Inside CoBlack
The label you never see
Workable’s documentation tags AI-assisted applications and lets employers filter them, while its robots.txt invites AI to read the posting. We read both.
On this page
There is a text file on apply.workable.com, the host where candidates hand their applications to employers hiring through Workable, and it tells AI crawlers they are welcome: every page may be read and fed to a model, just not used for training. The file is robots.txt, its Disallow field is empty, and it carries the line “Content-Signal: search=yes, ai-input=yes, ai-train=no”. Workable’s main site defines ai-input, citing contentsignals.org, as “inputting the content into one or more AI models (e.g. retrieval augmented generation, grounding)”. In plain terms, AI is expressly welcome to read the job posting. Anyone can check this in one fetch. We did, on September 27.
The same company maintains a second document. It is an article in its help center titled Managing AI-assisted and automated job applications, first published in September 2025 and kept current since. It is not an announcement. It is standing documentation, and it describes what happens to AI-assisted applications when they arrive on the other side of the same transaction.
What Workable tells employers
The article is direct. “Workable detects AI-assisted applications by analyzing application metadata and identifying patterns commonly associated with automated application tools.” When the detection triggers, “the candidate is tagged as AI-assisted.” If the application arrived under a machine-made address, “Workable will replace it with the candidate’s actual email from their resume, when available.”
The tag is not decoration. Employers can act on it: “You can filter your candidate list to include or exclude AI-assisted applications. The AI-assisted filter is available in the Pipeline view and the Candidates page.” The article offers more. “If you notice that a particular job board is generating this type of traffic, you can disable it on a per-job basis. Another effective strategy is to add auto-disqualifying questions relevant to your role, which quickly filter out low-quality or automated submissions.”
One more fact matters. The article describes the tag as something employers see and filter on. It says nothing about telling the candidate. The application is delivered, it appears in the pipeline, and the article does not describe any notice to the applicant that a label rode along with it.
The problem it is pointing at
Workable is not inventing a threat. Some mass-apply tools submit machine-generated relay addresses instead of the candidate’s own; the article’s example is [email protected]. One person can arrive under several different addresses, which creates duplicate profiles in an employer’s pipeline. That is a real mess, and a tracking system is entitled to clean it up.
What the documents do not say
Two things these pages do not claim, and we will not claim for them. The article describes tagging and filtering, not blocking; the tagged application still arrives in the pipeline. And neither document says how often employers use the exclusion filter, or what the label does to a real person’s odds. They describe a mechanism, not outcomes.
Set the two documents side by side anyway. One grants AI affirmative permission to read the job posting. The other tags the candidate when the application shows the patterns of an automated tool. The machine is invited in. The person whose help is noticed carries a mark no one promises to show them.
Help should not cost you your face
We believe help should not cost you your face. A person who gets help applying, from a tool, a friend, a career center, is still the same candidate with the same history. What the documentation is really describing is not help. It is industrial volume behind a mask: generated addresses, duplicate identities, the same spray sent everywhere. The candidate disappears behind the tooling.
Our approach is to keep the person visible in the first place. CoBlack builds each application from the member’s real profile and work history, tailored to the one opening it answers, never inflated past what is true. Openings come only from validated employer career pages and ATS feeds. One person, one application per opening, and nothing goes out that the member’s own record cannot back. The interview stays human.
Software will keep reading job postings, and job seekers will keep getting help. The version of that future worth building is the one where the person stays visible on both sides of the page.
Keep reading
More from Inside CoBlack →Where entry level went
Entry-level hiring in the most AI-exposed occupations: the entry-level share of postings fell from 29 percent to 10 percent between 2021 and 2026.
Four weeks in Silicon Valley
CoBlack was selected for Plug and Play's GOAL program in Silicon Valley: four weeks at the Plug and Play Tech Center in Sunnyvale, October 12 to November 6.
They automated the wrong half
No response after applying is the norm now: PeopleScout found 64 percent of candidates heard nothing from more than half the employers they applied to.
You were right
HR leaders now say AI resume screening may be hiding qualified candidates from them. What the admission changes, and where CoBlack chooses to stand.
The wasted evening has a witness
Ghost job laws are still bills in New York, Pennsylvania and New Jersey, and Texas is investigating. What the shift means for the candidate's wasted evening.
