200 US Software Engineer Postings, Checked for Ghost-Job Signs
Compiled by Sharif-Ali Asadullayev (GetDiogenes, Baku), with analysis and drafting assistance from Claude (Anthropic).
We ran 200 recent US "software engineer" postings through Diogenes' ghost-job check, twice. About 1 in 5 were flagged as high-risk, and almost all for one reason: a staffing agency advertising for a client it didn't name. This is our AI's judgement of the postings, not proof that any job is fake.
At a glance#
We took 200 recent "software engineer" postings from a large US job board and asked Diogenes' AI whether each one showed the signs of a ghost job. We ran the whole set twice.
| Run 1 (21 Oct) | Run 2 (22 Oct) | Average | |
|---|---|---|---|
| Flagged high-risk | 47 of 200 (23.5%) | 42 of 200 (21.0%) | 22.3% |
| Average risk score | 33.6 / 100 | 29.4 / 100 | 31.5 / 100 |
- About 1 in 5 postings were flagged. 30 postings (15%) were flagged in both runs; 59 (29.5%) in at least one.
- Almost every flag had the same reason. In run 2, the main reason for 40 of the 42 flags was that a staffing agency had posted the job for an unnamed client, not the employer itself. The other two: a role "contingent on contract award", and a posting set to close on 1 January 2100.
- Older postings were flagged far more often. Of postings under a week old, 7% were flagged; of those 8 to 30 days old, 47%.
"Flagged" means the AI judged the posting high-risk for being a ghost job. Nobody confirmed any verdict with an employer. Read every number here as "how often postings look like this", not "how many jobs are fake".
What got flagged, and why#
Each posting got a risk score from 0 to 100 and one main reason. The scores split cleanly: in run 2, no posting scored between 31 and 59, so the cut-off between "high-risk" and "not" did not decide many cases.
| Main reason (run 2) | Flagged postings |
|---|---|
| Staffing agency posting for an unnamed client, or a general talent pipeline | 40 |
| Role depends on the company winning a contract | 1 |
| Posting set to stay open until 1 January 2100 | 1 |
| Total | 42 |
A staffing-agency posting is not fake by default. Agencies fill real jobs. But a posting for "our client" with no named employer, no team and a generic description can also be a way to collect résumés for roles that don't exist yet. From the posting alone, you can't tell which, and the AI scored that uncertainty as risk. Six more postings were noted as agency posts but scored low; some of those named the employer they were hiring for.
In run 1, about 30 of the 47 flags gave the same main reason. Most of the rest were "vague description" flags, which we think were partly caused by the short text the AI was given (see the method section).
Older postings, more flags#
| Posted | Postings | Flagged, run 1 | Flagged, run 2 |
|---|---|---|---|
| In the last 7 days | 117 | 12 (10%) | 8 (7%) |
| 8 to 30 days before | 47 | 24 (51%) | 22 (47%) |
| 31 to 60 days before | 36 | 11 (31%) | 12 (33%) |
This is mostly the same finding as above, seen another way. Agency postings sat in the older listings: in run 2, 9 of the 117 newest postings were classed as agency posts, against 23 of the 47 from 8 to 30 days old. In this sample, a posting still up after a month was more often an agency's standing ad than an employer's open seat.
What to do with this#
If the posting is from a recruiter or staffing agency, ask before you send your résumé:
- Who is the client, and can I apply on their careers page as well?
- Is there an approved opening right now, or are you building a shortlist?
- Who is the hiring manager, and when do interviews start?
- Will you send my résumé anywhere without asking me first?
A good recruiter answers these easily. If they won't say whether a real opening exists, treat it as a pipeline, and don't count it as an application.
For any posting: check the date, look for the same role on the employer's own site, and be wary of end dates far in the future or roles that depend on a contract not yet won.
One view from Reddit#
When we shared the first run on Reddit, one commenter replied: "Most are data mining. They resell the info that you give them for free." Our data can't test that. It shows that most flagged postings came from agencies that didn't name the employer. It says nothing about what happens to the résumés afterwards.
Method, limits and citing#
- Postings: 200 US listings from the Adzuna job-search API for the term "software engineer", sorted by relevance, posted within the last 60 days (25 August to 21 October 2026). 181 were in Adzuna's IT category. They came from 45 different employers and agencies.
- Scoring: each posting was scored by Claude Sonnet 5 (Anthropic), the model behind Diogenes' Job Check, using the Job Check's own instructions. It returned a risk score, a high-risk yes/no, and one main reason. No web search was used.
- What the AI saw: the title, the company name, and the first 500 characters of the description, which is all the Adzuna API returns. Run 2 also gave the AI today's date, the posting date and the location, and told it that cut-off text is not a red flag. Run 1 had none of these, and four of its flags rested on wrong dates (for example, calling a posting from the day before "in the future"). We fixed that for run 2.
- Run to run: the same 200 postings got the same verdict in 171 cases (85.5%). The AI's judgement varies on borderline postings, so we show both runs and their average.
Limits:
- One search term, one country, one job board, and the first 200 results by relevance. This is not a random sample of US jobs, and the shares are not the share of all jobs that are ghosts.
- 500 characters is a short excerpt. The full posting may answer questions the AI raised.
- The verdicts are AI judgement. We did not contact any employer or agency to confirm them.
- We don't publish company names. A flag is a warning sign about a posting, not a finding about a company.
We built GetDiogenes, which checks job postings for ghost-job and scam signs. This report uses that same tool, so read it as our tool's view of these postings.
Cite as: Asadullayev, S.-A., with Claude (Anthropic). 200 US Software Engineer Postings, Checked for Ghost-Job Signs. GetDiogenes, October 2026. getdiogenes.com/research/200-software-engineer-jobs