Practical Guides

Does JobPosting Structured Data Help in AI Search? — What JobPosting Actually Means, and How to Verify the Effect

2026-08-09Reading time 22min

By Vaipm (which measures AI-space perception through a total of 25 stateless queries across multiple AI engines)

Key point

Does JobPosting structured data help in AI search? Google's guidance does not confirm it. What Google says, what hiringOrganization means, and how to verify.

Executive summary

The explanation that adding JobPosting structured data to a job page makes it more likely to be picked up by AI circulates widely. Within the scope of this review, we could not substantiate that claim in Google's official documentation.

What Google officially states goes as far as this: implementing JobPosting makes a job page eligible to appear in the job experience in search results. On generative AI features, the official Google guide updated on July 10, 2026 states that structured data is not something generative AI search requires and that there is no special schema.org markup to add. Google itself lists this among the items it presents to correct misunderstandings. Microsoft Bing, on the other hand, states in its official blog that structured content can help AI interpret and summarize a page. Positions differ by platform, and neither of them has verified any effect specific to JobPosting.

Concluding from this that JobPosting is pointless would be equally inaccurate. Alongside satisfying the requirements of the job search experience, JobPosting has a second job: declaring, in machine-readable form, who the employer is. hiringOrganization means the company that actually employs the person rather than the name of the hiring site, and the specification provides an official rule, confidential, for anonymous postings. Comparable requirements exist at Indeed and in Google Cloud Talent Solution, and under Japanese law, keeping job information accurate and current is an obligation under the Employment Security Act.

JobPosting, in other words, is not something you implement as an AI tactic. The accurate framing is that you implement it to satisfy the requirements of the job search experience and to declare employer attribution correctly. On top of that, because nobody officially promises how AI will handle your job postings, you have to measure that yourself.

What you will learn

  • How far the claim "JobPosting gives you an advantage with AI" can be substantiated, and where the substantiation stops
  • What Google and Microsoft Bing each say about the relationship between structured data and generative AI
  • The precise meaning of hiringOrganization, confidential, jobLocation, jobLocationType, directApply, and validThrough
  • How platforms other than Google also require the employer to be declared
  • A concrete 2026 example showing that structured data effects are not permanent
  • How to check the way AI is handling your own job postings
  • What we could not confirm in this review

Who this is for

Recruiting and employer-branding practitioners, IT teams that operate a company careers site, and product or planning staff at HR tech, ATS, and job-board companies. This article is not about how to write a job description; it is about how to handle the machine-readable part of a job page.

For the broader picture of how candidates research companies with generative AI, and what recruiting and employer-branding teams should assemble in response, see the hub article Candidates Are Researching You With AI. This article is a spoke of that hub and goes deep on JobPosting alone.

§0. Scope of this article — what we could confirm, and what we could not

In an article about a technical specification, it matters to state up front how far confirmation reaches and where it stops. If that boundary stays vague and the article moves on to practice, readers take as promised something that was never promised.

What we were able to confirm against primary sources is the specification of Google's JobPosting structured data (its required, recommended, and beta properties), how Google and Microsoft Bing each describe the relationship between structured data and generative AI, the handling of employer data that Indeed's partner documentation requires, the classification of business roles used by Google Cloud Talent Solution, the duty of accurate display of job information under Japan's Employment Security Act along with the related certification scheme, and a concrete case in which a display effect produced by structured data came to an end (FAQ rich results).

What we could not confirm is collected in §9. In short: within the scope of this review, we could not confirm either an official Google document stating that JobPosting works directly on recommendation or citation in generative AI answers, or a third-party controlled experiment in which implementation was the single variable.

Note that every specification described here reflects the state as of August 2026. Always check each company's current documentation before implementing. The role of this article is less the specification itself, which can change, than how to read a specification.

§1. "Add JobPosting and AI will pick you up" — starting with that assumption

Search for information about structured data on job pages and you will meet an explanation along these lines.

> Implement JobPosting structured data and your job page will not just appear in Google for Jobs; it will also gain an advantage in AI Overviews and become more likely to be cited by generative AI.

The two halves of that sentence rest on very different strengths of evidence.

The opening half holds up. Google states in its own documentation that adding JobPosting structured data makes a job page eligible to appear in a special user experience in Google Search results.

For the latter half, within the scope of this review, we could not find substantiation. We looked through Google's public documents for anything touching the relationship between JobPosting and generative AI features, and we could not confirm a statement to the effect that implementing JobPosting confers an advantage in generative AI answers. Google in fact states explicitly that structured data is neither a dedicated requirement nor a prerequisite for generative AI search (§2-2).

1-1. Why this misunderstanding arises

Explanations of structured data carry a confusing structure inside them. "Making something machine-readable" and "being favored as a result of a machine reading it" are separate matters, yet in everyday language both get described as "AI understands it." Implementing JobPosting does make the job information on the page machine-readable, but whether that raises the probability that your posting is selected in a generative AI answer is a different question. You can check the former yourself; the latter is platform behavior, and unless it is published, it cannot be seen from outside.

A second reason is generalization from past success. There genuinely was a period when adding FAQPage schema produced an accordion display in search results. The model "add schema, receive preferential display" was built from that experience. That the model itself was not permanent is what §6 examines.

1-2. Watch out for the error in the opposite direction

Swinging from here to "so JobPosting is unnecessary" gets it wrong twice over.

To appear in the job search experience, JobPosting is a requirement. Without it, a page does not show up in that experience. And as described below, JobPosting is also a means of declaring employer attribution. "It does not work on AI" and "it is unnecessary" are entirely separate statements.

§2. What Google actually says

The public Google documents that give you something to judge by fall into two broad groups: those about job postings, and those about generative AI features.

2-1. On the job search experience — as far as "becomes eligible to appear"

Google Search Central's JobPosting documentation (last updated December 18, 2025) explains that adding the structured data makes a job page eligible to appear in the job experience within search results. The benefits it lists are a display accompanied by a logo and by reviews and ratings, reaching more motivated applicants through filters on work location and job type, and increased opportunity for discovery and application.

The word to notice is eligible. The same documentation states plainly that satisfying all requirements, best practices, and policies does not guarantee that Google will crawl, index, or serve the content. Structured data can be a necessary condition for a display, but it is not a sufficient one.

The job search experience also has a stated availability footprint. In Asia it covers 16 countries and regions, including Japan; in Europe, 14 countries; North America, Central and South America, and Sub-Saharan Africa are covered as whole regions.

2-2. On generative AI features — stated to be "not a requirement"

The official Google guide on optimizing for generative AI features, updated July 10, 2026, has a section devoted to correcting misunderstandings, and one of the items it handles is structured data. There, Google states that structured data is not a requirement for generative AI search, is not specific to it, and calls for no special schema.org markup to be added. It adds, at the same time, that continuing to use structured data as part of an SEO strategy remains a good idea, because it bears on rich result eligibility.

The same section lines up further items as things not to worry about: machine-readable files such as llms.txt and AI-specific markup are unnecessary, because Google Search does not use them; there is no need to chop content into fragments; there is no need to rewrite prose for AI; and collecting unnatural mentions is less effective than it appears.

This is consistent with the older "AI features and your website" document (last updated December 10, 2025). That document explains that there are no additional requirements for appearing in AI Overviews or AI Mode and no special optimization needed, and that what is required is that the page be indexed and eligible to be shown with a snippet.

2-3. So what does Google recommend?

The same guide points to the fundamentals of conventional SEO as what actually works for generative AI features: holding a perspective that is not a rehash of what already exists, structuring pages so readers can follow them, not blocking crawling, providing important information as text, and keeping structured data consistent with the visible text on the page.

Google's position is consistent. Generative AI features are built on top of Google's core ranking and quality systems, and AI Overviews and AI Mode draw information from the same index as ordinary search. On the territory people call AEO or GEO, the guide states that, from the standpoint of Google Search, optimizing for generative AI search is optimizing for the search experience, and is still SEO.

Note also that labels such as AIO, GEO, and LLMO are practitioner terms rather than official specifications or standards. We cover that point in What Is AIO? and What Is LLMO?.

2-4. A technical precondition for generative AI features

Google's guide carries one more piece of technical detail worth attention. To be eligible to appear in generative AI features, beyond having the page indexed and displayable with a snippet, the site needs to be included in the scope for generative AI features in Search Console.

That is not an "add some markup" kind of action. It is a setting that bears on whether inclusion happens at all, which makes it an item to check before markup rather than after.

§3. So what is JobPosting for?

If it promises no preferential treatment in generative AI answers, what is the reason to implement JobPosting? There are two, and they are worth separating.

3-1. Role 1: satisfying the requirements of the job search experience

As set out in §2-1, this is the requirement for appearing in the job search experience, and it is an effect Google states officially. That experience is available in Japan as well. It comes with implementation constraints: the structured data belongs on a page that handles a single job posting, it does not belong on job listing or search results pages, and the content of the markup has to exist visibly on the page. The documentation states explicitly that these can be treated as policy violations subject to manual action.

3-2. Role 2: declaring, in machine-readable form, who the employer is

The second role is rarely discussed in practice, but it is unmistakable once you read the specification. JobPosting is also a format for declaring, in structured form, which company employs the person the posting is for.

This role carries meaning independently of whether the page appears in the job search experience. It is normal for job information to exist simultaneously in several places: a company's own careers site, job boards, the listing pages of recruitment agencies. When the posting entity varies and nothing declares which company is the employer for a given job, whoever aggregates the information has to infer it. And wherever inference enters, error enters with it. Misinformation and misattribution around job information follow the same structure we cover in AI Misinformation About Your Brand.

3-3. The two roles have different priorities

What matters in practice is that Role 2 has a longer life than Role 1. The specification and the availability footprint of the job search experience can change. Display effects can end, as §6 shows. But the value of accurate information about who the employer is does not depend on any particular display feature. In Japan, it is also a legal obligation (§5-3).

So if you place the motive for implementation on "AI tactics," you lose the reason to maintain the implementation the moment the premises change. Place it on "declaring job information accurately," and the implementation survives a change in premises.

§4. What the main properties actually mean

This is the core of the article. Everything below rests on Google Search Central's JobPosting documentation (confirmed August 9, 2026). Specifications change, so check the current version when you implement.

The properties Google supports fall into three tiers: required, recommended, and beta. Five properties are listed as required (datePosted, description, hiringOrganization, jobLocation, and title). There is an exception for jobLocation: it is not required for a 100% remote job that uses applicantLocationRequirements.

4-1. hiringOrganization — the company name, not the site name

This is the property where mistakes happen most often. The documentation states that it is the company that is offering the job position and that it has to be the name of the company. The contrast it draws is clear: a legal entity name such as "Starbucks, Inc" belongs here, and a hiring-location name such as "Starbucks on Main Street" does not.

"hiringOrganization": {
  "@type": "Organization",
  "name": "MagsRUs Wheel Company",
  "sameAs": "http://www.magsruswheelcompany.com"
}

sameAs can point to the company's website. Where companies share or resemble each other's names, a name alone does not resolve to one entity, and using this as an aid to identification is reasonable. Google does not, however, guarantee that effect in the JobPosting documentation.

4-2. confidential — the official rule for anonymous postings

Here sits a rule many practitioners do not know. The documentation specifies that when an organization is hiring anonymously — the examples given are a staffing or recruitment services provider posting on behalf of an anonymous employer, and an employer posting anonymously directly on a platform — you set hiringOrganization.name to the value confidential, in lower case.

"hiringOrganization": {
  "@type": "Organization",
  "name": "confidential"
}

What matters is that even an anonymous posting has a prescribed way to declare that it is anonymous. When an employer wants to stay unnamed, practitioners sometimes enter the posting company's name or a made-up name, or drop the property altogether; the answer the specification gives is confidential. Misrepresenting the employer is clearly prohibited by Google's job posting content policies, which list impersonating another organization, listings that do not accurately represent the actual job, postings for jobs that do not exist, and posting on behalf of another company without authorization among the violations.

Note that "the company that actually employs the person" is not a synonym for the company where the work happens. Under Japanese law, a temporary staffing worker is legally employed by the dispatching agency, not by the client company where the work is performed — so do not read hiringOrganization as "the workplace." How "who is posting this job" becomes structurally hard to see in postings routed through recruitment agencies, staffing firms, and applicant tracking systems is a topic in its own right. A separate article in this series (employer attribution in the staffing and recruitment industry) is planned to cover it; here we keep to the specification.

4-3. jobLocation — where the work happens, not where it was posted

jobLocation means the physical location of the business where the employee actually reports to work. The documentation states explicitly that this is not the location where the job was posted. addressCountry is required. Where there are multiple work locations, they are written as an array, and Google is described as selecting a location suitable for display.

4-4. jobLocationType and applicantLocationRequirements — expressing fully remote work

Fully remote jobs have their own dedicated expression. You set jobLocationType to TELECOMMUTE and state in the job description that the role is 100% remote. The requirement is strict: the documentation defines a TELECOMMUTE job as one that has to be completely remote, and prohibits this markup for arrangements that are not 100% remote, such as jobs where occasional work from home is permitted or where remote work is a negotiable benefit.

applicantLocationRequirements specifies the geographic area in which an applicant may be located. For a 100% remote job with no physical work location, this property specifies where applicants may apply from, and it needs to indicate a scope of at least one country. Where both a physical work location and remote work are permitted, one configuration uses the country given in jobLocation as the default scope.

"applicantLocationRequirements": {
  "@type": "Country",
  "name": "USA"
},
"jobLocationType": "TELECOMMUTE"

4-5. directApply — a property whose effect is stated to be still in development

directApply is a boolean indicating whether an applicant can apply directly from the job URL. The specification itself carries a caveat worth noticing: how this information will be used is still under development, and you may not see any immediate display or effect in Google Search.

By Google's definition, a direct apply experience is one that offers a short and simple application process without unnecessary intermediate steps; if the applicant is asked to click apply, fill out forms, or sign in multiple times, it is not a direct apply.

This property demonstrates the theme of this article from inside the specification. Implementing something does not necessarily produce a display or an effect — and here Google says so itself.

4-6. validThrough and handling closed postings — where manual action risk lives

validThrough is the date and time at which a job posting expires. It is described as required for jobs that have an expiry date, and as something not to specify for jobs where no expiry is known.

What matters is the handling when a posting closes. The documentation states that jobs no longer accepting applications have to be closed out by prescribed methods, and that failing to deal with expired postings in a timely manner may result in manual action. There are three methods — set validThrough to a past date and time, remove the page itself and return a 404 or 410, or strip the JobPosting structured data from the page. Google recommends removing expired postings from the site as the ideal, and describes setting validThrough to a past date and time as what to do when the page is not removed.

In practice this is the part most often overlooked. Job pages fail more readily at the close than at the launch. When integrating with an applicant tracking system, the closing flow needs to be designed with the same precision as the publishing flow.

4-7. Other recommended properties and beta properties

The recommended properties include baseSalary (the actual base salary the employer is offering, with unitText as one of HOUR through YEAR, and ranges given via minValue and maxValue), employmentType (FULL_TIME, PART_TIME, CONTRACTOR and others, eight values in total, multiple values permitted), and identifier (the employer's own job identifier). The four beta properties covering education and experience carry the same caveat as directApply.

The documentation also states that title takes the job title itself, and that job codes, addresses, dates, salaries, and company names do not belong in it.

4-8. Markup has to match the visible text

Separate from any individual property, one principle covers the whole. All information contained in the markup has to be present visibly on the job posting page.

The documentation gives as a violation the case where a salary appears in the markup but is not displayed on the page. This principle matches the recommendation in the guide on generative AI features. Structured data is not a place to make additional claims that the page does not make.

§5. Declaring the employer is required outside Google too

The requirement to make the employer explicit is not particular to Google. This is what reinforces the conclusion that JobPosting is not an AI tactic.

5-1. Indeed — the employer comes before the job

Indeed's partner documentation imposes a clear order on integrating parties such as ATS vendors: employer data has to be submitted before any job tied to that employer can be created.

An employer entity is managed as a pair: a type that uniquely identifies the partner's system, and an id that uniquely identifies the employer within that system. employerName looks optional on the input object, yet the documentation states that it is required when creating an employer. Attributes are defined across three scopes — global, country, and locale — and the global scope, where the value is the same in every country, holds employerName and employerType. Indeed is described as reviewing submitted employer data for completeness and appropriateness.

In short, Indeed manages the employer as an entity independent of the job posting and requires its creation before a job can be created.

5-2. Google Cloud Talent Solution — business roles are kept apart

The Job Search documentation for Google Cloud Talent Solution (last updated July 22, 2026) divides expected usage into four basic use cases: job boards, providers that supply careers-site services to client companies, staffing agencies, and applicant tracking systems that follow applicants through the hiring process.

In the job domain, the assumption that the entity posting a job and the entity employing the person can be different appears to be built into the design — that is what can be read from the platform's own classification. This is a reading of the classification, not a statement the documentation makes in those words.

5-3. Japan's Employment Security Act — accuracy is a legal obligation

In Japan, the accuracy of job information is not a recommendation but an obligation. The amended Employment Security Act (Japan), promulgated on March 31, 2022 and in force from October 1 of the same year, made accurate display of job-related information mandatory.

Article 5-4 of the Employment Security Act separates the obligation into three parts. Paragraph 1 applies to Public Employment Security Offices, designated local governments, employment placement business operators, persons conducting recruitment of workers, commissioned recruiters, businesses providing recruitment information and similar services, and labor supply business operators, and provides that they must not make false or misleading representations. Paragraph 2 is a direct obligation on persons conducting recruitment and on commissioned recruiters to keep information relating to the recruitment accurate and up to date. Paragraph 3 is an obligation to take measures, requiring employment placement business operators and providers of recruitment information and similar services to take the measures prescribed by Ministry of Health, Labour and Welfare ordinance to keep information accurate and up to date.

The structure of the obligation differs by actor. An operating company that recruits on its own site sits on the direct-obligation side of paragraph 2; job media and ATS vendors sit on the measures-obligation side of paragraph 3.

This is where the technical specification and the legal framework meet. The duty to keep information accurate and up to date demands the same thing as §4-6 on closing postings, approached from the other side. Leaving an expired posting in place is a risk on the search platform and a legal question at the same time.

One point must not be conflated, however. What the Employment Security Act requires is accurate display, accuracy, and currency of the job information itselfnot the implementation of JobPosting structured data. What the law requires is the substance of the information, not a schema.org property.

5-4. A statutory obligation and a voluntary quality certification coexist

The Japanese industry council for the proper handling of job information (Kyujin Joho Tekiseika Suishin Kyogikai) operated a declaration scheme for media conforming to its job information provision guidelines from June 1, 2018, and the council's own site indicates that the scheme ended on May 31, 2025. We could not confirm an official explanation of the reason, so this article does not speculate about it.

Meanwhile, the certification scheme for excellent providers of recruitment information (Yuryo Boshu Joho to Teikyo Jigyosha Nintei Seido), commissioned by the Ministry of Health, Labour and Welfare, is still in operation. Established in fiscal 2022, it certifies providers that meet standards covering legal compliance, accurate display of recruitment information, handling of personal data, information disclosure, screening, and complaint handling. Certification is granted per business operator and is valid for three years.

The framework around the quality of job information in Japan therefore has a shape in which a statutory obligation and a voluntary quality certification coexist.

5-5. Generative AI is starting to enter public services too

Hello Work Internet Service, Japan's public employment service portal, began trial operation of a generative AI chatbot on January 19, 2026. According to the Ministry of Health, Labour and Welfare, it uses generative AI to compose answers automatically to questions about using Hello Work, such as searching for work and how to submit a job listing, and 1,000 monitor users were recruited for the trial (recruitment has since closed).

Note that the published materials do not let us confirm that AI is recommending or summarizing individual job postings. What can be confirmed is that a trial in which generative AI is involved in job-search and application procedures has begun even within a public employment service.

§6. Display effects from structured data are not permanent

How long does the premise "add structured data and you receive a display benefit" actually hold? In 2026 there was a concrete case that answers the question.

6-1. The end of FAQ rich results — a staged disappearance

According to Google Search Central's documentation update history, FAQ rich results stopped appearing in Google Search from May 7, 2026, and Google added the deprecation notice to the FAQ structured data documentation the following day, May 8. Then, on June 15, 2026, Google removed the documentation for the feature itself.

Reporting at the time of the May notice indicated that support for the FAQ search appearance in the Search Console API was scheduled for removal in August 2026; because the notice document has itself been removed, we keep that single point as a statement resting on secondary reporting.

What matters is what disappeared. What ended was the display decoration in search results; the FAQPage schema.org type itself has not disappeared.

Display effects from structured data can end at the platform's discretion. The motive for an implementation needs to be designed with that in view (§3-3).

6-2. What correlation data shows, and what it does not

Vendors publish studies on the relationship between structured data and AI citation. They need to be read with care.

The study Semrush published in January 2026 analyzed 5 million URLs cited by ChatGPT Search and Google AI Mode. The schema types reported as present on cited pages include Organization at 25% (ChatGPT) and 34% (AI Mode), Article at 20% and 26%, Breadcrumb at 15% and 20%, and FAQ at 3% and 5.5%. By format, pages carrying schema.org (JSON-LD) accounted for about 40% of AI Mode cited pages and about 30% of ChatGPT cited pages.

Those figures carry two implications. To begin with, calculated from the proportions Semrush reported, the majority of cited pages do not carry schema.org JSON-LD. The claim "without schema you do not get cited" is not supported by this study.

Second, JobPosting is not among the schema types broken out in this study. That does not mean JobPosting does not get cited; it means the published breakdown does not cover it.

And these are correlations, not causation. Semrush notes this itself. The same correlation would arise under the explanation that pages likely to be cited also have the operational capacity to implement structured data. Semrush also supplies AI visibility tooling, which makes it an interested party in the subject it is studying.

6-3. What the academic research shows, and what it does not

Academic work to the effect that structured data improves AI accuracy does exist. The subject matter, however, is different.

SRAG (Structured Retrieval-Augmented Generation, arXiv:2503.01346) — a preprint that has not been peer reviewed — reports that organizing extracted entities into relational tables and then reasoning over them in tabular form improved accuracy on multi-entity question answering by 29.6% against existing methods. The survey of Retrieval And Structuring Augmented Generation published at KDD 2025 (arXiv:2509.10697), a formally published paper, likewise reviews across the literature how structured data such as knowledge graphs contributes to improved retrieval quality.

A distinction is needed here. What these studies address is how data is structured inside retrieval and generation systems — not how schema.org markup on your public pages changes what AI answers select. Using findings about the former as grounds for the latter substitutes one subject for another. Even where the general proposition "structuring helps AI" holds, it does not supply grounds for the particular proposition "the JobPosting markup on your page changes what AI answers."

6-4. Platforms take different positions

Platform positions on the relationship between structured data and AI are not uniform.

Google, as described above, states in its official guide that structured data is not something generative AI search requires and that no special schema.org markup needs to be added.

Microsoft Bing states a view in a different direction in its official Webmaster Blog. A November 2025 post says that giving content structure — product pages carrying schema markup, FAQ sections, comparison tables — makes it easier for AI systems to interpret a page and to condense what it says, and that this raises the likelihood of a page being cited, of readers clicking it, and of readers engaging with it inside answers assembled from several sources. The February 2026 post announcing AI Performance in Bing Webmaster Tools as a public preview likewise says that headings that are clear, together with tables and FAQ sections, help bring the important information to the surface and make a page easier for AI systems to reference without error.

These statements do not, however, verify or guarantee any effect specific to JobPosting. What Bing names is product pages, FAQs, and comparison tables; it does not present verification results for job structured data in particular.

The question "does it work on AI" does not have a single answer across platforms. Google says it is not a dedicated requirement; Microsoft says it can help interpretation. Arguing the question without specifying which AI is under discussion does not hold together at all.

§7. How to verify the effect — measuring what is not promised

One practical problem remains after everything above. Because Google does not promise an effect, how your job postings are actually treated in AI space is not something handed to you from outside. If you want to know, you have to measure.

7-1. Look at more than the counts — look at how you were described

Impressions and citation counts are provided as official metrics by both Search Console and Bing Webmaster Tools, and there is nothing meaningless about them. The problem is that they are not enough on their own for practical decisions about job information. What you need is not just impressions, but what was said.

The following are observation items this article proposes. They are not standard metrics derived from the studies or the official documentation cited here.

  • Citation and mention: when someone asks about your openings or your hiring conditions, does the AI mention your company, and does it name your pages as sources
  • Share within the answer: for the same question, how often do you appear relative to competitors
  • How you are described: is the content positive, negative, or neutral
  • Accuracy: do the work location, employment type, salary, and application method presented match your current postings
  • Freshness: are closed postings or superseded conditions still being described
  • Employer attribution: is your posting being described as a listing belonging to a job board or a staffing provider

The last two are specific to the job domain. As §4-6 and §5-3 set out, leaving an expired posting in place carries both a risk on the search platform and a legal question.

7-2. What official tools show, and what they do not

On June 3, 2026, Google announced that it had introduced generative AI performance reports in Search Console. They supply a dedicated view of impressions in AI Overviews, AI Mode, and the generative AI features in Discover. The announcement explains that this data was already included in the overall performance report, that what was added is a dedicated view limited to generative AI features, and that it is being rolled out to some sites in stages. The same applies to the inclusion setting described in §2-4. So you can see only within the range that is actually being shown for your own property.

On the Microsoft side there is AI Performance in Bing Webmaster Tools, which became a public preview in February 2026. What it shows is citation activity across the surfaces it covers: Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations.

Each platform shows you its own surfaces and no others. How you are handled in ChatGPT or Perplexity appears in neither. And impressions and citation counts do not tell you what §7-1 calls "how you were described" or "whether it was accurate."

On June 5, 2026, Google added guidance about third-party SEO tools, services, and advice, stating that people should be wary of third parties that promise ranking improvements or claim to use "internal metrics," and that third-party tools do not have access to Google's internal ranking or AI systems. That is a correct point, and a premise anyone doing measurement has to respect. What can be measured is the output — the answer the AI returned — not the internal mechanism.

7-3. Why repetition and statelessness are needed

Generative AI answers vary from run to run for the same question, and they are influenced by conversation history and by prior usage. Asking once and recording the result is not a measurement of durable perception.

As an operational design, Vaipm measures AI-space perception through a total of 25 stateless queries across multiple AI engines. Stateless means that each measurement runs without carrying over prior history. The design intent is that this removes conversation history as a confounder and improves comparability under the same conditions. Variation from model updates, the search index, region, time of day, and sampling remains. This count is also an operational design Vaipm uses; it is not an optimum derived from the studies cited in this article.

Why the design matters in the job domain is that a single check cannot separate a run that happened to answer correctly from a run that happened to answer with outdated conditions. What you want to know is what tends to happen when a candidate asks, not what happened on one occasion.

7-4. Practical notes for designing measurement of job information

  • Write the questions in a candidate's words: phrase them the way candidates actually ask — "what roles and locations is [company] hiring for," "does [company] allow remote work"
  • Build the answer key before you start: document your current posting conditions (role, location, employment type, salary range, application method, closing date). Without a reference, you cannot judge an answer right or wrong
  • Include closed postings: check whether a recently closed posting is still being described as open
  • Check employer attribution: look at whether your posting is being described as a listing belonging to a job board or a staffing provider
  • Record the date, time, and conditions of measurement: specifications change and so do answers. Without a record of when and under what conditions you measured, comparison is impossible

For the wider practice of managing perception in AI space on an ongoing basis, see What Is AI Perception Management?; for the supply-side question of what AI draws on as its sources, see What AI Cites About Your Company.

§8. So what should a recruiting team do?

Here is everything above, put into practical order.

8-1. Start with crawling and text

The conditions Google's official guide names for generative AI features are not about markup. They are about indexing and snippet eligibility.

  1. Whether your job pages are blocked from crawling by robots.txt or a CDN configuration
  2. Whether the important information in a posting (role, location, employment type, salary, application method) exists as text
  3. Where pages are rendered with JavaScript, whether the content is retrievable after rendering
  4. Whether the site is included in the scope for generative AI features in Search Console

Adding markup while these remain unmet gets the order backwards.

8-2. Next, satisfy the requirements of the job search experience

If you want to appear in the job search experience, implement JobPosting correctly (details in §4). Put it on a single job posting page, not on listing or search results pages; satisfy the five required properties; put the name of the company that actually employs the person in hiringOrganization, using confidential for anonymous postings; put the place where the work happens in jobLocation. For fully remote roles, set jobLocationType and specify where applicants may be located. Then confirm that everything in the markup exists visibly on the page, and validate with the Rich Results Test.

8-3. Design the closing flow with the same precision as the launch

As §4-6 and §5-3 set out, this is the part most likely to fail and the part with the largest consequences. When a posting closes, reliably do one of the following: set validThrough to a past date and time, remove the page and return a 404 or 410, or strip the JobPosting structured data. Confirm that closure propagates automatically through your applicant tracking system integration, and that where the same job is posted across several outlets, there is a path by which closure reaches every one of them.

8-4. Do not implement it as an "AI tactic"

Moving the conclusion of this article straight into practice looks like this.

  • Do not justify a JobPosting implementation internally on the grounds of preferential treatment in generative AI answers. The grounds cannot be confirmed officially, so the justification collapses later
  • Do not adopt llms.txt or AI-specific markup as measures that work in Google
  • Do not present FAQPage schema as a way to obtain display decoration in search results (that effect ended on May 7, 2026)
  • Do not put anything in markup that the page does not say

8-5. And then measure

An implementation being correct and AI handling it correctly are separate things. Posting conditions change and so does AI behavior, so check periodically along the lines of §7.

§9. What we could not confirm in this review

The following are matters we could not confirm and cannot state definitively at this point.

1. An official Google document stating that JobPosting works on recommendation or citation in generative AI answers

Within the scope of this review, we could not confirm one. We do not assert that none exists, because Google's documentation gets updated. What the current guide does state explicitly is that structured data is neither a dedicated requirement nor a prerequisite for generative AI search.

2. A third-party controlled experiment in which the presence of a JobPosting implementation was the single variable

We could not confirm published research that varied the implementation alone while holding other conditions constant. The studies that do exist are observations of correlation, with factors other than implementation confounded into them.

3. Peer-reviewed research measuring the effect of schema.org markup on public pages on what an LLM selects for its answers

As §6-3 sets out, research on structured data and LLMs centers on structuring inside the systems themselves.

4. Platform verification results for effects specific to JobPosting

As §6-4 sets out, Microsoft Bing states in its official blog the view that structured content can help AI interpretation, but what it names is product pages, FAQs, and comparison tables, and it does not verify any effect specific to JobPosting. On the Google side, we could not confirm published results verifying the relationship between JobPosting and generative AI features in particular.

5. Primary research on AI visibility specific to the job domain in Japan

Several surveys of generative AI usage among job seekers exist, but within the scope of this review we could not confirm domestic primary research connecting the technical implementation of job pages to how they are treated in AI space.

FAQ

Q1. If I implement JobPosting, will AI be more likely to pick me up?

Within the scope of this review, we could not find an official Google document saying so. What Google states officially goes as far as becoming eligible to appear in the job search experience. On generative AI features, the official guide updated on July 10, 2026 states that structured data is not a requirement for generative AI search, is not specific to it, and calls for no special schema.org markup. The accurate place to put your reasons for implementing is satisfying the requirements of the job search experience and declaring employer attribution accurately.

Q2. So is it fine not to implement JobPosting?

No. If you want to appear in the job search experience, JobPosting is a requirement, and Japan is among the regions where that experience is available. It is also a means of declaring, in machine-readable form, who the employer is. "Does not promise preferential treatment in generative AI answers" and "is unnecessary" are separate statements.

Q3. What should go in hiringOrganization?

The name of the company that actually employs the person. Google's documentation states that this has to be the name of the company and must not be the name of an individual hiring location. You can also point sameAs at the company's website. Using that as an aid to distinguishing companies with the same name is reasonable, but Google does not guarantee that effect.

Q4. What should I do for a posting where I want to keep the employer unnamed?

Google's specification has an official rule. When hiring anonymously — a staffing or recruitment services provider posting on behalf of an anonymous employer, or an employer posting anonymously directly on a platform — you use confidential, in lower case, as hiringOrganization.name. Entering the posting company's name or a made-up name, or omitting the property, is not the answer the specification gives.

Q5. For a job routed through a recruitment agency or a staffing firm, who is written as the employer?

On the specification, hiringOrganization means the company that actually employs the person, and confidential is used when the employer is to remain unnamed. Note that under Japanese law a temporary staffing worker is legally employed by the dispatching agency, not by the client company where the work is performed. How "who is posting this job" becomes structurally hard to see in the staffing and recruitment industry is a topic in its own right, and a separate article in this series (employer attribution in the staffing and recruitment industry) is planned to cover it.

Q6. How do I describe a remote job?

For 100% remote jobs and no others, set jobLocationType to TELECOMMUTE and state in the job description that the role is fully remote. Where there is no physical work location, use applicantLocationRequirements to specify where applicants may be located, indicating a scope of at least one country. Where both a physical work location and remote work are permitted, one configuration uses the country in jobLocation as the default scope. Marking up a job that permits occasional work from home, or where remote work is a negotiable benefit, in this format is clearly prohibited.

Q7. How should I handle a posting whose recruitment has ended?

There are three methods: set validThrough to a past date and time, remove the page and return a 404 or 410, or strip the JobPosting structured data — with removal recommended as the ideal. The documentation states that failing to deal with expired postings in a timely manner may result in manual action. In Japan, on top of that, keeping recruitment information accurate and up to date is an obligation under the Employment Security Act.

Q8. If I put FAQPage schema on a job page, will it stand out in search results?

No. FAQ rich results stopped appearing in Google Search from May 7, 2026; Google added the deprecation notice to its documentation the following day, May 8; and on June 15, 2026 it removed the documentation for the feature itself. The FAQPage schema.org type itself has not disappeared, but you cannot present it as a measure for obtaining display decoration in search results.

Q9. If I publish llms.txt, will AI read me more readily?

As far as Google Search is concerned, the official guide denies it clearly. It states that there is no need to create new machine-readable files, AI text files, markup, or Markdown in order to appear in Google Search, generative AI features included, and that Google Search does not use them. Maintaining such a file for other services is unobjectionable, and it neither harms nor helps visibility or ranking in Google Search.

Q10. I saw a study showing that pages with structured data get cited by AI more often.

Studies reporting that correlation do exist, but they need to be read with care. In the 5-million-URL study Semrush published in January 2026, pages carrying schema.org JSON-LD accounted for about 40% of Google AI Mode cited pages and about 30% of ChatGPT cited pages. Calculated from the proportions the company reported, the majority of cited pages do not carry JSON-LD. The study itself notes that these are correlations, not causation. It is also a study by a company that supplies AI visibility tooling, so its position as an interested party belongs in how you read it.

Q11. Can I see how AI is handling my job postings in Search Console?

Partly. On June 3, 2026, Google announced that it had introduced generative AI performance reports in Search Console. A dedicated view of impressions in AI Overviews and AI Mode is provided, but it is being rolled out to some sites in stages, and its coverage is limited to Google's own surfaces. On the Microsoft side, AI Performance in Bing Webmaster Tools shows citation activity across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations. Both deal in impressions and citation counts, and neither tells you how your company was described inside an answer or whether the content was accurate. To know that, you have to check the answers the AI returns, continuously.

Q12. If I check once and find no problems, is that enough?

It is not. Generative AI answers vary from run to run for the same question and are influenced by conversation history. A single check cannot separate a run that happened to answer correctly from a run that happened to answer with outdated conditions. Posting conditions change, and so do search platform specifications. As an operational design, Vaipm measures AI-space perception through a total of 25 stateless queries across multiple AI engines. Variation caused by model updates, the search index, and similar factors remains even under that design.

Summary

JobPosting is a requirement for appearing in Google's job search experience. It is not a promise of preferential treatment in generative AI answers. Google's official guide states explicitly that structured data is neither a dedicated requirement nor a prerequisite for generative AI search. Positions differ by platform, though: Microsoft Bing states in its official blog the view that structured content can help AI interpretation. Neither of them has verified any effect specific to JobPosting.

JobPosting has a second role. It declares, in machine-readable form, who the employer is. That value does not depend on any particular display feature. Display effects are not permanent: FAQ rich results stopped appearing on May 7, 2026. Put the motive for implementation on "AI tactics" and you lose the reason to maintain it when the premises change; put it on "declaring job information accurately" and the implementation survives a change in premises.

And because no effect is officially promised, how you are actually treated is something you have to measure. Not just impressions, but what is being said and how, whether the conditions are accurate, whether closed postings are lingering, whether employer attribution is right. These sit outside the range of the official tools.

For the picture recruiting and employer-branding teams need in an era when candidates research companies with AI, see Candidates Are Researching You With AI.

Sources

Primary sources and official documentation

  1. Google Search Central, "Job posting (JobPosting) structured data for Job Search" (last updated December 18, 2025 / confirmed August 9, 2026) https://developers.google.com/search/docs/appearance/structured-data/job-posting
  2. Google Search Central, "Optimizing your website for generative AI features on Google Search" (last updated July 10, 2026 / confirmed August 9, 2026) https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
  3. Google Search Central, "AI features and your website" (last updated December 10, 2025 / confirmed August 9, 2026) https://developers.google.com/search/docs/appearance/ai-features
  4. Google Search Central Blog, "Introducing Search Generative AI performance reports in Search Console" (June 3, 2026) https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
  5. Google Search Central, "Latest Google Search Documentation Updates" (deprecation notice for FAQ rich results / confirmed August 9, 2026) https://developers.google.com/search/updates
  6. Google Cloud, "Job Search basics — Cloud Talent Solution" (last updated July 22, 2026 / confirmed August 9, 2026) https://docs.cloud.google.com/talent-solution/job-search/docs/basics
  7. Indeed Partner Docs, "Create an employer — Employer Data API" (confirmed August 9, 2026) https://docs.indeed.com/employer/operations/create-employer
  8. Bing Webmaster Blog, "How AI Search Is Changing the Way Conversions are Measured" (November 2025 / confirmed August 9, 2026) https://blogs.bing.com/webmaster/November-2025/How-AI-Search-Is-Changing%E2%80%AFthe%E2%80%AFWay%E2%80%AFConversions%E2%80%AFare-Measured
  9. Bing Webmaster Blog, "Introducing AI Performance in Bing Webmaster Tools – Public Preview" (February 2026 / confirmed August 9, 2026) https://blogs.bing.com/webmaster/February-2026/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview
  10. Ministry of Health, Labour and Welfare (Japan), guidance issued under the Employment Security Act on the display of worker recruitment advertisements (excerpt of Article 5-4 of the Employment Security Act / confirmed August 9, 2026) https://www.mhlw.go.jp/stf/seisakunitsuite/bunya/koyou_roudou/koyou/haken-shoukai/r0604anteisokukaisei1_00006.html
  11. Ministry of Health, Labour and Welfare (Japan), on the 2022 amendment to the Employment Security Act (promulgated March 31, 2022; in force October 1, 2022 / confirmed August 9, 2026) https://www.mhlw.go.jp/stf/seisakunitsuite/bunya/0000172497_00003.html
  12. Kyujin Joho Tekiseika Suishin Kyogikai (Japanese industry council for the proper handling of job information), declaration scheme for media conforming to the job information provision guidelines (the council's site indicates that the scheme has ended / confirmed August 9, 2026) http://tekiseika.jp/compatibility-system/
  13. Official site of the certification scheme for excellent providers of recruitment information (Yuryo Boshu Joho to Teikyo Jigyosha Nintei Seido), a project commissioned by the Ministry of Health, Labour and Welfare (established in fiscal 2022 / confirmed August 9, 2026) https://yuryonintei.com/authorization/introduction/
  14. Ministry of Health, Labour and Welfare (Japan), news release on the certification of 15 companies as excellent providers of recruitment information (confirmed August 9, 2026) https://www.mhlw.go.jp/stf/newpage_32215.html
  15. Ministry of Health, Labour and Welfare (Japan), Shokuba Joho Sogo Site, notice to employers and job seekers on the recruitment of monitor users for a generative AI chatbot (trial operation began January 19, 2026; recruitment closed. Updated January 19, 2026) https://shokuba.mhlw.go.jp/110/20251218150748.html

Peer-reviewed and other academic literature

  1. "SRAG: Structured Retrieval-Augmented Generation for Multi-Entity Question Answering over Wikipedia Graph," arXiv:2503.01346 (2025). A preprint that has not been peer reviewed. Reports a 29.6% improvement over existing methods in multi-entity question answering by structuring entities into relational tables. Its subject is structuring inside retrieval and generation systems; it does not demonstrate an effect of markup on public pages https://arxiv.org/html/2503.01346v1
  2. "A Survey on Retrieval And Structuring Augmented Generation with Large Language Models," published at KDD 2025 / arXiv:2509.10697. Reviews across the literature how structured data such as knowledge graphs contributes to improved retrieval quality. As above, its subject is structuring inside the systems themselves https://arxiv.org/pdf/2509.10697

Interested-party studies and industry reporting (indicative values)

  1. Semrush, "How Do Technical SEO Factors Impact AI Search? [Study]" (January 5, 2026). Analyzes 5 million URLs cited by ChatGPT Search and Google AI Mode. A study by an interested party that supplies AI visibility tooling; the company itself notes that these are correlations, not causation https://www.semrush.com/blog/technical-seo-impact-on-ai-search-study/
  2. Search Engine Journal, "Google Drops FAQ Rich Results From Search" (May 2026). The scheduled removal of Search Console API support (August 2026) rests on this reporting, because the notice document has already been removed https://www.searchenginejournal.com/google-drops-faq-rich-results-from-search/574429/

About this article

The statements in this article are based on information as of August 9, 2026. We distinguish the parts confirmed against primary sources from the parts resting on secondary sources and inference.

  • Confirmed against primary sources: Google's various documents and its documentation update history, Microsoft Bing's official Webmaster Blog, Indeed's partner documentation, the Google Cloud Talent Solution documentation, the text of Japan's Employment Security Act and materials published by the Ministry of Health, Labour and Welfare, and the official information on the certification scheme for excellent providers of recruitment information
  • Resting on secondary sources: the scheduled removal of Search Console API support for the FAQ search appearance (based on reporting, because the notice document has already been removed)
  • This article's own proposals and framing: the observation items in §7-1, and the framing in §3-3 that Role 2 has a longer life than Role 1

Search platform specifications and generative AI behavior change. Always check each company's current documentation before implementing. Where an effect could not be confirmed, we do not assert that it does not exist; we state that we could not confirm it within the scope of this review. We also do not treat studies showing correlation as grounds for causation.

By Vaipm (which measures AI-space perception through a total of 25 stateless queries across multiple AI engines)

Related articles