90.2% of the job postings we read say nothing about visa sponsorship

We hold 761,645 US job postings. 668,970 of them are live today. We have read the full description text of 139,508 of those live postings, which is 20.9% of them, and counted how many say anything at all about visa sponsorship. Everything below was counted on 20 September 2026.

The read set is not a random sample of the live set. These are the postings our pipeline had reason to open in full, and what gives a posting that reason correlates with the employer, with the sector, and with how the posting reached us in the first place. So the figure that follows is a true statement about the postings we have read. It is not a safe estimate for every job advert in America. We are not going to launder it into one.

The count

Of the 139,508 live postings we have read in full:

• 125,906 say nothing at all about visa sponsorship. That is 90.2%.
• 12,193 say explicitly that sponsorship is not available. That is 8.7%.
• 1,409 say explicitly that sponsorship is available. That is 1.0%.

Of the adverts we have read, nine in ten are silent on the one question that decides whether an international candidate is wasting the afternoon.

The shape repeats on postings that have since closed. We have read 11,827 of those. 10,731 were silent, 1,044 said no in writing, and 52 said yes. Those are a different set of adverts, for roles nobody can apply to any more, and silence dominates them just as heavily.

Silence is not a no

This is the part that gets misread, so here it is flatly.

A posting that says nothing about sponsorship may come from an employer who sponsors every year and never thought to write it down. It may come from an employer who will end your application at the first screening question. It may come from one that decides case by case, on the role, on the candidate, and on what their lawyer says that quarter. The advert does not tell you which. You cannot tell from the outside.

The 90.2% measures one thing, and it is not the thing people quote it as. It measures how little information the advert gives you.

It does not measure how many employers sponsor. Those are two different quantities and only the first one is in our data. The single figure here that maps to employer behaviour is the 8.7% who took the trouble to rule it out in writing. Those employers told you something real. The other 125,906 told you nothing.

What employers filed, against what they wrote

We also hold 97,662 approved H-1B sponsorship records covering fiscal years 2024 through 2026, our copy of the public employer disclosure data published by US Citizenship and Immigration Services.

Put that next to 1,409. The two do not sit comfortably together.

Sponsorship happens often enough to fill a public dataset with 97,662 records across three fiscal years. It shows up as an offer in 1,409 of the 139,508 adverts we read. Both of those statements are true at the same time, so employers can sponsor at some scale while rarely writing it into the advert.

The two are measuring different things and that is the point rather than a problem. The immigration records describe what employers did, after the fact, under a filing obligation. Our counts describe what they wrote in the advert. Behaviour and text are not the same variable, and nothing in our numbers contradicts the immigration data.

We cannot join the two reliably. The employer name on a job advert and the employer name on an immigration filing are often different strings for the same company, and the failure mode of a loose match is not a missing row, it is a wrong row that reads like a fact. So we cannot tell you what share of the silent 90.2% comes from employers who do in fact sponsor. That is the number you actually want. We do not have it, and until the name matching survives being attacked we are not going to publish an estimate of it.

One more limit on the cross-check itself. It sits between two datasets we hold, not against an independent count of sponsorship language in job adverts. That is weaker than being checked by somebody else's measurement, and it is the main reason to read 90.2% as ours rather than as the market's.

What to do with this if you need sponsorship

Filtering a job board on "sponsorship available" throws away about nine in ten of the postings we have read. Almost all of what you throw away is silence, not refusal. Most of that filtering buys you no protection at all. It hides 125,906 postings whose position on sponsorship nobody has told you either way.

The 8.7% are the different case. Those employers wrote the answer down and the answer was no. Filter those out and you have removed the ones that really are closed to you, which is what a filter is for.

Then ask, early, before you have spent weeks on a pipeline that cannot close. The question is usually put to you by the software the employer collects applications through rather than by a person reading your CV, and on the form it reads something like "will you now or in the future require sponsorship" (this is how that software works). Answer it honestly. A recruiter telling you no in week one costs you one email. Screening yourself out of nine in ten of the postings we have read, in advance and on no evidence, costs you the search.

If you would rather start from employers than from postings, /companies lists the ones we track alongside their live roles. Our August 2026 read of the US tech market covers the same corpus from the seniority angle, and the résumé guidance here is written against the same postings.

How we counted, and what it cannot tell you

What was counted: the description text of live US job postings we hold, classified into three buckets. Says sponsorship is available. Says it is not. Says nothing either way. A posting lands in the third bucket if the description never raises the subject.

What was excluded: every live posting we have not read in full, which is most of them. Also anything outside the description text. If an employer keeps its sponsorship position in the application questions, or on a benefits page, or in a recruiter's head, our count reads that posting as silent, because from the text of the advert it is.

The window: a full census of the read set as it stood on 20 September 2026. Within that set nothing is sampled. The set itself is the sample problem, and it is not a random one.

What it cannot tell you:

• Whether any particular employer sponsors. Silence in the advert is not evidence in either direction.
• What the true rate is across all 668,970 live postings. We have read 20.9% of them, chosen by our pipeline rather than by a coin.
• How often the classifier is wrong. Our AI does the reading, a machine reading of text is not a perfect reading of it, and we have no audited error rate to quote for these three buckets.

So: filter out the explicit no, keep the silence, and ask the question in the first conversation you get. The 8.7% is the only one of the three numbers that tells you what an employer will actually do.