FAQPage, JSON-LD and schema in Hebrew: what actually helps answer engines

Schema — schema.org, in JSON-LD format — is a structured layer of information that explains to a machine what is on the page: who the business is, what the questions and answers are, what the steps in a process are. It does not promise a mention in a ChatGPT answer, and no provider has declared it a ranking factor in a model's answers. What it does do: it reduces the chance that your information is understood wrongly, and some of the types still earn rich results in Google. In Hebrew this matters especially, because the volume of Israeli text the models have seen is smaller, and every ounce of clarity is worth more.

What JSON-LD is, in two paragraphs

JSON-LD is a JSON block that sits inside a script tag on the page and describes the content in the terms of an agreed vocabulary — schema.org. It changes nothing on screen and is not mixed into the HTML itself.

That is also the practical point: because it is separate from the display, it does not break when someone changes the design. The older format — microdata, which scatters attributes across the HTML tags — breaks exactly that way, which is why JSON-LD is what Google recommends and also what is worth writing today.

Does schema affect ChatGPT and Gemini?

Here two providers have to be told apart, because the situation is not the same.

OpenAI: nothing has been published. There is no statement from OpenAI that schema affects ChatGPT's answers. Anyone who tells you otherwise is presenting a guess as a fact.

Google: published, and the statement is sober. In the official AI optimization guide Google addresses the subject directly, under the heading "Over-focusing on structured data", and writes that structured data is not required for AI features in Search and that there is no special schema.org markup you need to add — but that it is worth carrying on using it as part of a general SEO strategy, because it helps with eligibility for rich results.

That is a statement worth reading twice, because it puts out two fires at once: both the promise that schema will get you into AI answers, and the opposite claim that there is no point to it. Schema is good infrastructure. It is not a lever.

On the same page Google also warns against third-party tools that promise success in rankings or claim to use its "internal" metrics. We are a third-party tool, so we will say it explicitly: we measure through the providers' APIs — not through anyone's internal metrics — and what we measure is not byte-for-byte what a person sees in the consumer app.

What changed: FAQ and HowTo no longer produce a rich result

This is the part most Hebrew guides have still not been updated about, and it changes whether the investment is worth making.

• HowTo — Google removed the documentation in September 2023, on the grounds that the rich result is no longer displayed, neither on desktop nor on mobile.
• FAQPage — in the same month it was narrowed to recognised government and health sites only. In May 2026 Google announced the deprecation, and the feature stopped appearing on 7 May 2026; in June 2026 the documentation was removed.

Both of these types are entirely valid under the schema.org specification — Google's current feature list simply no longer includes them. The practical conclusion: if you implemented FAQPage for the rich result, that reason has gone. If you implemented it in order to describe the page in a way a machine can read — it still does exactly that, and that is why this page carries it too.

The three types worth having for every Israeli business

Organization or LocalBusiness

The only one of the three that still produces a rich result, and therefore the first one to implement.

{
  "@context": "https://schema.org",
  "@type": "LocalBusiness",
  "name": "משרד לוי ושות'",
  "inLanguage": "he-IL",
  "url": "https://example.co.il",
  "telephone": "+972-3-1234567",
  "address": {
    "@type": "PostalAddress",
    "streetAddress": "רחוב הרצל 10",
    "addressLocality": "תל אביב",
    "postalCode": "6120101",
    "addressCountry": "IL"
  },
  "areaServed": { "@type": "City", "name": "תל אביב" }
}

Note PostalAddress as an object and not as a free-text string, and addressCountry as the country code and not as a full name. The values stay in Hebrew — משרד לוי ושות' ("the Levy & Co. firm"), at רחוב הרצל 10 ("10 Herzl Street") in תל אביב ("Tel Aviv") — because the page they describe is in Hebrew.

FAQPage

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "inLanguage": "he-IL",
  "mainEntity": [{
    "@type": "Question",
    "name": "כמה עולה ייצוג בתביעת פיטורים?",
    "acceptedAnswer": {
      "@type": "Answer",
      "text": "שכר הטרחה נקבע לפי מורכבות התיק והשלב שבו הוא נמצא."
    }
  }]
}

Every question here has to appear as visible text on the page. That is not a stylistic recommendation — see the mistakes section. The question in the block asks "כמה עולה ייצוג בתביעת פיטורים?" ("how much does representation in a dismissal claim cost?"), and the answer says the fee is set by the complexity of the case and the stage it has reached.

HowTo

{
  "@context": "https://schema.org",
  "@type": "HowTo",
  "name": "איך מגישים תביעה לבית הדין לעבודה",
  "inLanguage": "he-IL",
  "step": [
    { "@type": "HowToStep", "position": 1, "name": "לאסוף מסמכים",
      "text": "תלושי שכר, חוזה העסקה והתכתבויות רלוונטיות." },
    { "@type": "HowToStep", "position": 2, "name": "להגיש כתב תביעה",
      "text": "כתב התביעה מוגש לבית הדין האזורי לעבודה שבאזור מקום העבודה." }
  ]
}

The procedure in the block is "איך מגישים תביעה לבית הדין לעבודה" ("how to file a claim with the labour court"): first "לאסוף מסמכים" ("gather documents") — payslips, the employment contract and relevant correspondence — and then "להגיש כתב תביעה" ("file a statement of claim"), with the regional labour court for the area of the workplace.

The type specific to your industry

Industry	The type
Law firm	LegalService
Estate agency and property	RealEstateAgent
Insurance agency	InsuranceAgency
Financial advice	FinancialService

They all inherit from LocalBusiness, so all the fields above apply — just swap the @type.

Common mistakes in Hebrew

The section people come back to:

• inLanguage missing or set to en. It comes from a template default. Schema with Hebrew values that declares itself to be in English hands over contradictory information.
• lang="he" or dir="rtl" missing on the html element. The schema is valid, and the page itself declares neither a language nor a direction.
• Schema that describes content which is not on the page. The only mistake on the list that can do harm and not merely fail to help: Google's guidelines treat structured content that does not represent the page as a violation, not as a bug.
• Mixing English and Hebrew in the same entity. "name": "Levy & Co" next to an address in Hebrew makes it harder to recognise that this is one entity. Choose one language per entity.
• An address as a string. "address": "רחוב הרצל 10, תל אביב" ("10 Herzl Street, Tel Aviv") is text. PostalAddress is data.
• A FAQPage whose questions do not appear on the page. A particular case of the dangerous mistake, and the most common one — because it is easy to generate the block from a data file and forget to render it.

The pattern behind almost all of them: schema is written once and forgotten. It does not appear on screen, it does not fail a build, and nobody looks at it again until somebody checks. The only approach that holds up over time is to generate the schema and the visible content from the same data source, so that the gap is impossible rather than forbidden.

What schema will not do

Short, because it matters:

• It will not create content. If the site holds no answer — no tag will invent one.
• It will not get you into an answer you do not belong in. It describes, it does not persuade. What decides is which text ties you to the need.
• It will not make up for a brand no source mentions. Excellent schema describes very well a business nobody has heard of.

Schema is a multiplier, not a source. Multiply zero and you get zero.

And what about llms.txt?

A question that gets asked alongside schema, and there is an explicit answer to it. In the same AI guide Google writes that llms.txt files are not required for its search, that it ignores them, and that they will neither harm nor help ranking — but that it is perfectly fine to keep them for other systems that do use them.

We keep a file like that on exactly that logic: not in order to influence Google, but so that a system reading it finds precise definitions instead of guessing. If someone is selling you llms.txt as a ranking trick in Google — the provider itself has already answered that.

And how do you measure whether any of this worked

Two ways, and both are legitimate. Google Search Console has a performance report for Google's AI features, and it is free and comes from the provider itself — if your question is about Google, that is the place to start. If the question is who gets mentioned in Hebrew answers across several engines, that is what we measure, and here is the explanation of what AEO is in the first place.

In both cases the principle is the same: measure before, change one thing, and measure again with exactly the same questions.

And finally — this page

This article itself carries Article, HowTo and FAQPage. The steps and the questions below are generated from the same source that renders them on screen, so the schema cannot describe content that is not here — not because someone remembered to check, but because the architecture does not allow anything else.

An article about schema whose own schema is broken is the worst possible advertisement for the idea.

If in a few months you want to know whether any of this moved the needle, our check is free and takes about a minute.