How many tech jobs are actually remote? We counted three ways: 13% to 18%

In August we published a remote-work number, said it was probably too high, and guessed at why. This is the check. We hold 857,052 live job postings read from employers' own career pages, and 32,329 of them fall into the eleven technical roles we track. Most of those carry three separate answers to the question of how many jobs are actually remote, and they do not agree.

The employer's own remote field says 18.0%. What the employer typed into the job title or location line says 13.2%. Our AI's read of the full description, on the 19,020 of those postings it has read, says 12.7% remote and another 20.4% hybrid.

So for tech the answer is somewhere between 13% and 18%, depending on which of the three you believe. All figures were pulled on 27 September 2026.

The short version of the correction: the per-role numbers we published in August mostly held up. The outside benchmark we measured them against was the wrong row of someone else's data, and our explanation for the gap was wrong too.

What we published in August, and where

Two places.

The August 2026 market report post said remote was 17.2% of ML engineering postings, 15.9% of data science and 24.8% of backend. It then said Indeed Hiring Lab puts "fully remote tech roles at around 8% in Q1 2026", that we were "more than double that", and that some of the gap was "probably us, reading 'remote-friendly' in a description as remote." It told readers to treat our remote numbers as an upper bound.

Our report methodology page carried the same comparison as a table row: "Share of tech postings advertised as fully remote", ours 19.5%, Indeed's about 8%, marked as the one place we were out of line.

Three things in that are wrong, and one is right.

The Indeed number was the wrong row. Indeed publishes its remote tracker as open data on GitHub, and we downloaded it this week. The US series that sits near 8% is the share of all Indeed postings, nurses and truck drivers included, that mention remote or hybrid work. It averaged 8.8% across Q1 2026 and read 8.4% on 31 August. It covers every kind of job, and it counts hybrid. The same file has a software development row, and that row averaged 33.9% in Q1 2026 and was 30.8% on 31 August.

The 19.5% was mislabeled. It was never a figure for all tech postings. It was the software engineer role alone, from the 20 August snapshot: 1,249 of 6,408 postings. And it was built by taking our AI's read of the description and, where the description was silent, falling back to the employer's remote field. That fallback matters, and the next section shows why.

Our explanation could not have caused the per-role figures. The 17.2%, 15.9% and 24.8% came from the location line. A posting counted as remote if its location said "Remote" and not "hybrid". No description was read to produce them, so "remote-friendly" in a job description had nothing to do with it.

What we got right was the suspicion itself. One of the ways we counted does run high.

Three signals, three answers

Here is each signal across the 32,329 tech postings.

The remote field is a yes/no box on the hiring system the employer posts through. It is the easiest to count, and it runs highest.

The text signal is what the employer actually wrote where a candidate looks first. It lands about five points lower.

Our read of the description is the only one of the three that can say "hybrid", and it says it a lot. Of the 19,020 tech postings it has read, 12.7% are remote, 20.4% hybrid, 20.2% on-site, and 46.7% never say clearly. That last share is up from the roughly 40% we published in August.

The fourth row is the strict version. The remote field says yes, nothing in the location line or the description contradicts it, and either the description agrees or the location line says remote or names only the country. That is the rule we already use before our own job pages are allowed to call a role fully remote. It gives 14.2%.

Now the software engineer figure, redone the same way it was built in August. Read-the-description-else-trust-the-field gives 15.8% today, on 9,875 read software engineer postings. The description alone, on those same postings, gives 10.6%, and the strict rule applied across all 16,863 software engineer postings, read or not, gives 10.5%.

So the method behind the 19.5% runs about five points high today. The gap is 509 postings where the description was silent and the remote field said yes, which is the fallback and nothing else. We cannot rerun 20 August, so we will not claim the August figure should have read 10.6%. We can say the method leans high. It also still feeds the remote chart on each of our role report pages, where the September figure for software engineers reads 16.6% "explicitly remote". That label overstates it for the same reason, and the chart has not been changed yet.

Where the signals disagree

5,812 tech postings have the remote field set to yes.

Most of the time something backs it up: in 4,033 of those (69.4%), the title or location line also says remote. In 1,718 (29.6%) nothing does. The location names a city, and a yes/no field behind it says remote.

Our AI has read 3,459 of the postings flagged remote. It agrees on 1,699 (49.1%). On 1,355 it found nothing either way. On 345 the description says hybrid and on 60 it says on-site. That is 405 postings, 11.7%, where the employer's own field and the employer's own description contradict each other.

It fails in the other direction as well. Of the 2,419 read postings our AI calls remote, 720 (29.8%) have the remote field set to no.

Hybrid is where the yes/no box breaks. A hybrid role has to be filed as one or the other, and our read found 3,883 hybrid tech postings. Of those, 3,427, or 88.3%, say nothing about it in the title or location. You only find out how many office days it wants by reading the body.

One more thing we checked, because it is tempting. Across all 857,052 postings, not just tech, the remote field says yes on 8.0%. That sits right next to Indeed's 8.4% all-jobs figure. It is a coincidence of two different measurements and we are not going to use it as a match.

Measured against outside sources, properly this time

Indeed first, like for like. Their tracker counts a posting as remote or hybrid if it uses remote, hybrid or flexible-work language, or if the location is listed as remote. The closest thing we have is remote plus hybrid from our description read: 33.1% of read tech postings, and 30.3% of read software engineer postings. Indeed's software development row read 30.8% on 31 August and averaged 33.9% in Q1 2026.

That is close, and nowhere near double.

The keyword rules differ and our read subset is not a random sample, so read it as the same range. We would not quote it to the decimal.

The number we genuinely cannot reconcile is Robert Half's. Their Q2 2026 analysis of US job postings, published 31 July 2026 and read by us on 27 September 2026, puts technology at 85% fully on-site, 11% hybrid and 4% fully remote. Our strict remote figure for tech is 14.2%, more than three times theirs. Their postings are mapped to the job titles in their own salary guide and come from a different source, and their technology bucket likely includes IT support and help desk roles we do not count as tech. That is our guess at the gap. We have not tested it, and the row stays open.

What this means if you only want remote work

• Trust the location line over a remote filter. A filter built on the yes/no field would show you postings where 29.6% of the time the employer never wrote the word remote anywhere a candidate looks first.
• Search the description for "hybrid" and "days". Across the read tech postings, one in five is hybrid, and almost nine in ten of those only say so in the body.
• Expect roughly one tech posting in seven to be remote once two signals agree. That is the 14.2% strict figure. For backend engineering it runs higher, 35.5%. For the software engineer title it is 10.5%.
• A flagged-remote posting in a named city is worth one question before you spend an hour on the application. Ask the recruiter whether the role is remote for your state.

If you care about visas as well as location, most postings are silent on that too. We counted how often in our sponsorship disclosure post. And if you are sorting by age at the same time, read the posting lifespan numbers before you skip a remote listing for being two weeks old.

By role

Same four measures, one row per role. The first three columns are shares of every posting in the role. The last two are shares of postings our AI has read.

Measured the way the August post measured them, from the location line alone, ML engineering is now 16.2% against the 17.2% we published, data science 12.9% against 15.9%, and backend 30.4% against 24.8%. Two moved down, one moved up. The populations differ, since August counted 12,166 read postings and this counts all 32,329, and a month has passed. None of it looks like a figure inflated by misreading descriptions.

How we counted, and what it cannot tell you

What was counted. Every live posting in our corpus on the evening of 27 September 2026, 857,052 of them, counted in full in two passes about ten minutes apart. The corpus is live, so the total moved by 3,566 between passes. The tech count, 32,329, did not. The total is also far above the 668,970 in our September report a week earlier. That is real growth, not a changed definition: about 300,000 of today's postings arrived after that report, mostly from employers on two career-site systems we started reading this month, and only about 3% of those new postings are technical roles. So the tech base grew much less than the headline. Tech means one of eleven role families matched from the job title, the same classifier behind the August report.

The three signals. The remote field is whatever the employer's hiring system stores as its yes/no remote flag, taken as given. The text signal is a word match on the title and location line for remote, work from home, telecommute and similar, where any mention of hybrid wins and a negation like "not remote" does not count. The description read is our AI's classification of the full job description into remote, hybrid, on-site or unclear.

What that cannot tell you. The description read covers 19,020 of 32,329 tech postings. Which postings get read depends on our own job matching, not on chance, so its shares describe the postings we have read. They may not hold for the rest. The text match is crude by design. "Remote" in a location line like "New York or Remote" counts as remote even when the employer would prefer New York. And none of the three knows what a hiring manager will actually agree to once you are talking.

This is one evening's snapshot. It replaces both the August paragraph and the methodology row as the figures we stand behind, and we will redo it when the next market report runs.