Who Does AI Think Is the Employer? Attribution in Placement, Dispatching, and ATS

Where a job is handled through a placement agency, a staffing company, or a job board, a single job posting can involve several parties. There is the company that opens the position, the operator that intermediates it, and the medium that publishes it. When a job seeker asks a generative AI what kind of company a given job belongs to, the AI points somewhere along that chain. Which point it picks is not self-evident.

In worker dispatching in particular, the legal employer is the dispatching business operator, not the client company where the work is performed. Within the same job posting, the party that holds the employment relationship and the party that holds the place of work sit in separate fields. This structure existed long before AI. What AI introduced is a new path: when that structure is summarized into natural language, the two parties can be folded into one.

1. The parties in the chain through which job information travels, and how Google, Indeed, and Google Cloud Talent Solution handle employer attribution
2. Why the employer and the place of work diverge structurally in worker dispatching
3. Five paths along which attribution can break down
4. What the Employment Security Act requires of intermediary operators
5. How to check the way the jobs your company handles are attributed in AI space

Practitioners at placement agencies, staffing companies, job boards, and ATS/HR-tech vendors — that is, employment placement business operators and recruitment information providers under the Employment Security Act, and dispatching business operators under the Worker Dispatching Act. The perspective of the company that opens its own positions is covered separately in Does JobPosting schema help AI?.

Let us state first what this article does not claim. Two things could be confirmed. First, that Japan's Employment Security Act and Worker Dispatching Act distinguish the parties involved in circulating job information and impose different duties on each. Second, that Google, Indeed, and Google Cloud Talent Solution have built into their specifications the premise that the party publishing a job and the party employing the worker may be different. Both can be verified against primary sources.

One thing could not. A public benchmark that measures the rate at which generative AI, answering in natural language, attributes a job to the wrong employer — within the scope of this review, we could not confirm one.

Adjacent research does exist. There is work that learns semantic representations of company names from a corpus of job ads in order to solve company entity matching (CompanyName2Vec, IEEE DSAA 2022), and work on entity linking that maps skill descriptions in job text to occupational classifications (Findings of EACL 2024). However, the former addresses entity matching, and the latter's target is skills, not companies. AI visibility surveys on the practitioner side likewise measure the frequency with which a brand is mentioned or cited; they do not measure whose job an AI said a posting was.

This article therefore presents no figures on frequency of occurrence. What it presents are the structurally possible paths that can be deduced from the legal framework and platform specifications, together with a method for checking them for yourself.

This article is not an interpretation of the law. For individual cases, please consult a certified social insurance and labor consultant or an attorney.

1-1 "Intermediation" is not one thing

Before a single job reaches a job seeker, the following roles are typically involved. Even within "intermediation," employment placement and worker dispatching are opposites as to whether an employment relationship exists.

Within the single term "the staffing and recruiting business," a role that holds no employment relationship and a role that holds one sit side by side. That is the starting point of this article.

1-2 The Employment Security Act defines this distinction in statute

Article 4, paragraph (1) of the Employment Security Act defines employment placement as accepting job offers and job applications and brokering the formation of an employment relationship between the job offerer and the job seeker. The party doing the brokering is not a party to the employment relationship; even where a placement agency handles a job, the employer is the hiring company.

Paragraph (6) of the same Article, meanwhile, enumerates "provision of recruitment information, etc." as four types of act, and the Ministry of Health, Labour and Welfare organizes these as business categories 1 through 4. The axis of division is whether the information was provided on request or collected by the operator itself. The 2022 amendment made explicit that services that crawl and collect job information for provision, and services that republish content posted on other job media, also fall within the business of providing recruitment information, etc.

In other words, Japan's legal framework decomposes what tends to be lumped together as "handling job information" along the axis of whether it was requested or self-collected. That decomposition existed long before AI appeared.

1-3 Why "whose job is this" stops being self-evident

A job seeker wants to know which company a job belongs to. Seen from the information side, however, that question has at least three answers. Who is the counterparty to the employment contract? Where is the work actually performed? Where did this information come from?

When a person reads, context fills the gaps. If a job board's logo sits at the top and the placement agency's name sits at the bottom, the reader processes them as separate things. But once the information is reduced to a machine-readable form, summarized, and regenerated as natural language, that visual distinction is lost. What remains is only which field held which value. That is why the definitions of the fields matter.

2-1 hiringOrganization means the company that actually employs

Google's documentation on job posting structured data defines hiringOrganization as the organization offering the position, and states that it must be the name of the company and not the name of the individual location that is hiring. The example the documentation gives is "Starbucks, Inc" rather than "Starbucks on Main Street."

The same documentation defines jobLocation as the physical location of the business where the employee reports to work (an office or a work site, for instance), and not the place where the job was posted.

Google therefore treats the employing party, the place of work, and the place of posting as separate things. That the specification needed this distinction means the possibility that these do not coincide is assumed.

2-2 confidential — an official format for anonymous postings

Google also provides for the case where an organization hires anonymously. The examples given are a staffing services provider posting on behalf of an anonymous employer, and an employer posting directly on a provider's platform; in that case confidential is specified as the value of hiringOrganization.name. That a staffing services provider heads the list of examples reflects that the anonymous-posting mechanism was designed on the premise that intermediation exists.

2-3 Indeed — the employer must be created before the job

Indeed's design goes further. As a prerequisite of the Employer Data API, Indeed requires that an employer be created before any job associated with that employer is created. employerName and employerType are treated as global attributes that take the same value in every country.

The employer is designed not as one attribute of the job but as an independent object that precedes the job. The design philosophy shows in that ordering.

2-4 Indeed's requirements for Japan call for intermediation to be made explicit

More concrete still are the requirements Indeed publishes for its partners in Japan under "Placement companies." The documentation defines a placement company as one whose customer is the employer and which acts as an intermediary between the employer and the candidate, and states that a posting may be rejected if the required fields are not sent. What is required is as follows.

The sample job description in the documentation includes a sentence, in Japanese, stating that the posting is a placement handled by a named employment placement business operator.

What is happening here is that a single job posting simultaneously holds two company names in two fields that carry different meanings. Moreover, the intermediation relationship itself is declared in machine-readable form as the boolean isPlacement, and Indeed additionally requires it to be stated in the natural-language description.

Indeed also provides a separate employment-type code for dispatched workers in Japan (8YWGX, Third-party contract) and requires the attribute 5T857 to be used alongside it for indefinite-term dispatching. Contracting work is T65DZ and shokutaku (contract) employees carry the attribute 9HCQR: a different value is assigned to each form of contract. The judgment that forms of contract need to be distinguished is reflected in the code system.

2-5 Google Cloud Talent Solution — roles themselves as units of design

The basics documentation for Google Cloud Talent Solution classifies customer sites into four basic use cases: job boards (returning job listings to job seekers), career site providers (providing career site services to client companies), staffing and placement providers (supplying short-term or long-term talent to clients), and applicant tracking systems (ATS) (tracking applicants through the hiring process).

The service also has an intermediate layer, the tenant, which owns the jobs and companies objects. The uses the documentation gives are a job site provider building a job site for an organization with multiple subsidiaries, and a placement agency building an applicant tracking system for multiple companies. The release notes also record the addition of an event type for the case where a placement agency submits an application on behalf of a candidate.

2-6 What the existence of these specifications shows

The specifications reviewed so far share one point.

> That the publishing party and the employing party can differ is explicitly assumed by design.

The fact that these specifications exist shows that the need to distinguish is real. It does not show that writing in accordance with them will make generative AI recognize the employer correctly. What a specification guarantees extends only to storing information in the sense each platform has defined. How that is treated at the stage where it is summarized into natural language remains a separate question.

3-1 The legal employer is the dispatching business operator

Article 2, item (i) of the Worker Dispatching Act — formally, the Act for Securing the Proper Operation of Worker Dispatching Business and Protecting Dispatched Workers — defines worker dispatching as follows.

> Having a worker whom one employs engage in work for another person, under that employment relationship and under the direction and instruction of that other person; this does not include arrangements under which it is promised that the worker will be employed by that other person.

As the provision states twice — a worker "whom one employs," and "under that employment relationship" — the employment relationship is between the dispatching business operator and the dispatched worker. The client company gives direction and instruction; it is not a party to the employment relationship.

The Ministry of Health, Labour and Welfare explains it the same way. It describes the worker dispatching business as a dispatching business operator having workers it employs engage in work for a client company, under that client company's direction and instruction, as a business. And its guide on distinguishing worker dispatching from contracting work writes the responsible parties separately as "the employer (the dispatching business operator or the contracting business)," "the client company," and "the ordering party." That an administrative document uses "employer" for the dispatching business operator is an important point of confirmation in practice.

Whether a given contractual arrangement constitutes worker dispatching or contracting work is judged on the substance of the arrangement rather than its form. For judgments on individual cases, please consult a certified social insurance and labor consultant or an attorney.

3-2 Within the same posting, two parties sit in different fields

Overlaying the employment relationship under the Worker Dispatching Act with the property definitions from Google reviewed in §2, this article organizes an ordinary dispatching job posting as follows.

★This mapping is this article's own. Google has not stated it specifically for worker dispatching in Japan. What Google states extends only to the definition of what each property refers to.

Written as structured data, this takes the following form.

{
  "@context": "https://schema.org/",
  "@type": "JobPosting",
  "title": "Customer Support",
  "datePosted": "2026-08-10",
  "validThrough": "2026-10-10T00:00",
  "hiringOrganization": {
    "@type": "Organization",
    "name": "(Name of the dispatching business operator holding the employment relationship)"
  },
  "jobLocation": {
    "@type": "Place",
    "address": {
      "@type": "PostalAddress",
      "addressLocality": "(Location of the business where the work is performed)",
      "addressCountry": "JP"
    }
  }
}

A single posting stores side by side the name of the company that holds the employment relationship and the address of a company with which that company has neither a capital nor an employment relationship. This is not an error; it is the result of following the definitions in the specification faithfully.

This example omits employmentType because Google's employmentType has no value that directly expresses "worker dispatching" as it exists under Japanese law. TEMPORARY means only that the position is of limited duration; it is not a code for the legal relationship of worker dispatching (indefinite-term dispatching also exists). This is where the difference from Indeed's code system in §2-4 shows.

3-3 The gap between everyday phrasing and the legal employment relationship

The problem arises when this structure is rendered into natural language.

In everyday Japanese, it is an entirely ordinary turn of phrase for someone working through dispatching to say "I work at Company A." The particle points to the place of work, not to the employer. On the face of the sentence, however, it is indistinguishable from a permanent employee saying the same thing.

A process that summarizes a job posting cannot obtain that distinction from anything other than the definitions of the fields. Once the two are folded into natural language, the basis for the distinction is left to context.

What has been described so far is the structure; as for the proportion of cases in which the distinction is actually lost, within the scope of this review, we could not confirm it. That a path exists and how far that path is traveled are separate questions.

3-4 What happens in employment placement

In employment placement, the employment relationship is formed between the hiring company and the job seeker (Article 4, paragraph (1) of the Employment Security Act). The placement agency is not the employer. Accordingly, what belongs in hiringOrganization is the hiring company.

In practice, however, confidential job postings that withhold the hiring company's name are widely handled. What is provided for this case is confidential from §2-2, together with isPlacement and sourceType from §2-4. Neither is a mechanism for cases where "the employer is unknown"; both are mechanisms for making it explicit that the employer is being withheld.

Note that in dispatching with a view to direct employment (Article 2, item (iv) of the Worker Dispatching Act), the employer during the dispatching period is the dispatching business operator, but once employment placement is completed and the client company employs the worker directly, the party to which the job is attributed itself changes. This article's account concerns the dispatching period.

What follows organizes the paths that can be deduced from the legal framework and platform specifications. For none of them do we present figures on frequency of occurrence.

4-1 Identical names, similar names, and group companies

Company names are not unique. Group companies, former corporate names, branch names, trade names, and unrelated companies bearing the same name coexist.

This problem is not specific to AI. The leaflet the Ministry of Health, Labour and Welfare publishes for recruitment information providers states the following about the recruiter's name, as a point of caution regarding misleading representation.

> Do not present information in ways that confuse the hiring company with group companies.

What the leaflet gives as an instance of this is prominently featuring information about a group company with a strong track record. The inappropriate example given is displaying, for a job at Company B — a group company of Company A — a statement that Company A needs people with advanced IT engineering skills. Errors around names within a group are a point on which the administration has issued explicit cautions, before AI ever reads them.

On the technical side, the ambiguity of company names is a known problem. One study learned semantic representations of company names from a corpus of job ads and solved company entity matching with an average success rate of 89.3% (CompanyName2Vec, IEEE DSAA 2022). Even for a method trained specifically on company names in the job domain, that figure means roughly one in ten remains unsolved. Research using large language models has also reported error types in which a mention cannot be resolved for lack of context, and in which a mention is misinterpreted and assigned to a different class (Pons et al., ISWC 2024).

Both concern accuracy on the task of entity matching, not a measurement of whose job a generative AI says a posting is when answering in natural language. What they share, however, is the structure that a match on company name alone does not establish identity.

4-2 Substitution of parties in dispatching

This is the structure described in §3. The dispatching business operator goes in hiringOrganization and the client company in jobLocation, and a path remains along which the two can be folded together at the stage of summarizing into natural language.

4-3 Anonymous postings lose their subject

In a posting that specifies confidential, the employer's name has been deliberately withheld. What can remain as the subject when the posting is summarized is the name of the publishing medium, the name of the placement agency, or nothing at all. There are cases where the decision to withhold is lawful and necessary, but it needs to be recognized that the party withholding cannot decide what becomes the subject as a result.

4-4 Multiple postings of the same job

It is not unusual for the same job to exist simultaneously on the hiring company's career site, on the placement agency's site, and on several job media. The problem is not the existence itself but the case in which who is written as the employer is not aligned from page to page. Google asks that a canonical URL be specified where the same job exists at multiple URLs, and asks third-party job sites to provide job titles as received, without rewriting them.

4-5 Persistence after the posting closes

Google's job posting content policies do not permit expired job postings, and failing to address them can become the subject of a manual action. The methods given for closing a posting are setting validThrough to a date and time in the past, removing the page so that it returns 404 or 410, and removing the JobPosting structured data.

Even after a page has been taken down, however, third-party articles that cited it, caches, and republished copies can remain. "The job has been closed" and "descriptions of that job have left circulation" are not the same thing.

4-6 The paths can be explained; the frequency cannot be stated

All five can be traced from the legal framework and platform specifications. As for public data showing which paths are actually traveled and to what extent, within the scope of this review, we could not confirm any. With no industry-wide figures available, there is no alternative to checking the jobs your own company handles for yourself. §8 covers how.

5-1 Article 5-4 of the Employment Security Act — the three paragraphs address different parties

The amended Employment Security Act, which took effect on October 1, 2022, added Article 5-4, on accurate representation of information concerning job offerings and the like. The three paragraphs each address different parties.

What matters is that paragraphs (2) and (3) apply according to the legal role being performed in the situation, not the industry label of the company. Recruitment information providers such as job boards, and employment placement business operators, are addressed by paragraph (3); meanwhile, where a dispatching business operator recruits its own dispatched workers, it may fall under "a person conducting labor recruitment" in paragraph (2). The same company can be addressed by both, depending on the function. For ATS and HR-tech operators as well, whether they fall within the business of providing recruitment information, etc. varies with the service functions they provide.

The other difference lies in the nature of the duty. Whereas paragraph (2) requires a state of the information itself, paragraph (3) requires that measures be taken. The Ministry of Health, Labour and Welfare's published Q&A on the amended Employment Security Act organizes it the same way: providing information that is not accurate or up to date does not immediately constitute a violation of paragraph (3), but failing to take the measures provided by ordinance does.

5-2 The specific measures set out in the Enforcement Ordinance

Article 4-3, paragraph (4) of the Ordinance for Enforcement of the Employment Security Act gives concrete form to the measures under paragraph (3). The Ministry of Health, Labour and Welfare's leaflet organizes them by business category.

Measures every recruitment information provider needs to take

• Respond without delay where asked to stop providing, or to correct, job information or job-seeker information
• Where it is confirmed that job information or job-seeker information is not accurate or up to date, either check without delay with the party that requested the provision whether there is a correction, or stop providing the information

Measures of which operators that provide information on request (categories 1 and 3) need to take one

• Ask the party requesting provision of job information to give prompt notice where recruitment has ended or the content of the job information changes (by email or in writing, preferably)
• Make clear the point in time to which the job information or job-seeker information refers

Measures of which operators that provide self-collected information (categories 2 and 4) need to take one

• Collect and update job information and job-seeker information regularly, and make the frequency clear
• Make clear the point in time to which the job information or job-seeker information refers

Measures of which employment placement business operators need to take one

• Check regularly with job offerers and job seekers whether the job information or job-seeker information is up to date
• Make clear the point in time to which the job information or job-seeker information refers

Employment placement business operators are likewise required, in common with the others, to respond without delay to requests for correction or for provision to stop, and to check for corrections or stop providing where they confirm that information is inaccurate or outdated. The leaflet adds, for every category, that it is desirable to take all of these measures so far as possible. What is worth noting is that operators that crawl and collect job information are asked to make their update frequency clear. The design discloses metadata about freshness rather than freshness itself.

5-3 "Posting under another company's name" is listed as an example of false representation

On the false representations prohibited by paragraph (1), the Ministry of Health, Labour and Welfare's leaflet lists examples that may fall within it. The first is this.

> Posting a job under a company name other than the one actually recruiting.

This is precisely the attribution problem this article addresses. Posting a job under a company name other than that of the employer is not a matter of how well it is worded; it is conduct that may constitute false representation under the Employment Security Act, and the administration lists it explicitly.

The same leaflet gives the confusion of group companies from §4-1 as an example of misleading representation, and also treats as inappropriate the conflation of freelance (commissioned) recruitment with recruitment under an employment contract. Two issues — whether an employment relationship exists, and who the counterparty to it is — are both subject to the representation rules. That the sample entries for job information include an employment-type field for the case of employing someone as a dispatched worker likewise reflects that stating whether a job is dispatching was assumed as a component.

5-4 The certification system, and disclosure regarding display order

A note on the surrounding framework as well.

The certification system for excellent recruitment information providers is a Ministry of Health, Labour and Welfare commissioned program that certifies recruitment information providers meeting set standards on legal compliance, accurate representation, the handling of personal information and the like, disclosure, screening, and the handling of complaints. It began in fiscal 2022. Certified operators are published, and a list as of April 1, 2026 can be checked on the program's official site.

The other point that cannot be overlooked in relation to AI is disclosure of business information. The Employment Security Act asks recruitment information providers, as a duty of effort, to disclose business information via the internet and other means. The items subject to disclosure include the following.

Matters concerning accurate representation of information on job offerings and the like; matters concerning the protection of personal information; matters concerning the handling of complaints; and matters taken into account in determining the display order of search results for job information and job-seeker information — four items in all.

On the last item, the Ministry of Health, Labour and Welfare adds that where algorithms, AI, and the like are used to determine display order in accordance with the items above, disclosure of the details of those algorithms or AI is not required. Disclosure of the matters taken into account is required; the internals of the algorithm are not — a division that connects to the approach to measurement discussed below.

6-1 Criteria for deciding who goes in hiringOrganization

The judgment is clearest when it starts from the employment relationship.

1. Determine who enters into the employment contract. For worker dispatching, the dispatching business operator; for employment placement, the hiring company. Where the judgment is unclear, confirm with a certified social insurance and labor consultant or an attorney.
2. Use that party's formal legal name. Do not substitute a location name, a brand name, or the name of the group parent. Use confidential only where the employer needs to be withheld (§6-2).
3. When sending to Indeed, put the client company's name and your own company's name in different fields (§2-4).

6-2 When to use confidential, and when not to

It may be used where the employer wishes to remain anonymous and is withheld on the basis of that intent (§2-2). It should not be used in cases such as the following.

• Where your company does not know who the employer is and wants to avoid determining it
• Where the information obtained from the republishing source does not include the employer and you want to save the effort of checking
• Where you withhold the employer deliberately in order to make your own name more prominent

The first and second may raise an issue under the law, in the sense that the measures required by paragraph (3) have not been taken. The third falls in territory that may touch the representation rules in §5-3.

6-3 Sorting out multiple postings

As in §4-4, three things need sorting out.

• Specify the canonical URL (Google asks for this)
• Align the employer name. Check that the name in hiringOrganization is aligned across your own site, partner media, and ATS-mediated routes
• Avoid rewriting titles (Google asks that they be provided as received)

6-4 Operating the close of a posting

Design the closing process in three stages.

1. Closure in the specification. Reliably carry out one of the three methods in §4-5.
2. A communication route with the requesting party. As a measure under paragraph (3), ask the party requesting provision to notify you of the end of recruitment or changes in content (categories 1 and 3), or make clear the frequency of regular collection and updating (categories 2 and 4).
3. Observation after closure. After a set period, check whether a job you closed is still being described in AI answers (§8).

The third is not a requirement of the law, but it is a way of confirming whether the closing process reached beyond your own perimeter.

6-5 What not to do

The last row is based on the fact that Google's job posting content policies list, among violations, posting a job on behalf of an organization or company without authorization, postings or content that represent the employer in a way that is not accurate, realistic, and truthful, and providing false location data that does not match the actual place of work. The content overlaps with the false representations prohibited by Japan's Employment Security Act.

This section sets out the issues this article could not address. Each means "not yet known," not "there is no problem."

• The rate at which generative AI states the wrong employer for a job. Within the scope of this review, we could not confirm a public benchmark measuring this (for adjacent research, see §0).
• Which generative AI favors which party. Within the scope of this review, we could not confirm public data comparing which of hiringOrganization and jobLocation tends to be chosen as the subject where both are given.
• The identifying effect of sameAs and identifier. These are properties that can help distinguish companies bearing the same name, but within the scope of this review, we could not confirm public data showing that they improve identification accuracy in generative AI answers.
• What AI takes as the subject where confidential is specified. Within the scope of this review, we could not confirm data examining this systematically.
• Industry-association guidance on AI attribution. The Japan Staffing Services Association published, in May 2026, a survey on generative AI covering member companies and certified excellent dispatching business operators, 141 companies; and Keidanren (Japan Business Federation) published, in April 2026, a report on the use of AI in HR departments. Both, however, address issues on the side of using AI in operations, and within the scope of this review, we could not confirm that they address the issue of how AI describes one's own company or client companies.

The last item expresses where this article sits. Interest on the industry side is directed at the use of AI; the accuracy of attribution in AI space has not yet been framed as a shared issue.

8-1 Why you have to measure it yourself

As for industry-wide figures, within the scope of this review, we could not confirm any (§7). At the same time, taking measures to keep information accurate and up to date is a duty under the law (§5).

> Rather than waiting for an industry average, observe for yourself how the jobs your company handles are described in AI space.

As set out in §5-4, the Employment Security Act does not require disclosure of the internals of the display-order algorithm; it requires disclosure of the matters taken into account. Working with conditions and results rather than internal mechanisms is an approach that applies equally to measuring perception in AI space.

8-2 What to observe — putting attribution accuracy at the center

★What follows is not a legal obligation required by Article 5-4, paragraph (3) of the Employment Security Act. What that paragraph requires are the measures in §5-2, not the monitoring of generative AI answers. From here on is an optional external monitoring approach that this article proposes.

The common metrics for AI visibility are exposure: whether a brand was mentioned, whether it was cited. This article proposes observing not only exposure but attribution accuracy in particular. Five items work well as the axes of observation.

As the third column shows, the observation items correspond one to one with the structures organized in this article. The order matters: the observation items are derived from the structure. It is because the paths along which attribution can break down have been identified that the points to observe are determined.

8-3 Aligning measurement conditions

Generative AI answers vary from run to run even for the same question. A single check cannot distinguish whether an observed event is a tendency or a coincidence. The conditions to control include at least the following.

• The wording of the prompt. "Tell me about Company A's job openings" and "What jobs is Company A recruiting for" imply different answers.
• Session state. Observation needs to take place under conditions that carry in no conversation history and no personalization.
• Number of repetitions. A single answer is one sample. Only repetition under identical conditions makes a tendency readable.
• Date of observation. Both the AI models and the information they reference are updated. Observation results without a date cannot be compared.

8-4 Vaipm's measurement design

Vaipm measures AI-space perception through a total of 25 stateless queries across multiple AI engines.

Two points are worth stating. The first concerns the number of queries. This is not an optimum derived from research; it is Vaipm's operational design. It is the result of a judgment about where to strike the balance between practical repeatability and precision, given the variability of answers. The second is statelessness: observing without carrying in conversation history or personalization is the core of controlling conditions, and it addresses the session-state issue in §8-3.

What is observed is not exposure alone. Alongside citations and mentions, it includes how the company was treated within the answer, which party the job was attributed to, and whether errors or outdated information were present. In the context of this article, the five observation items in §8-2 become the measurement items directly.

The thinking behind this is covered in What is AI Perception Management (AIPM)?, and the overall picture of the measurement field in What is AIO (AI search optimization)? and What is LLMO?.

8-5 What the results are used for

Observation is not the goal. The output connects to three things.

1. Correcting representations. Where a mismatch in the employer name, a lingering expired posting, or confusion with a group company is found, go back to the operations in §6 and fix it. This also amounts to verifying for yourself the effectiveness of the measures in §5.
2. Maintaining the pages used as the basis. Knowing which pages the AI presented as its basis determines which pages need to be kept accurate. The supply structure of third-party information sources is covered in What AI cites about your company.
3. Responding to errors. Responding to a wrong attribution does not mean a removal request alone. Save a reproducible prompt and the date and time, identify where it contradicts information you have already published, correct the gaps and outdated pages on your side, and check again after a set period. For details, see Responding to AI misinformation.

How candidates use generative AI to research companies is collected in the hub article How candidates research you with AI. This article carves out of it the issues facing the intermediating side.

The circulation of job information lines up the company that opens the position, the operator that intermediates it, and the medium that publishes it. Who the employer is is not self-evident, and in worker dispatching the dispatching business operator, which is the legal employer, and the client company, where the work is performed, sit in different fields of the same posting. The major platforms have built this ambiguity into their specifications as an explicit design assumption, and Japan's Employment Security Act, at Article 5-4, paragraph (3), requires those who intermediate and those who publish to take measures to keep information accurate and up to date.

Attribution accuracy is not a new problem born in the age of AI. What was already a requirement of the law is being tested along a new path: summarization by AI. As for public data showing how far generative AI actually gets it wrong, within the scope of this review, we could not confirm any. That is precisely why the practice of observing, under aligned conditions, the jobs your company handles is required.

Q1. For a job handled by a placement agency, whose job does AI say it is?

Within the scope of this review, we could not confirm a public benchmark that answers this question. What can be confirmed extends only to this: several parties line up in the chain through which job information travels, and which party is treated as the "employer" is set by explicit rules that differ by platform. Accordingly, there is at present no method of higher confidence than checking, for the jobs your own company handles, what actually happens. The procedure for checking is organized in §8.

Q2. For a dispatching job, which party is the legal employer?

Article 2, item (i) of the Worker Dispatching Act defines worker dispatching as having a worker whom one employs engage in work for another person, under that employment relationship and under the direction and instruction of that other person. The employment relationship is between the dispatching business operator and the dispatched worker, and the client company is in the position of giving direction and instruction. The Ministry of Health, Labour and Welfare likewise writes the responsible parties separately as the employer (the dispatching business operator or the contracting business), the client company, and the ordering party. Because whether an individual contractual arrangement falls within a category is judged on its substance, please consult a specialist.

Q3. Should hiringOrganization hold the dispatching business operator or the client company?

Google defines hiringOrganization as the organization offering the position, and as the name of that company rather than the name of the individual location that is hiring; and it defines jobLocation as the physical location of the business where the employee actually reports to work, rather than the place where the job was posted. Overlaying the employment relationship under the Worker Dispatching Act with those definitions, this article treats the former as the dispatching business operator and the latter as the business where work is actually performed. Google has not, however, stated this mapping specifically for worker dispatching in Japan, and because applying it involves a legal judgment, please confirm with a specialist.

Q4. When is confidential used?

Google gives, as examples of an organization hiring anonymously, a staffing services provider posting on behalf of an anonymous employer and an employer posting directly on a provider's platform, and provides for confidential to be specified as the value of hiringOrganization.name. This is a format for cases where there is an intent to withhold the employer. It is not an escape route for cases where it is unknown who the employer is, or where one wants to save the effort of checking.

Q5. What happens legally if a job is posted under a company name other than the employer's?

The leaflet the Ministry of Health, Labour and Welfare publishes for recruitment information providers lists, first among examples that may constitute false representation, posting a job under a company name other than the one actually recruiting. Article 5-4, paragraph (1) of the Employment Security Act prohibits false representations and misleading representations, and the parties addressed include employment placement business operators and persons engaging in the business of providing recruitment information, etc. An error in attribution is not a problem of expression; it is an issue under the law.

Q6. Is it a problem to describe a group company's job under the parent company's name?

The same Ministry of Health, Labour and Welfare leaflet states, as a point of caution on misleading representation regarding the recruiter's name field, that information must not be presented in ways that confuse the hiring company with group companies; and it treats as inappropriate the example of displaying, for a job at Company B, a group company of Company A, that Company A needs such people. Using names distinctly is, before AI ever reads them, first an issue under the law.

Q7. What is the difference between paragraphs (2) and (3) of Article 5-4 of the Employment Security Act?

The parties addressed and the nature of the duty differ. Paragraph (2) requires persons conducting labor recruitment and commissioned recruiters to keep recruitment information accurate and up to date. Paragraph (3) requires Public Employment Security Offices, specified local governments and employment placement business operators, persons engaging in the business of providing recruitment information, etc., and labor supply business operators to take measures to keep it accurate and up to date. Application is divided not by industry label but by the legal role being performed in the situation. Where a dispatching business operator recruits its own dispatched workers, it may fall under paragraph (2).

Q8. What specifically are the "measures" required by paragraph (3)?

According to the Ministry of Health, Labour and Welfare's leaflet, what is common to every recruitment information provider is responding without delay where asked to stop provision or to correct, and, where it is confirmed that information is not accurate or up to date, checking without delay whether there is a correction or stopping provision. In addition, operators that provide information on request are required either to ask for notice of the end of recruitment or of changes in content, or to make clear the point in time to which the information refers; and operators that collect information themselves are required either to make their update frequency clear or to make clear the point in time to which the information refers.

Q9. If JobPosting structured data is written correctly on a job posting, will AI recognize the employer correctly?

Within the scope of this review, we could not confirm grounds for saying so. Whether structured data works on display and citation in AI search is covered separately in "Does JobPosting schema help AI?". What this article states is that, given that specifications such as hiringOrganization make explicit which party should be written, writing in accordance with those definitions is itself a requirement of accuracy, and holds independently of any effect on AI.

Q10. What is the problem if a closed job posting persists?

It becomes a problem at two levels. First, Google's job posting content policies do not permit expired job postings, and failing to address them can become the subject of a manual action. Second, the measures required by Article 5-4, paragraph (3) of the Employment Security Act and the Enforcement Ordinance consist precisely in keeping information up to date. In addition, where the page or a cache remains after the posting has closed, a path remains along which AI refers to it and continues to state the old attribution.

Q11. Should the same job existing at multiple URLs be avoided?

It is not unusual for the same job to exist simultaneously on the hiring company's career site, on the placement agency's site, and on several job media. The problem is not the existence itself but that who is written as the employer is not aligned from page to page. Google asks that a canonical URL be specified where the same job exists at multiple URLs, and asks third-party job sites to provide job titles as received.

Q12. How should we measure how AI describes the jobs our company handles?

We recommend putting attribution accuracy, rather than exposure, at the center of the observation items. Specifically, five points: whose job it was said to be; whether the employer and the place of work are being mixed together; whether it is still being described after the posting closed; whether it is being mistaken for a company bearing the same name or for a group company; and which page was given as the basis. Because generative AI answers vary from run to run, what is needed is not a single check but repeated observation under aligned conditions. For details, see §8.

Sources are classified into three tiers. Tier 1 is primary material from statutes and government agencies, Tier 2 is official documentation from the platform providers, and Tier 3 is industry associations, academic work, and other sources. All were checked on August 10, 2026.

Tier 1｜Statutes and government agencies (primary)

1. Employment Security Act (Act No. 141 of 1947), Article 4 and Article 5-4 — Japanese Law Translation Database System (Ministry of Justice)

https://www.japaneselawtranslation.go.jp/ja/laws/view/4486

2. Act for Securing the Proper Operation of Worker Dispatching Business and Protecting Dispatched Workers (Act No. 88 of 1985), Article 2 — Japanese Law Translation Database System (Ministry of Justice)

https://www.japaneselawtranslation.go.jp/ja/laws/view/4664

3. Ministry of Health, Labour and Welfare, leaflet for recruitment information providers on the changes to the operating rules for the business of providing recruitment information, etc. (Employment Security Act as amended, effective October 1, 2022)

https://www.mhlw.go.jp/content/11600000/000983825.pdf

4. Ministry of Health, Labour and Welfare, leaflet for employment placement business operators on the changes to the operating rules for the employment placement business (Employment Security Act as amended, effective October 1, 2022)

https://www.mhlw.go.jp/content/001250188.pdf

5. Ministry of Health, Labour and Welfare, Q&A on the Employment Security Act as amended (2022) (including the description relating to Article 4-3, paragraph (4) of the Ordinance for Enforcement of the Employment Security Act)

https://www.mhlw.go.jp/content/001250191.pdf

6. Ministry of Health, Labour and Welfare, "Worker Dispatching Business"

https://www.mhlw.go.jp/stf/seisakunitsuite/bunya/roudoushahakennjigyou.html

7. Ministry of Health, Labour and Welfare, on the "Guide to Conducting Worker Dispatching and Contracting Work Properly"

https://www.mhlw.go.jp/stf/seisakunitsuite/bunya/0000077386_00020.html

8. Certification system for excellent recruitment information providers (Ministry of Health, Labour and Welfare commissioned program), official site

https://yuryonintei.com/

Tier 2｜Official platform documentation (primary)

9. Google Search Central, "Job posting (JobPosting) structured data for Job Search" (hiringOrganization／jobLocation／confidential／job posting content policies／treatment of expired postings. Last updated December 18, 2025)

https://developers.google.com/search/docs/appearance/structured-data/job-posting

10. Indeed Partner Docs, "Placement companies (Japan)" (requirements for companyName／sourceName／sourceType／isPlacement. Last updated July 2, 2026)

https://docs.indeed.com/job-sync-api/for-japan-partners/placement-companies

11. Indeed Partner Docs, "Employer Data API" (requirement to create the employer entity before the job; scope of employerName／employerType)

https://docs.indeed.com/indeed-plus/employer-data-api/

12. Indeed Partner Docs, "Create an employer"

https://docs.indeed.com/employers/operations/create-employer

13. Indeed Partner Docs, "Job types" (employment-type codes including dispatched worker 8YWGX／indefinite-term dispatching 5T857／contracting work T65DZ. Last updated July 21, 2026)

https://docs.indeed.com/job-sync-api/reference/job-types

14. Google Cloud, "Job Search basics" (the four categories of job board／career site provider／staffing and placement provider／ATS, and multi-tenancy)

https://docs.cloud.google.com/talent-solution/job-search/docs/basics

15. Google Cloud, "Tenant basics" (the structure in which a tenant owns the jobs and companies objects)

https://docs.cloud.google.com/talent-solution/job-search/v4beta1/docs/tenants

16. Google Cloud, "Cloud Talent Solution release notes" (event type for the case where a placement agency submits an application on behalf of a candidate)

https://docs.cloud.google.com/talent-solution/docs/release-notes

Tier 3｜Industry associations, academic work, and other

17. Japan Staffing Services Association (JASSA), publication of survey results on the business impact and actual use of generative AI at staffing companies (May 19, 2026. Responses from member companies and certified excellent dispatching business operators, 141 companies)

https://www.jassa.or.jp/information/6453/

18. Japan Institute for Labour Policy and Training (JILPT), on Keidanren's "Report on the Use of AI and the Like in HR Departments": proposals including securing safety, fairness, and transparency in responding to AI use in HR departments (Business Labor Trend, June 2026 issue)

https://www.jil.go.jp/kokunai/blt/backnumber/2026/06/kokunai_02.html

19. Keidanren (Japan Business Federation), "Report on the Use of AI and the Like in HR Departments" (April 14, 2026)

https://www.keidanren.or.jp/policy/2026/016.pdf

20. Ran Ziv, Ilan Gronau, Michael Fire, "CompanyName2Vec: Company Entity Matching Based on Job Ads," Proc. IEEE 9th International Conference on Data Science and Advanced Analytics (DSAA 2022), IEEE, 2022 (peer-reviewed. DOI: 10.1109/DSAA54385.2022.10032350)

https://arxiv.org/abs/2201.04687

21. Gerard Pons, Besim Bilalli, Anna Queralt, "Knowledge Graphs for Enhancing Large Language Models in Entity Disambiguation," Proc. 23rd International Semantic Web Conference (ISWC 2024), LNCS, Springer, 2024 (peer-reviewed. DOI: 10.1007/978-3-031-77844-5_9)

https://arxiv.org/abs/2505.02737

22. Mike Zhang, Rob van der Goot, Barbara Plank, "Entity Linking in the Job Market Domain," Findings of the Association for Computational Linguistics: EACL 2024, pp. 410–419 (peer-reviewed. The target is the mapping of skill descriptions in job text to the ESCO classification, not companies)

https://aclanthology.org/2024.findings-eacl.28/

This article organizes the issue of employer attribution in AI space for placement agencies, staffing companies, job boards, and ATS/HR-tech vendors. It is a spoke of the hub article How candidates research you with AI, and differs in reader and subject from Does JobPosting schema help AI?, which covers the perspective of the company that opens its own positions.

This article does not present an interpretation of the law, nor does it advise on individual cases. On whether a form of contract falls within a category, and on the lawfulness of a representation, please consult a certified social insurance and labor consultant or an attorney.

AIO, GEO, and LLMO are not names of official standards; they are terms used in practice. On that premise, this article confines itself to the issue of attribution accuracy.

By Vaigate Inc. (which operates Vaipm, measuring AI-space perception through a total of 25 stateless queries across multiple AI engines)

Published: August 10, 2026 ／ Sources checked: August 10, 2026