Free plan

← All guides

Agent Policy for Job Boards and Recruiters

Published November 14, 2026 · Reviewed November 2026 · By the Ingenious AI team

The short answer

Job boards live on a two-sided flow agents now mediate from both ends: candidates send agents to search and shortlist roles, and recruiters send agents to source and screen. The job-board pattern welcomes the candidate side's search work completely, fences the application itself — an application must be the candidate's act, like a university's — and treats listing data as the commercial asset it is, refusing the bulk extraction that feeds rival boards and lead-spam operations.

Generate your policy — free · Validate an existing one

What is a job board's policy deciding?

Three things. Discoverability: whether candidate agents can search the board's listings richly enough to shortlist — the demand side of the marketplace. Application integrity: where assistance ends and submission begins, with the board's credibility to employers resting on applications being real acts by real candidates. Listing economics: who may take the board's listings away in bulk, and on what terms — the question that decides whether the board's inventory builds its own audience or everyone else's. The pattern below is the marketplace pattern retuned, with application integrity in the seat the stores pattern gives to checkout.

The candidate side: search without friction

Grant search, browse, and read broadly — candidate agents filtering by location, salary, seniority, and remote terms are doing the board's matching work at the candidate's direction. Freshness terms matter acutely: roles close, and an agent shortlisting a filled role wastes its candidate's hope, so state the cache discipline (listings quotable for 24 hours; status confirmed at the listing) and keep your own status data honest. Salary and benefits data are the board's most-quoted content; require attribution in the terms so shortlists name the board — the recruiter-facing proof that the board's inventory is where candidates are found.

The application fence

submit-form — the application act — is disallowed for agents in the standard pattern, for reasons employers will state for you if you hesitate: they are hiring a person, and the application is evidence about that person. Agent assistance upstream remains welcome and valuable — matching, tailoring under the candidate's direction, alerting — but it happens in the candidate's tools; the submission happens in the board's flow, by the candidate. Boards experimenting with agent-submitted applications for specific programmes should do it the governed way: named rules, employer consent in the arrangement, the experiment's existence public in the file. Anything less converts the board's employer trust — its scarcest asset — into an unpriced experiment.

The recruiter side and the scraping line

Recruiter tooling — sourcing agents working under a recruiter's account and agreement — belongs to the named-rule layer with the SaaS pattern's credentials logic, never to the public grant. The public line to hold is against extraction: crawl-full-site disallowed, listing resale refused in the terms, catch-all slow and conditional. Rival boards and CV-database builders announce themselves by needing the whole inventory rather than a search; the policy's structure forces that need into the open, where it becomes a licensing conversation or a refusal on the record. Boards that license listings to aggregators should mirror those licences as named rules — syndication is a product, and the file is its price list's cover page.

Boards should watch one signal above the rest in the policy's first year: employer reaction. Candidates will adopt agent assistance regardless of any board's posture — the assistants are theirs, not the board's to permit. What the board controls is whether that assistance arrives through channels the board shapes and can vouch for: search that respects freshness, shortlists that attribute, applications that remain the candidate's own act. Employers who see agent-assisted applications arriving better matched, with the board's integrity fences visibly intact, become the board's advocates in the next procurement round. The policy is how a board makes that outcome a design rather than an accident — and in a market where every board faces the same agent wave, the designed outcome is the differentiator.

Worked example: a tech job board's candidate-first file

Fictional example. DevRoles, a fictional technology job board carrying 26,000 live roles, publishes 7 rules. Candidate side: catch-all plus 3 named agents granting search and read at 40 requests per minute, 24-hour listing cache, attribution required. Application: submit-form disallowed in every rule, with the notice stating applications are made by candidates in the board's flow. Recruiter side: 2 named integration rules under its recruiter programme. Licensing: 1 aggregator rule mirroring a paid syndication deal at 100 per minute with download permitted. First-year results: agent-attributed applications (candidates arriving with agent-built shortlists) became the board's fastest-growing quality segment per employer feedback; 2 scraping operations were identified against the published terms and moved — one to a licence, one to a refusal that stayed refused.

The job board checklist

  1. Open search completely. Candidate agents are your matching engine's newest front end — verbs generous, freshness stated.
  2. Fence the application act. submit-form refused for agents; the candidate applies, in your flow, as a person.
  3. Attribute the shortlist. Salary and role quotes carry the board's name — your recruiter-facing evidence of value.
  4. Refuse extraction, license syndication. Crawl disallowed publicly; aggregators as named, priced rules.
  5. Keep recruiter agents contractual. Named rules under the recruiter programme; credentials never in the file.

Keep reading

Agent Policy vs Bot Blocking: Declaration Meets Enforcement · Who Should Own the Agent Policy on a Team? · How to Tell AI Agents What They May Do · Spec samples · Conformance · Pricing · All guides

Frequently asked questions

Why refuse agent-submitted applications if candidates want the convenience?

Because the application's evidentiary value to employers collapses if its authorship is automated and undisclosed — and employer trust is the board's business. The pattern preserves the convenience that matters (search, matching, shortlisting, alerting) and reserves the act that certifies the candidate. Boards that judge the trade differently should run the experiment openly, with named rules and employer consent.

Can agents alert candidates to new roles under the policy?

Alerting is reading plus the agent's own notification machinery — the board's side of it is simply being searchable and current. Boards that want to support it well keep listing status ruthlessly accurate and state a sane cache window, so alerts fire on real openings. Some boards add named rules for alert services whose quality they have vetted; the catch-all search grant already covers the rest.

How do we stop rival boards copying our listings?

Declare the boundary in the standard way — crawl-full-site disallowed, resale refused, catch-all conditional and slow — and license the legitimate aggregators as named rules so licensed copying is visibly distinct. A rival that proceeds anyway is handled in the enforcement layers, with your published terms as the opening document, exactly as in the marketplace pattern.

Should salary data be in quotes?

If you publish it, agents will quote it — the choice is whether they quote it with your board's name attached and your freshness terms around it. Attribution and a cache window turn salary transparency into the board's marketing. Boards in jurisdictions with pay-transparency rules will find the policy a natural place to state how their published ranges should be carried.

What pace does candidate search need?

Search fan-out is moderate: filters do the heavy lifting server-side, so an agent's session is dozens of requests, not thousands. Thirty to fifty per minute serves named candidate agents comfortably; the number to watch is not pace but status accuracy — a fast agent quoting closed roles is worse than a slow one quoting live ones.

Generate a policy — free · Validate a policy · What an agent access policy is

← Back