Key takeaways
- Most ideas are not bad, they are unfinished. 76.0% land in the caution zone (40-69), not the no-go pile, and the median score is 54, not 30.
- The top killer changed in 2026. Early-stage risk overtook go-to-market as the number one failure signal, 36.9% vs 25.7%. Most failed validations are not ideas yet.
- A crowded market is a green flag. Ideas flagged for competitive pressure average 64.5, the highest score of any major risk group. An empty market usually means no demand.
- A 70+ score puts you in the top 18%. Only 17.5% of the benchmark set earned a "go" verdict, and fewer than 0.2% broke 90.
- Rescanning works. 734 founders fixed flagged issues and ran their idea again. Average gain: 8.9 points. 75.2% of them improved.
The median startup idea scores 54 out of 100. Most validation tools would have told you 78. That gap is why I publish this report: these benchmarks come from more than 6,000 completed analyses run through Preuve AI, free scans and paid deep analyses combined, each one scored against live data: competitor databases, Google Trends, Reddit, community forums, and more.
This is the third edition. The first covered 1,000 ideas; the second, in April 2026, rebuilt the tables from 4,000+. This July 2026 edition computes every number on a cleaned, like-for-like comparison set, and the methodology section spells out exactly what I cleaned out and why. No synthetic data, no ChatGPT-generated ideas. Cite freely.
The Preuve AI Score
The Preuve AI Score is a 0-to-100 startup viability rating computed by 10 AI agents cross-validating claims against 60+ live data sources including competitors, demand signals, market size, community sentiment, hiring trends, and patent activity. Every point is source-linked. A score above 70 indicates launch-ready signals. Below 40 means fundamental validation work remains.
Overall Viability Distribution
The average viability score across the benchmark set is 56.3 out of 100. The median is 54. The distribution is still centered in the 50-59 range, where a third of all ideas land.
| Score range | % of benchmark set |
|---|---|
| 0-9 | 0.2% |
| 10-19 | 0.3% |
| 20-29 | 1.8% |
| 30-39 | 4.1% |
| 40-49 | 23.8% |
| 50-59 (peak) | 33.4% |
| 60-69 | 18.9% |
| 70-79 | 15.4% |
| 80-89 | 2.0% |
| 90-100 | 0.2% |
76.0% of all ideas land between 40 and 69. This is the caution zone, where the idea has a kernel of something real but is missing at least one thing that matters: a distribution plan, a clear competitor picture, or enough demand evidence from outside the founder's head.
Fewer than 0.2% of ideas score above 90. In the entire benchmark set, 9 ideas broke 90 and only 2 ever hit 100. A score that high means live market data shows strong demand, a gap competitors have left open, and multiple converging signals. It remains exceptionally rare.
Eight months, one distribution
New in this edition: I tracked the monthly average across the whole benchmark window. It never left the 53-58 band. December 57.0, January 53.1, February 57.3, March 56.3, April 55.7, May 58.4, June 56.0, July 55.3.
Why that matters to you: a score you earn in July means the same thing as a score earned in January. The scoring did not drift softer or harsher month to month. If a tool's median creeps upward over time, its scores are inflating. This one holds still, which is what makes the benchmarks below usable as a reference.
Verdict Breakdown: Go, Caution, No-Go
Every analysis ends with one of three verdicts based on the composite viability score:
17.5%
Go
Avg score: 74.5
76.0%
Caution
Avg score: 54.3
6.4%
No-Go
Avg score: 30.1
The no-go rate went up since April, from 4.8% to 6.4%. The system got harsher, not softer, and I consider that a feature. A validation tool that fails almost nothing is a compliment machine.
The gap between a go idea and a caution idea is an average of 20.2 points. In most cases, that gap comes down to 2-3 fixable issues: unclear positioning, no competitor awareness, no distribution plan. The idea itself is rarely the problem.
Why do most startup ideas fail validation in 2026?
A primary risk label was captured on 4,995 of the analyses in the benchmark set. Here is the distribution, and it contains the biggest shift since the April edition:
| Top risk | % of tagged ideas | Avg score |
|---|---|---|
| Early-stage risk | 36.9% | 51.7 |
| Go-to-market | 25.7% | 57.9 |
| Regulatory risk | 12.0% | 57.1 |
| Competitive pressure | 11.5% | 64.5 |
| Funding requirements | 6.6% | 59.3 |
| Production challenges | 3.5% | 58.0 |
| Team execution | 2.2% | 46.6 |
Minor categories like technical complexity, legal risk, and brand risk account for the remaining 1.7% of tagged analyses.
Early-stage risk overtook go-to-market as the number one killer. In April it was flagged on 25.6% of analyses; it is now 36.9%. These are ideas too vague to properly evaluate: no defined buyer, no pricing reference, not even a competitor list. Their average score of 51.7 is the second lowest of any major risk group. My read on the shift: AI tools made building feel instant, so more people scan a sentence-long concept before it has become an actual idea. Validation cannot score what does not exist yet.
Go-to-market risk drops to second at 25.7%, but the pattern behind it has not changed. The founder has a product concept but no distribution plan and no channel strategy. Nothing shows the target buyer is even reachable. Roughly 1 in 4 flagged ideas still has no realistic path to a first customer.
The competition paradox
Competition ranks fourth as a risk factor (11.5%), but among the major risk categories it has the highest average score at 64.5. That is counterintuitive until you realize: competition means there is a real market. Ideas with no competitors often have no customers either. And genuinely empty markets barely exist. The agents surfaced three or more named competitors in more than 98% of analyses; only 16 reports in the entire set came back with zero.
What happens when founders fix the flags and rescan?
This stat did not exist in the first two editions, and it changed how I think about scores. The benchmark set contains 734 improved iterations: a founder got a score, fixed flagged issues, and ran the analysis again on the sharpened idea.
734
re-scored iterations
+8.9
average score gain
75.2%
improved their score
An 8.9-point average gain is bigger than the 5-point band most ideas miss "go" by. Three out of four founders who did the homework moved up. A first score is not a verdict on you or your idea. It is a snapshot of how much validation work is left, and the rescans prove the work moves the number.
Do founders who pay for deeper analysis have better ideas?
Preuve AI offers two tiers: a free quick scan and a paid deep analysis that unlocks competitor research, community signals, and investor-grade sections.
Free Scan
55.7
Average score
Deep Analysis (paid)
66.6
Average score
Paid users average 66.6 vs 55.7 for free users, a gap of 10.9 points. That is not because paying makes your idea better. It is selection bias. Founders who invest $29 in validation tend to be more serious, have already done some homework, and are working from a clearer concept with a more defined market in mind. The gap was 13.0 points in April; it narrowed, but it did not close.
The practical takeaway: if you are going to pay for validation, do the basic homework first. Define your buyer. Name three competitors before you spend a cent. Check what price range the market actually tolerates. That baseline preparation is worth more than any tool.
What separates a 50 from a 75?
After reviewing thousands of analyses, the gap between a mid-range score and a high score almost always traces back to the same pattern:
1. Named competitors with pricing
In this edition's set, ideas where the analysis could not surface a single named competitor averaged 42.9. Ideas with three or more named competitors averaged 56.3, a 13.4-point gap. "No competitors" almost always means "I have not looked."
2. Demand evidence from outside the founder's network
Reddit threads, review complaints, Google Trends upticks, community forum posts. External demand signals show that strangers already care about the problem. Without them, the analysis can only score what the founder claims.
3. A specific buyer, not "everyone"
"Small business owners" is not a target market. "Solo accountants in the US who still use spreadsheets for client onboarding" is. Narrow targeting makes every other validation signal sharper: competitors get easier to name, demand gets easier to measure, and pricing gets a real reference point instead of a guess.
The difference between a 50 and a 75 is not a better idea. It is better homework. Most ideas in the 40-60 range have a real kernel of value, they just have not been pressure-tested against the market yet.
How to read this benchmark
One caveat before you compare your own number to these: the set is self-selected. Founders who run a scan are already further along than founders who never test anything, so the true population of startup ideas probably scores lower than this benchmark, not higher.
And if you are staring at a 54 right now, that is the median, not a rejection. The rescan data above says founders who treated their report as a to-do list gained 8.9 points on average. That is how I want this report used.
Methodology
This report starts from more than 6,000 completed analyses processed by Preuve AI as of July 30, 2026. Every table is computed on a cleaned, like-for-like comparison set. Previous editions said "cleaned" and left it there; this one names the cuts, because a benchmark you cannot audit is just marketing:
My own scans are out. I run hundreds of test and demo analyses while building the product. All of them are excluded, so the set contains only real founders' ideas.
One promotional batch is out. A batch of analyses from a regional promotion ran on a lighter AI model that scored systematically lower. Keeping them would have dragged every band down and broken the like-for-like comparison, so they are excluded.
Agency scans and investor packages are out. The set covers free scans, paid deep analyses, and follow-up improved iterations tied to founder ideas. Nothing else.
The viability score (0-100) is a composite of market size signals, competitive density, demand evidence, timing, and risk factors. The verdict (go/caution/no-go) follows from the score, and every data point links to its original source within the report. The primary-risk section uses the subset of analyses where a top-risk label was explicitly stored. All data is anonymized. No individual ideas, founder names, or company names are disclosed.
For the full validation methodology, read how Preuve AI validates startup ideas. For the scoring formula, see how viability scores work.
FAQ
What percentage of startup ideas pass validation?
In the current benchmark set, 17.5% of startup ideas scored 70 or above. Another 76.0% landed in the caution zone between 40 and 69, while 6.4% scored below 40.
What is the most common reason startup ideas fail validation?
Early-stage risk is now the top failure signal, flagged in 36.9% of tagged analyses: the idea is too vague to evaluate. Go-to-market risk ranks second at 25.7%, and competitive pressure ranks fourth at 11.5%, behind regulatory risk at 12.0%.
What is a good startup viability score?
The current benchmark set has a median viability score of 54 and an average score of 56.3. Ideas scoring 70 or above are high-viability, and fewer than 0.2% score above 90.
What is the startup failure rate in 2026?
In the current 2026 benchmark set, 6.4% of startup ideas receive a no-go verdict with a score below 40, up from 4.8% in the April edition. Most ideas, 76.0%, land in the caution zone, and 17.5% pass validation with a score of 70 or higher.
Do startup ideas score better after iteration?
Yes. Across 734 re-scored iterations in the benchmark set, founders who fixed flagged issues and ran the analysis again gained 8.9 points on average, and 75.2% of them improved their score.
How does Preuve AI calculate startup benchmarks?
Preuve AI scores startup ideas from 0 to 100 against live data sources including competitor databases, Google Trends, Reddit, Crunchbase, and community forums, with paid deep analyses cross-checking 50+ of them. This report starts from more than 6,000 completed analyses and computes every table on a cleaned, like-for-like comparison set, with the exclusions listed in the methodology.
Vincent
5 years in B2B growth, building Preuve AI in public. 82% of ideas it scores aren't ready, the point is finding out in 8 minutes, not 3 months.
Follow on X →
Building is expensive. Validation is free.
Run your idea through 10 AI agents before you write a line of code. Every claim source-linked.
