How ready are YC startups for AI search?
TrySEO ran its free site checker on 1,000 randomly chosen Y Combinator startup websites to see whether ChatGPT, Claude, Perplexity and Google can read them. Here's what we found.
Checked on 1 November 20261,000 sites24 checksDownload the data (CSV)
43%
publish an llms.txt file for AI assistants
17%
show AI agents fewer than 150 words of text
44%
have no structured data on their home page
83
is the median AI readiness score, out of 100
What we found
Each finding is a complete sentence, so you can quote it as it stands. Please link to this page when you do.
43%
43% of the YC startup websites we checked publish an llms.txt file, the plain-text summary written for AI assistants.
It's a young convention, first proposed in 2024. Sites without one leave assistants to work out what they do from their pages.
Free fix: llms.txt generator47%
YC startups from 2026 batches are more likely to publish an llms.txt file than those from 2022: 47% of the 2026 startups we checked do, against 36% of those from 2022.
The share rises with every batch year in between. The table further down has each year's figures.
Free fix: llms.txt generator<1%
Fewer than 1% of the YC startup websites we checked use robots.txt to block at least one of the crawlers ChatGPT, Claude and Perplexity use to find and open pages.
A blocked search crawler keeps a site out of that assistant's answers. The other 997 sites let every one of them in.
Free fix: AI crawler checker7%
7% of the YC startup websites we checked block at least one AI training crawler in robots.txt, and Bytespider is the one blocked most often.
Blocking training crawlers keeps pages out of model training, not out of AI answers. 6% of sites block training crawlers while still letting every search crawler in.
Free fix: AI crawler checker17%
17% of the YC startup websites we checked serve a home page with fewer than 150 words of text in its HTML, and 7% send an empty app shell that only JavaScript fills in.
Most AI agents read the HTML and don't run JavaScript, so on these sites they have little or nothing to read.
Free fix: Site checker44%
44% of the YC startup websites we checked have no JSON-LD structured data on their home page.
Structured data states who you are and what you sell in a format search engines and AI read without guessing.
Free fix: Schema markup generator40%
40% of the YC startup websites we checked describe their company with Organization markup that links to its profiles elsewhere, such as LinkedIn or X.
Those links let AI tie a brand name to its accounts, reviews and press, instead of guessing which company is meant.
Free fix: Schema markup generator22%
22% of the YC startup websites we checked have no XML sitemap that we could find at /sitemap.xml or in robots.txt.
Without one, crawlers find pages only by following links, and new pages take longer to show up.
Free fix: Site checker21%
21% of the YC startup websites we checked offer a Markdown version of their home page to AI agents that ask for one.
Agents that ask for Markdown get clean text instead of HTML to pick apart. It's newer than llms.txt and less common.
Free fix: Site checker6%
6% of the YC startup websites we checked publish their prices as plain text at /pricing.md or /pricing.txt for AI agents to quote.
A plain-text /pricing.md lets agents comparing options quote prices exactly, without decoding a pricing page.
Free fix: Site checker39%
39% of the YC startup websites we checked have a meta description that is missing, too short or too long for search results.
When the description doesn't fit, search engines and AI pick their own snippet from the page.
Free fix: Title and meta writer27%
27% of the YC startup websites we checked have no H1 heading on their home page, or more than one.
One clear H1 tells crawlers what the page is mainly about.
Free fix: Site checker83
The median YC startup website we checked scores 83 out of 100 for AI readiness, and 1% pass every check we score.
The score counts a pass as a full point and a check that needs work as half, across the checks in the table below.
Free fix: Site checker
Which AI crawlers get blocked
The share of the 1,000 sites whose robots.txt shuts each crawler out of the home page. Search and user crawlers decide whether an assistant can cite a site; training crawlers only decide whether its pages train models.
AI search
- OAI-SearchBot · OpenAI<1%
- Claude-SearchBot · Anthropic<1%
- PerplexityBot · Perplexity<1%
Fetches for a user
- ChatGPT-User · OpenAI<1%
- Claude-User · Anthropic<1%
- Perplexity-User · Perplexity<1%
AI training
- GPTBot · OpenAI6%
- ClaudeBot · Anthropic5%
- Google-Extended · Google6%
- Applebot-Extended · Apple5%
- CCBot · Common Crawl6%
- Bytespider · ByteDance6%
- Meta-ExternalAgent · Meta5%
Bars run from 0 to 10% of sites.
How the sites score
Each site's AI readiness score, out of 100: a passing check counts fully, one that needs work counts half and a failing one counts nothing. The median is 83.
- 0 sites scoring 0 to 9
- 0 sites scoring 10 to 19
- 0 sites scoring 20 to 29
- 1 sites scoring 30 to 39
- 21 sites scoring 40 to 49
- 72 sites scoring 50 to 59
- 141 sites scoring 60 to 69
- 199 sites scoring 70 to 79
- 235 sites scoring 80 to 89
- 331 sites scoring 90 to 100
Sites in each score range
Newer startups and older ones
The same measures for each year's YC batches. Each year has fewer sites than the whole sample, so small gaps between years can be chance.
| Batch year | Sites | llms.txt | Markdown version | Structured data | Under 150 words in the HTML | Block an AI search crawler | Median score |
|---|---|---|---|---|---|---|---|
| 2022 | 169 | 36% | 15% | 55% | 12% | 0% | 80 |
| 2023 | 138 | 41% | 20% | 51% | 16% | 0% | 83 |
| 2024 | 195 | 42% | 22% | 54% | 21% | 1% | 80 |
| 2025 | 219 | 44% | 24% | 55% | 19% | 0% | 83 |
| 2026 | 279 | 47% | 23% | 61% | 16% | 1% | 85 |
All 24 checks, totalled
The share of sites with each result. Not scored marks an optional extra the site doesn't have, or a choice that's up to the owner, such as blocking AI training.
| Check | Passed | Needs work | Failing | Not scored |
|---|---|---|---|---|
| Crawlers and access | ||||
| HTTPS | 100% | – | <1% | – |
| Search indexing | 100% | – | 1% | – |
| robots.txt | 85% | 15% | – | – |
| AI search crawlers | 100% | – | <1% | – |
| AI training crawlers | – | – | – | 100% |
| Sitemap | 78% | – | 22% | – |
| Ready for AI agents | ||||
| Content in the HTML | 83% | 5% | 12% | – |
| llms.txt | 43% | 57% | – | – |
| llms-full.txt | 13% | – | – | 87% |
| Markdown version | 21% | – | – | 79% |
| pricing.md | 6% | – | – | 94% |
| Page tags | ||||
| Title tag | 78% | 21% | 1% | – |
| Meta description | 61% | 34% | 5% | – |
| H1 heading | 74% | 11% | 16% | – |
| Canonical tag | 69% | 30% | 1% | – |
| Social preview | 77% | 11% | 12% | – |
| Page language | 96% | 5% | – | – |
| Structured data | ||||
| Structured data | 56% | – | 44% | – |
| Organization details | 40% | 9% | 51% | – |
| What you offer | 24% | 76% | – | – |
| Content and trust | ||||
| Subheadings | 79% | 21% | – | – |
| Image descriptions | 37% | 23% | 40% | – |
| About and contact | 43% | 35% | 22% | – |
| Server response | 80% | 18% | 2% | – |
How we ran the study
- Sample
- 2,785 active companies from Y Combinator's Winter 2022 to Fall 2026 batches that list a website of their own, from the public YC company directory (the open yc-oss copy, downloaded on 1 November 2026). We put them in a random order with a fixed seed and checked them in that order until 1,000 had loaded. 36 of the 1,036 we tried wouldn't load for our checker (down, timing out or blocking automated visits) and were replaced by the next in line.
- When
- Every site was checked on 1 November 2026.
- What we checked
- The same 24 checks as our free site checker: each site's home page, robots.txt, XML sitemap, llms.txt, llms-full.txt, a Markdown copy of the home page and /pricing.md. Every check is a fixed rule; no AI was involved. We re-fetched a sample of the llms.txt files and Markdown copies and read them, to make sure they were real files rather than error pages.
- How we visited
- A few sites at a time, from one location, with a user agent naming TrySEO-Research and linking to this page. Response times include the distance from that location to each server.
- Limits
- We looked at home pages only, and at the robots.txt rules that cover them. A host or CDN can also block AI crawlers at the network level, which robots.txt doesn't show, so real blocking may be higher than these figures. Sites that turned our checker away were replaced, so sites with strict bot protection are under-represented. The YC directory shows the website each company lists, which may not always be current.
- Privacy
- We publish totals only and never name a site. Per-site results aren't published.
Use the data
The totals behind every figure on this page, including each check, each AI crawler and each batch year, are free to use under CC BY 4.0. Credit TrySEO and link to this page.
Download the data (CSV)Cite this study
TrySEO (2026). How ready are YC startups for AI search? An AI readiness study of 1,000 Y Combinator startup websites. https://tryseo.org/research/ai-readiness-2026
Run the same 24 checks on your site
The free site checker behind this study. See where your site stands in about 10 seconds, with the fix for every issue. No sign-up.
Rather have it all done for you?
TrySEO writes the content AI assistants quote, builds backlinks and tracks how they talk about you, every month. Try it free for 3 days.