Newshot

AI assistants never named 74% of the 224 YC startups we tested

Original text(via HN AI)

Citefold studyUpdated November 18, 2026YC batches W24 to F26

How visible are YC startups to AI assistants?

For each of 224 Y Combinator companies we wrote 4 questions a buyer in its category would ask, and put them to ChatGPT, Perplexity and Gemini. That gave us 2,688 answers. 74% of the companies were not named in any of them.

SharePost to XPost to LinkedIn

Is your startup in the study?

The numbers

Each percentage is a share of the 224 companies we scanned.

  • 74% were never named166 of the 224 companies did not appear in any of their 12 answers. 36 appeared once. Only 7 appeared three times or more, and the best result was 5 of 12.

74% were never named

166 of the 224 companies did not appear in any of their 12 answers. 36 appeared once. Only 7 appeared three times or more, and the best result was 5 of 12.

  • Without web search, a young company does not existAnswering from what the model already knows, ChatGPT named the company in 0.9% of answers and Gemini in 0.1%. These companies are from 2024 to 2026 batches, so most are newer than the training data. Perplexity searches the web for every answer and named them in 9.2%. We reran 39 of the companies on ChatGPT with web search switched on: it named them in 9.0% of answers, and 23% of those companies at least once.

Without web search, a young company does not exist

Answering from what the model already knows, ChatGPT named the company in 0.9% of answers and Gemini in 0.1%. These companies are from 2024 to 2026 batches, so most are newer than the training data. Perplexity searches the web for every answer and named them in 9.2%. We reran 39 of the companies on ChatGPT with web search switched on: it named them in 9.0% of answers, and 23% of those companies at least once.

  • 48% already publish an llms.txtThat is far more than we expected. It goes with a small difference: 30% of companies with an llms.txt were named at least once, against 22% of those without. The wider gap is for comparison pages. 45% of companies with a "vs" or alternatives page were named, against 20% of those without, and 77% have no such page. These are differences in one sample and do not prove cause.

48% already publish an llms.txt

That is far more than we expected. It goes with a small difference: 30% of companies with an llms.txt were named at least once, against 22% of those without. The wider gap is for comparison pages. 45% of companies with a "vs" or alternatives page were named, against 20% of those without, and 77% have no such page. These are differences in one sample and do not prove cause.

  • The answers are built from other people's pagesWhen Perplexity searched, it cited github.com in the reports of 92 companies, dev.to in 46 and g2.com in 36. An assistant that searches reads pages like these first. A startup that is absent from them is usually absent from the answer.

The answers are built from other people's pages

When Perplexity searched, it cited github.com in the reports of 92 companies, dev.to in 46 and g2.com in 36. An assistant that searches reads pages like these first. A startup that is absent from them is usually absent from the answer.

How many of the 12 answers named the company

Each company got 12 answers. The number above a bar is how many companies were named in that many of them.

图表:Answers that named the company, out of 12

How often each engine names them

The bar is the share of that engine's answers that named the company the question was written for.

  • ChatGPTmodel knowledge, 896 answersNamed 3% of companies at least once0.9%
  • Perplexitylive web search, 896 answersNamed 24% of companies at least once9.2%
  • Geminimodel knowledge, 896 answersNamed 0.4% of companies at least once0.1%

Same questions, ChatGPT with web search on (39 companies, 156 answers per engine)

  • ChatGPT with searchlive web searchNamed 23% of companies at least once9.0%
  • Perplexitylive web searchNamed 15% of companies at least once4.5%
  • ChatGPT without searchmodel knowledgeNamed 3% of companies at least once0.6%

What the named companies have in common

Each pair compares companies that have something on their site with companies that do not. The bar is the share of each group that was named at least once.

These are differences in a small sample, not proof of cause.

Comparison or alternatives pages

  • Has them51 companies45%
  • Has none173 companies20%

llms.txt

  • Publishes one108 companies30%
  • Has none116 companies22%

Structured data on the homepage

  • Has it76 companies28%
  • Has none83 companies23%

Question-and-answer content

  • Has it82 companies33%
  • Has none99 companies21%

XML sitemap

  • Has one172 companies27%
  • Has none52 companies21%

How each batch does

The share of companies in each batch that were named at least once.

  • W2418 companies, average score 3511%
  • S2417 companies, average score 3629%
  • W2514 companies, average score 3521%
  • X2518 companies, average score 3939%
  • S2519 companies, average score 3126%
  • F2514 companies, average score 3821%
  • W2628 companies, average score 3329%
  • X2619 companies, average score 3432%
  • S2640 companies, average score 3125%
  • F2634 companies, average score 3224%

How each industry does

The share of companies in each industry that were named at least once.

  • B2B201 companies, average score 3425%
  • Fintech18 companies, average score 3233%

How the scores spread

The AI visibility score runs from 0 to 100. It combines how often the engines name a company, whether AI crawlers can read its site, and whether its content answers buyer questions. The median is 32.

图表:AI visibility score, share of companies

What their sites get right and wrong

The share of companies that passed, got a warning on, or failed each technical check.

PassWarningFail

  • robots.txt lets AI crawlers in72% pass
  • Firewall lets AI crawlers through91% pass
  • Content is readable without JavaScript80% pass
  • llms.txt present48% pass
  • Structured data (schema.org)34% pass
  • XML sitemap77% pass
  • Title and meta description82% pass
  • One clear H185% pass
  • Question-and-answer content37% pass
  • Comparison and alternatives pages23% pass
  • Clear company identity21.2% pass
  • Server response time100% pass

Who the engines recommend instead

The brands named most often. Answers is how many of the 2,688 answers named the brand. Companies is how many of the 224 company reports it appeared in.

BrandAnswersCompanies
LangSmith8819
Arize Phoenix7013
Cursor5511
Langfuse5411
Claude Code5112
ChatGPT4816
OpenAI4815
Claude4817
Helicone4712
Braintrust4514
Fireworks AI449
OpenRouter439
Zapier4110
Together AI418
Modal3810

The sources they cite

The domains cited most often when an engine searched the web. Citations is the number of times the domain was cited. Companies is how many of the 224 company reports cited it.

DomainCitationsCompanies
github.com67992
dev.to21746
arxiv.org17837
claude.com17219
g2.com13236
sourceforge.net11227
microsoft.com10229
slack.com908
salesforce.com8414
hubspot.com7918
openai.com7820
amazon.com7313
f6s.com6718
zapier.com6619
linkedin.com6114

The 7 most visible YC startups

Ranked by AI visibility score. Named is how many of the answers written for a company's category named it.

RankCompanyBatchCategoryNamedScore
1Locus paywithlocus.comF25pay-per-call API marketplace for AI agents5 of 1268
2Tracecat tracecat.comW24open source security automation platform (SOAR)4 of 1264
3Context.dev context.devS26web scraping and search api for ai agents4 of 1262
4Bluejay getbluejay.aiX25testing and monitoring platform for voice AI agents3 of 1260
5Aravolta aravolta.comX25data center infrastructure monitoring software3 of 1258
6Rebulk rebulk.comS25transload and rail operations software3 of 1248
7Saphira AI saphira.aiS24AI safety engineering software for robots3 of 1244

How we did it

Updated November 18, 2026

Sample: 224 Y Combinator companies from the W24 to F26 batches, taken from the public YC directory. We started with US companies that list a contact address on their own site, and 201 of them are B2B. This is not a random sample of YC.

Questions: for each company we read the homepage and wrote 4 questions a buyer in its category would ask before knowing the company exists, such as "what is the best X for Y". The questions do not contain the company name. They were written by a language model and a few will have missed the category.

Engines: each question went to ChatGPT (GPT-5.6) and Gemini (Gemini 3.1) through their APIs with no web search, and to Perplexity (Sonar), which always searches. That is 12 answers per company. With no search, a model can only name companies it saw in training, so near-zero results for 2024 to 2026 startups are partly a training-cutoff effect.

Search subsample: we picked 39 of the companies at random and asked ChatGPT the same 4 questions with web search on (156 answers). On those same companies Perplexity named 4.5% of answers and ChatGPT without search named 0.6%.

API answers can differ from what the consumer apps show, and answers change from run to run. Each question was asked once per engine.

Named means the company or product name appears in the answer as an option. A link to the site without the name does not count.

Site checks: 12 automated checks on each homepage and its robots.txt, sitemap and llms.txt. A check we could not verify is reported as a warning and is not counted as a fail.

All scans ran on 18 November 2026. Every company in the study has a report page, and founders can look theirs up on this page. If your report got your category wrong, email [email protected] and we will rerun it.

Questions or corrections: [email protected]

SharePost to XPost to LinkedIn

Run the same scan on your site

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by HN AI
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗