so within their first year or two; survival past that point flattens the curve. If a tool you rely on launched recently, its riskiest period is right now.
Does launch-day hype protect you?
Partly — and this is the finding that surprised us most. We expected launch-day placement to be noise. It isn’t:
By finishing rank that day
| Top 5 of the day | 9% dead |
| Rank 6 and below | 25.9% dead |
By launch-day upvotes
| Under 100 | 26.4% |
| 100–249 | 23.3% |
| 250–499 | 7.1% |
| 500+ | 12.9% |
A top-5 finish cuts the death rate roughly 2.9×. The likeliest explanation is not magic exposure — it’s selection: products capable of winning a competitive day tend to have real teams and real momentum behind them. But hype is a discount on risk, not insurance — the casualty table below is full of products that had huge launch days and are DNS-dead anyway.
AI vs everything else: no difference
The prevailing story says AI tools are uniquely ephemeral — thin wrappers doomed to die when the model underneath changes. Our data says otherwise: AI products are dead at 24.5%, non-AI at 24.4% (754 and 1,312 verified products respectively — a statistically meaningless gap).
One honest nuance: products carrying Product Hunt’s specific “Artificial Intelligence” topic tag die at 33.7% — above average. But so do comparable non-AI SaaS segments (see the category table below), and our broader AI definition — which also catches AI products filed under Writing, Developer Tools, or Marketing — lands exactly on the overall average. The boom changed what people build; it did not make software businesses die faster than they always have.
The deadliest — and safest — categories
Product Hunt topic tags with at least 40 verified products in the sample:
Highest death rate
| Writing | 42 verified | 45.2% |
| Hiring | 40 verified | 40% |
| Artificial Intelligence | 86 verified | 33.7% |
| Knowledge base software | 45 verified | 33.3% |
| SaaS | 67 verified | 32.8% |
| Developer Tools | 41 verified | 29.3% |
Lowest death rate
| Open Source | 45 verified | 8.9% |
| iOS | 64 verified | 12.5% |
| Design resources | 40 verified | 15% |
| AI Generative Media | 48 verified | 18.8% |
| Fintech | 40 verified | 20% |
| Design Tools | 139 verified | 20.9% |
Two patterns stand out. Writing tools are a killing field (45.2% dead) — the segment where a thin layer over a language model was easiest to ship and easiest to abandon. Open-source products barely die (8.9%): when the code is public, a project can outlive its founder’s attention span. If longevity matters to you, an open repository is the single strongest survival signal in this dataset.
Notable casualties
Each of these collected hundreds of upvotes on launch day. As of July 18, 2026, their domains fail DNS, sit parked, or serve expired certificates. (Paste any of them into the Wayback Machine to see what they looked like alive.)
| Product | Launched | Upvotes | What’s there now |
|---|---|---|---|
| Use ChatGPT | 2023-03 | 663 | Domain no longer exists |
| CostGPT.ai | 2024-10 | 662 | Domain no longer exists |
| Butter 2.0 | 2023-02 | 621 | Domain no longer exists |
| Humalect | 2023-04 | 535 | Domain no longer exists |
| Belly Buddy | 2024-02 | 463 | Broken security certificate — site abandoned |
| What to Meme | 2023-04 | 412 | Domain no longer exists |
| YC Rejection | 2023-11 | 398 | Domain no longer exists |
| Clearmind | 2023-08 | 348 | Server no longer responds |
| Boeing Be Gone | 2024-07 | 323 | Domain no longer exists |
| Devlopea 2.0 | 2025-02 | 294 | Domain no longer exists |
The absorbed: 94 products now redirect somewhere else
4.5% of verified products don’t 404 — their original domain quietly forwards somewhere else: an acquirer, a rebranded successor, or the founder’s next project. We don’t guess which from the outside — the dataset records the redirect, not the backstory. What’s certain is that the product as launched is gone from its original home. Live-verified examples:
- Algomo (2023-01, 662 upvotes) — algomo.com now forwards to newmode.ai
- AI Traffic Analytics (2023-09, 371 upvotes) — its link now forwards to writesonic.com (folded into the parent product)
- Dashcam (2023-08, 366 upvotes) — dashcam.io now forwards to testdriver.ai
- Breeew (2023-04, 280 upvotes) — breeew.com now forwards to getorchestra.com
- Plus for Google Slides (2023-02, 254 upvotes) — plusdocs.com now forwards to plusai.com (rebrand)
- Mermaid Whiteboard (2023-06, 246 upvotes) — mermaidchart.com now forwards to mermaid.ai (rebrand)
What this means when you pick tools
If you adopt a tool from a launch site today, this dataset says there’s roughly a 1-in-4 chance its website won’t exist in ~2 years — with your data, workflows, and integrations inside. Practical takeaways from the numbers:
- Prefer products with an open-source core — the 8.9% death rate is the best survival signal we found.
- Treat the first 12–24 months as the danger window — the 2025 cohort is already 20% dead. Demand an export path before you commit real workflows.
- Weight evidence of a real business over launch-day noise: a top-5 finish helps (9% vs 25.9%), but pricing pages, changelogs and shipping cadence help more.
- Check before you buy.This study exists because we track this continuously: we re-verify ~8,000 AI tools and score each one’s viability — including an at-risk list — precisely so you don’t build on a tool that’s about to join this page. Browse tools by category, compare them head-to-head, or see the broader State of AI Tools report this study belongs to.
Methodology — fully auditable
Sample. 59 complete Product Hunt daily leaderboards drawn across January 2023 – April 2025 (~20 featured launches per day, all ranks — not just #1s). After de-duplicating relaunches (a product keeps its earliest appearance) and dropping products with no usable website link, the cohort is 2,291 products.
Probe (run July 18, 2026).Each product’s website was fetched with a standard browser user-agent, 15-second timeout, and retries with backoff. DNS was checked separately and double-checked to absorb resolver blips.
Verdicts. Dead = no DNS record, connection refused/reset, expired/invalid security certificate, or HTTP 404/410 at the domain level. Parked = a domain-parking sales page (counted with dead). Absorbed= the domain redirects to a different company’s domain; same-name domain moves count as alive. Unknown = anti-bot walls, timeouts, vanished app-store listings, and deep links that 404 while the root domain still responds — 225 products (9.8%), excluded from every percentage and never counted as dead.
A second, stricter pass. Before publishing we re-audited every 404-based verdict: products whose link pointed at an app-store listing or a deep page on a bigger site were re-probed at the domain root, and anything ambiguous was moved to unknown. This lowered the headline from a preliminary 27% to the published 24.4% — we’d rather under-claim than defend a soft number.
AI classification. A product counts as AI when Product Hunt tags it with an AI topic or its name/tagline/description contains AI terms (844 of 2,291 products).
Reproducibility. Before publishing, we re-probed the entire cohort a second time, end to end, hours after the first run. The headline reproduced exactly (24.4%), every reported statistic landed within ±0.6 points, and 99.4% of individual verdicts were identical — the remaining handful are sites flickering in real time, which is what the internet does.
Known limitations.A responding website is a proxy for an operating product — some “alive” sites are zombies with dead backends, so the true product death rate is likely higher than our headline. The sample covers featured (leaderboard) launches; unfeatured launches likely fare worse. The 2025 cohort covers sampled days through April 2025 only.
Download the dataset
Per-product rows with the verdict and its evidence, as of July 18, 2026(CC BY 4.0 — free to use with attribution).
Cite as: RightAIChoice, “The Product Hunt Graveyard” (2026) — https://rightaichoice.com/state-of-ai-tools/graveyard · License: CC BY 4.0
FAQ
How did you decide a product is "dead"?
Hard evidence only. We fetched every product's website with a normal browser user-agent. "Dead" means the domain no longer resolves in DNS, the connection is refused, the security certificate has expired, the homepage returns HTTP 404/410, or the domain is parked for sale. Anything ambiguous — anti-bot walls, timeouts, an app-store listing that vanished, a deep link that 404s while the company's main domain still works — is classed "unknown" and excluded from every percentage. Excluding the ambiguous cases can only make the true death rate higher than our headline, never lower.
Does a redirect count as dead?
Never in the headline. A product that moved domains but kept its name (deeto.ai → deeto.com) is counted alive — that's a migration, not a death. A product whose domain now forwards to a differently-named destination (algomo.com → newmode.ai) is reported separately as "absorbed" — that bucket includes acquisitions, rebrands, and founders pointing a dead product's domain at their next project. We record the redirect, not the backstory, and none of it counts toward the death rate.
What exactly is the sample?
Every product on 59 complete Product Hunt daily leaderboards sampled between January 2023 and April 2025 — about 20 featured launches per day, every rank, not just the winners. After de-duplicating relaunches and dropping products with no usable website link, the cohort is 2,291 products, of which 2,066 produced a definitive verdict as of July 18, 2026. Each product's evidence line ships in the downloadable dataset, so any individual verdict can be audited.
Do AI products die faster than non-AI products?
Under our broad AI definition (AI topic tag on Product Hunt, or AI terms in the name/tagline/description) the rates are statistically identical: 24.5% for AI products vs 24.4% for everything else. One nuance worth knowing: products carrying Product Hunt's specific "Artificial Intelligence" topic tag die at 33.7% — the thin-wrapper era left real casualties — but comparable non-AI SaaS segments die at similar rates, and the broader AI population lands exactly on the overall average.
Does a big launch protect a product?
Partly, yes — and it surprised us. Products that finished in their day's top 5 are dead at only 9%, versus 25.9% for rank 6 and below. But success is no guarantee: several products with 500–700 launch-day upvotes now fail DNS entirely.
Why do so many top launches die?
The data can't prove motives, but the pattern in the casualties is consistent: single-purpose products built quickly on someone else's platform — browser extensions, one-feature "wrappers", weekend projects riding a trend — dominate the dead list, while open-source products die at only 8.9%, because a community can keep a project alive after its maker moves on. A launch spike is attention, not revenue.
Can I verify a specific product myself?
Yes — that's the point of shipping the evidence. Download the dataset below, find the product's row, and check the evidence column (the exact DNS error, redirect chain, or HTTP status we recorded). You can also paste any dead domain into the Internet Archive's Wayback Machine to see what the site looked like when it was alive.
Can I use this data?
Yes. The full per-product dataset (JSON and CSV, with evidence per verdict) is CC BY 4.0 — free to use, including commercially, with attribution. Cite "RightAIChoice — Product Hunt Graveyard study" with a link to this page.
Picking tools that will still exist next year?
We re-verify ~8,000 AI tools continuously and score their viability — that’s why we built this study.