One answer in sixteen: how often AI search names a young company

When a buyer asks an AI assistant what tool does a job, how often does the answer name a company that launched in the last few years? We measured it on a fixed cohort, with a published method, and the number is one answer in sixteen.

What we measured

A cohort of 118 Y Combinator companies from the 2021 to 2026 batches, drawn by a seeded random draw from YC's public directory, and 18 of YC's own designated top companies as the control. Both legs from the same institution, so the comparison is young against graduated rather than startups against household names, and founder quality, funding access and admission standard are held roughly constant.

Each company got thirty buyer questions built from its own homepage with the brand removed, asked of Google AI Overviews, Perplexity, Gemini and ChatGPT on the same day. A mention is the company's name appearing in the answer. Detection is recall-first: a close call counts as a mention, which if anything flatters the young leg.

The number

A gap of 5.7 times, 30.3 points, and it holds on every engine: the young leg sits between 6.0% and 6.8%, the graduates between 30.6% and 39.4%.

The distribution says more than the average. Forty-eight of the 118 young companies were never named once, across 76 answers each about what they sell. One young company in 118 cleared 50%. Seven of the eighteen graduates did.

On the eight questions every engine was asked identically, 84% of young-company questions were named by no engine at all. For the graduates, 40%.

Time helps, then it stops

The young leg is split evenly across six batch years, so time since launch is a variable rather than a confound.

Companies a year old sit near 2.5%, and two thirds of the newest batch was never named at all. By two to three years out the rate has climbed to roughly one answer in ten. Then it stops: the 2021 batch is no higher than 2023. Time alone lifts a company from invisible to one answer in ten and no further. What carries the graduates to four in ten is something other than age.

Among the graduates the same slope shows from the other end. YC's top companies from 2009 to 2013 run 37% to 67%; from 2014 to 2017, 17% to 36%. Even YC's designated best from the last decade mostly sit under 40%.

Who clears the bar young

Nine young companies were named in a quarter or more of answers. They share two shapes. A narrow, nameable category where the engines have few names to reach for: voice-agent testing, architecture diagramming, government contracting software, exam preparation. Or an open-source project with a public repository. The one young company over 50% is a 2022 fintech API.

The engines do not disagree about young companies, which is its own finding: for the young leg they all sit within three points of one another. Engine choice barely matters while you are unknown, and starts to matter once you are known, where they spread across twelve points.

What this does and does not say

• It is YC. It does not generalise to startups outside YC, and no figure here claims it does.
• The frame contains only survivors: YC marks the rest acquired or inactive. The older the batch, the more selected it is. Roughly three in ten of the 2021 batch is already out of the comparison, so the 2021 figure is about the companies that made it.
• The question is the instrument. Every figure is from questions built from homepage text with the brand removed. A different construction gives different numbers; our own pilot showed the swing between two constructions can be sevenfold. The construction is versioned and published.
• Perplexity's answers for most companies were collected some hours after the other engines', after an account quota lapse mid-wave was topped up and re-fired. Every pooled figure excludes errored answers rather than counting them as zero.
• The control leg is 18 companies. Its per-stratum figures are directional.

The questions to ask of any figure like this

Where did the questions come from? How many answers, not scans? What counts as a mention, and how generous is it? Who was in the sample, and how old are they? Our answers are on the method page; the cohort list and the question set are published with it.

AfterLaunch runs this measurement continuously for the companies it works with. The method above is the one behind every figure on this site.