The Problem in One Sentence
Most AI job boards optimise for the number of listings, not for the number of people who actually get hired, and everything that follows from that incentive is noise.
If you have searched for an AI role lately, you have seen the symptoms. The same job appears under three company names. A posting has been open for eight months. A "quick apply" turns into a ten-minute questionnaire that ends by asking for your CV with no employer attached. You apply, hear nothing, and the role is reposted the following week with a new date.
This is not bad luck. It is the predictable output of a system where listings are scraped, republished, and counted as inventory. Once you know the patterns, they are easy to see. This piece walks through the five we see most often, then explains what we changed to avoid them.
Five Patterns That Fill a Feed With Noise
1. The same job under many company names
A recruiter or aggregator takes one real opening, rewrites the title slightly, and publishes it across several employer names or generic "talent" fronts. The job is real, but you are now competing against your own application three times, and you cannot tell which version is the actual employer.
The giveaway is a company name you cannot verify, no company domain, and an apply link that goes to a form rather than a careers page.
2. Evergreen listings that never close
Some postings stay open for months, refreshed so they always look new. Sometimes the role is genuinely hard to fill. Just as often it is a pipeline listing: the company wants a steady flow of CVs for a position that may or may not exist, or the listing simply never expires on the source board.
Age is the honest signal here. A role first published eight months ago is not the same opportunity as one published last week, even if the feed shows both as "new".
3. The "application" that is really a CV harvest
You complete a long form, upload a CV, and the trail ends. There is no named employer, no interview process, and no reply. The page exists to collect candidate data, not to fill a role. This is the most damaging pattern for job seekers because it costs time and hands over personal data for nothing.
A real listing points at a real company, and the apply path ends at that company or its applicant tracking system.
4. Keyword-stuffed "AI" titles
As "AI" became a hiring keyword, unrelated roles started wearing it. Account managers, generic software engineers, and sales roles get an "AI" prefix to ride the search volume. The result is a feed where the word AI no longer tells you anything about the work.
The fix is to read the responsibilities, not the title. If nothing in the description involves models, data, evaluation, or systems that support them, the "AI" label is decoration.
5. Stale feeds
The quietest problem is the worst one. A board built from a single scrape can keep serving the same snapshot for months. Every card looks plausible. Nothing tells you that half the roles closed weeks ago. You spend your time on listings that were already dead when you found them.
A job board is only as good as its refresh rate, and almost none of them publish it.
Why This Keeps Happening
The incentives are stacked against the reader. Aggregators are paid to show volume, so they scrape and republish. Source boards rarely enforce expiry, so old listings stay fetchable. Scraping LinkedIn is technically fragile and against its terms, but the data is attractive enough that people keep trying.
None of these actors are necessarily malicious. They are each optimising their own metric. The problem is that nobody in the chain is optimising for the only metric that matters to you: did this listing lead to a real conversation with a real employer.
That is why we stopped treating LinkedIn as the source and built the board around feeds that are meant to be read. You can read the full approach on our About page.
What We Do Differently
AI Work Portal pulls from public job APIs and company applicant tracking systems, not from scraped LinkedIn. That single decision removes most of the patterns above, and the rest are handled by explicit rules:
- Refresh twice a week. The feed is rebuilt on a Monday and Thursday schedule, so a listing cannot silently age for months.
- Hide anything older than 45 days. If we cannot stand behind a role as current, we do not show it. An empty board is more honest than a padded one.
- Block harvest and aggregator firms. Known offenders are filtered out at ingest, and checked again when the page renders.
- Stamp remote status from the source. We record whether a role is remote, hybrid, or onsite from the source data instead of guessing from the description text.
- Credit the source. Every card shows where it came from, so you can see whether a role arrives via a company board or a public feed.
The board defaults to remote because that is what most people search for, and the off-state is shareable so a link keeps the filters you meant to send.
The result is a smaller feed. That is the point. A board that shows you fifty current roles is more useful than one that shows you five hundred, half of which are dead.
How to Spot a Bad Listing Yourself
These checks take about thirty seconds and will save you hours:
- Check the company. Real roles name a real employer with a real domain. No company, no application.
- Follow the apply link. It should end at the company careers page or its ATS, not a generic form that only collects a CV.
- Look at the posting age. Anything open for many months deserves scepticism. Ask directly when the role was opened.
- Read past the title. Confirm the responsibilities actually involve the work you want.
- Search the exact title in quotes. If it appears under several unrelated company names, you have found a repost.
- Be wary of urgency. Pressure to apply immediately, with no named employer, is a data-collection signal, not a hiring one.
None of these rules are exotic. They are just rarely applied by the boards themselves. If you want more on the remote side of this, our guide to remote AI jobs goes deeper on what to look for.
Conclusion
A noisy job feed is not an accident. It is the natural product of counting listings instead of outcomes. The patterns, reposts, ghost roles, harvest forms, keyword inflation, and staleness, all follow from that one choice.
We went the other way: fewer sources, fetched properly, refreshed on a schedule, and filtered hard. It means a smaller board, and we are fine with that. If a listing is on AI Work Portal, it is current, it names a real employer, and it points somewhere real.
👉 Start with remote AI jobs or browse all current listings.