AI recommended just 98 of 1,437 YC startups to their own buyers
When a buyer asks AI what to use, most young startups aren’t in the answer. We asked AI the question each of 1,437 Y Combinator startups’ buyers would ask, without naming anyone. It recommended 98 of them (6.8%). For 1,262, it recommended other companies and never the startup.
Key findings
- Most startups are left out. AI recommended 98 of 1,437 YC startups from the 2024–2026 batches. For 79% of startups it named at least 8 other companies across its two answers.
- The younger the startup, the less AI recommends it: 12.8% of 2024 startups, 6.5% of 2025’s and 2.7% of 2026’s.
- AI rarely looks at a startup’s own site. Its searches reached the startup’s website in 85 of 1,437 checks. When they did, it recommended the startup 74% of the time; when they didn’t, 2.6%.
- It recommends what it already knows: OpenAI, Anthropic, Cursor, LangGraph, HubSpot and Salesforce came up again and again for startups’ buyer questions.
- Its answers are built from other people’s pages: Medium, YouTube, Reddit and GitHub were the sites it read most.
1. Most startups aren’t in the answer
For each startup we asked the question its buyers would ask, like “what’s the best tool for X”, twice. AI recommended the startup in at least one answer for 98 of 1,437 (6.8%), and in both answers for only 51 (3.5%). For 1,262 startups (88%) it recommended other companies instead, and for the remaining 77 it gave general advice without naming anyone.
That’s the moment a buyer builds a shortlist. If AI names eight other tools and not yours, you’re not on it.
2. The younger the startup, the less AI recommends it
Startups from 2024 batches were recommended 12.8% of the time (53 of 414). That fell to 6.5% for 2025 (30 of 460) and 2.7% for 2026 (15 of 563). Older startups have had more time to collect reviews, roundups, tutorials and threads, which is what AI searches for. A startup that launched this year has almost none of that yet.
3. When AI finds a startup’s own site, it usually recommends it
AI’s searches reached the startup’s own website in only 85 of 1,437 checks (5.9%). In those, it recommended the startup 63 times (74%). In the other 1,352, it recommended the startup 35 times (2.6%).
This is a correlation, not proof that being read causes the recommendation: AI probably reads a site because it already looks relevant. But it shows where the gap is. For most startups, nothing AI finds when it searches the buyer’s question leads back to them.
4. Who AI recommends instead
Across 1,437 checks AI named 6,080 different companies and products instead of the startup. These came up most often, counted as the number of startups whose buyers were pointed to them:
| Recommended instead | Startups |
|---|---|
| LangGraph | 44 |
| OpenAI | 43 |
| Cursor | 41 |
| Anthropic | 38 |
| ChatGPT | 36 |
| Claude | 34 |
| CrewAI | 30 |
| Claude Code | 29 |
| n8n | 29 |
| HubSpot | 26 |
| Salesforce | 24 |
| Langfuse | 23 |
| GitHub Copilot | 23 |
| vLLM | 22 |
| AutoGen | 22 |
Most are large, well-documented companies and open-source projects. For a young startup, the real competition in an AI answer is usually not another startup. It’s the name everyone has already written about.
5. Where AI’s answers come from
In 1,232 checks AI cited the pages it read, on average about 9 different sites each. Medium alone was read for 221 startups’ buyer questions (18% of checks with sources). These are the pages that decide the answer, and getting onto them, through a guest post, a tutorial, a comparison or an honest thread, matters as much as your own site.
6. By category
| Category | Recommended | AI picked most often instead |
|---|---|---|
| Security | 6 of 44 | NeMo Guardrails, Llama Guard, GitHub Advanced Security |
| Legal | 4 of 33 | Harvey AI, Spellbook, Lexis+ AI |
| Finance and Accounting | 4 of 41 | HighRadius, Vic.ai, BlackLine |
| Supply Chain and Logistics | 4 of 42 | Flexport, Alvys, NetSuite |
| Consumer (general) | 7 of 76 | ChatGPT, Google Gemini, Claude |
| Infrastructure | 13 of 146 | E2B, Modal, Anthropic |
| Insurance | 3 of 34 | FurtherAI, Gradient AI, Akur8 |
| Sales | 3 of 36 | Clay, Gong, Apollo.io |
| Engineering, Product and Design | 18 of 246 | Cursor, Claude Code, Windsurf |
| Fintech (general) | 3 of 44 | Wise, Payoneer, 3Commas |
| Operations | 3 of 51 | n8n, Jobber, ServiceNow |
| B2B (general) | 14 of 369 | Anthropic, LangGraph, Microsoft Copilot Studio |
| Analytics | 1 of 34 | BigQuery, Databricks, Snowflake |
| Marketing | 1 of 35 | Claude, ChatGPT, HeyGen |
| Productivity | 1 of 55 | LangGraph, n8n, Aider |
By industry the gap is small: Consumer 8.2%, B2B 6.7% and Fintech 6.4%. Category pages have every startup’s result.
What founders can do
- Write the page for your buyer’s question. One clear page that answers “what’s the best tool for X” honestly, with you in it and a fair comparison.
- Make sure AI can read it. GPTBot and most AI crawlers don’t run JavaScript. Check what they see with the free GPTBot view.
- Get onto the pages AI already reads. Roundups, tutorials, Medium and dev.to posts, GitHub READMEs and real community threads in your category.
- Check again. Answers change. Track the question and see whether you start getting named.
Method
- Startups: 1,437 companies from the public Y Combinator directory (yc-oss list), batches 2024–2026, industries B2B, Fintech and Consumer. Biotech and hardware were left out because their buyers ask research questions no product answers.
- The question: written once per startup by AI from the startup’s own description, as its buyers would ask it, never naming the startup.
- The answers: an AI model with live web search, asked each question twice, between 4 and 12 November 2026 (the same model for all but 2 setup checks).
- “Recommended”: the startup is named in an answer, confirmed both by an AI grader and by the name appearing in the answer text. The grader also listed who was named instead (up to 8 per check).
- Sources: the domains of the pages the AI’s web search returned and it cited; the startup’s own site is counted separately.
Limits. One AI model, two answers per startup, one question each, one point in time. ChatGPT, Perplexity and Claude may answer differently, and AI answers vary from run to run. Questions written by AI may not match every buyer’s wording. We don’t publish what AI said about individual startups from memory: those are a model’s guesses about named companies.
Data and citing this study
The totals behind every number here are in ai-startup-study-2026.csv (CSV, CC BY 4.0). Live results for each startup and category are on the startup study pages.
PatchMySEO, “AI recommended just 98 of 1,437 YC startups to their own buyers”, November 2026. https://patchmyseo.com/blog/ai-recommends-yc-startups-2026
Does AI recommend your startup?
YC startups can look up their result. Anyone else can add their startup and we’ll check on day 0, 7 and 14 whether AI recommends it.
Find your startup