Do we need schema markup for AI assistants to understand our site?

Your web developer has mentioned "schema" in a quote, your agency says the site "needs structured data for AI", and nobody has explained what either means or whether it will make ChatGPT more likely to name your firm. The useful answer is smaller than the sales pitch.

Structured data is a short block of code, invisible to visitors, that tells machines plainly what a page is: this is a company, its name is X, it is based in Leeds, it provides these services, this article was written by this person on this date. Schema markup is the shared vocabulary for writing it, published at schema.org and backed by the major search engines.

What is on record is narrower than most agencies claim. Google's guide to AI features (updated December 2025) says there is "no special schema.org structured data that you need to add" to appear in AI Overviews or AI Mode. Microsoft's Bing team says schema helps "search engines and AI systems understand your content". OpenAI and Anthropic have not said whether they read it. And in an SE Ranking study of 216,524 pages, reported by Search Engine Journal in November 2025, pages with FAQ schema averaged 3.6 ChatGPT citations against 4.2 for pages without it.

So schema is not a ranking trick. It is how you state your facts in a form machines cannot misread, and that is still worth an afternoon.

#What is schema markup, in plain English?

It is information written in a fixed, machine-readable format rather than in sentences, usually in a format called JSON-LD in the page's code.

Compare two ways of saying the same thing. In prose: "Founded in 2015, our team of thirty has helped over two hundred manufacturing businesses improve their cash position." In structured data: founding date 2015, number of employees 30, area of expertise manufacturing. The prose is better for a human. The structured version tells a machine at once that these are three separate facts and what each one is.

#Does ChatGPT actually read schema markup?

Nobody outside OpenAI knows, because OpenAI has not said. What is on record comes from Google and Microsoft.

Google's AI optimisation guide, last updated in July 2026, calls "overfocusing on structured data" a myth, then adds that "it's a good idea to continue using it as part of your overall SEO strategy". Microsoft's statement comes from a Bing product manager's guide to AI search answers (October 2025).

Two practical points follow. ChatGPT search has drawn on Bing: in a small Seer Interactive test (February 2025), 87% of its citations matched Bing's top results. Markup that helps Bing read you may therefore help ChatGPT indirectly, though that is our inference, not something OpenAI has confirmed.

And how the markup reaches the page matters. Vercel and MERJ found (December 2024) that GPTBot, ClaudeBot and PerplexityBot fetch pages but do not run JavaScript. Markup injected by a tag manager after the page loads can be read by Googlebot and missed by the AI crawlers, so put it in the HTML the server sends. Our guide to how AI crawlers read your website explains what they see.

#Why does structured data help, if nobody requires it?

Because it removes doubt about the basic facts, and doubt is what makes an assistant hedge, guess or leave you out. Prose is ambiguous: "an established firm" could mean five years or fifty. Markup states the founding date, the company number and the exact services once, in a standard format.

It also makes contradictions visible. If your markup says you serve software companies and the page talks about retail, trust in both falls, which is why Google asks for markup that matches the visible text. Why ChatGPT describes firms wrongly shows what inconsistency costs. When we reviewed TCC Group, part of what failed was "machine-readable data that declared a name, a logo and nothing else."

What markup will not do is get thin content named, raise your Google ranking on its own or fix a site AI crawlers cannot reach. Content, mentions on other sites and clear positioning decide whether you are named; markup makes them easier to read.

#Which schema types does a B2B firm need?

Four types do most of the work: Organization, Person, Article and Service. The rest apply only where they describe you.

Spelling matters. Schema.org uses American spellings, so the type is Organization with a z; a hand-typed "Organisation" is not a schema.org type and validators will flag it. SoftwareApplication and BreadcrumbList are single words with two capitals each.

#Organization: the block every firm needs

Here is the Organization block on this site's homepage, trimmed (the full version also lists contact details, our three services and all three founders):

The @id gives the organisation a permanent address that every other block can point to, so Article and Person blocks say "this organisation" without repeating its details.

#sameAs: how machines join your site to facts elsewhere

Schema.org defines sameAs as the "URL of a reference Web page that unambiguously indicates the item's identity". Google's Organization markup documentation says the markup helps it "disambiguate your organization in search results". List only profiles that are yours and kept current: your LinkedIn company page, your Companies House record, a Wikidata entry if one exists and is accurate, Crunchbase, professional body or regulator listings, and review profiles such as Clutch or G2. If two of them disagree on your name or what you do, fix them first; why ChatGPT sounds unsure about your firm explains why assistants check.

#Person: markup for your authors

Person markup ties each guide to a real, checkable person, which supports the signals covered in what E-E-A-T is and why AI cares about it. It needs a name, jobTitle, worksFor and sameAs. This is the author block for David Adams on the guides he writes:

The blocks for Ashley Kenerson and Ross Williams follow the same pattern with their own jobTitle and LinkedIn profile. A founder who is a company director can also add their Companies House officer page to sameAs, which ties the person to the company on a public register.

#Article: use the right date fields

Article markup needs a headline, an author, datePublished and dateModified. There is no "last reviewed" field on Article: lastReviewed belongs to WebPage. Change dateModified only when the content changes; our guide on how old is too old for AI assistants covers why. This guide's own Article block, simplified:

#Service, not ProfessionalService

Add one Service block to each service page, with a name, a description, the provider (your Organization's @id) and where it is offered. Avoid ProfessionalService: schema.org's own page says it "was deprecated due to confusion with Service".

#Organization or LocalBusiness if clients never visit?

Use Organization. Google recommends a LocalBusiness subtype when "your site is about a local business, for example a restaurant or a physical store". A consultancy or software firm whose clients never walk in is not that. If clients do visit your offices, a specific subtype such as LegalService or AccountingService on each office page is accurate, alongside the Organization block.

#Is FAQ schema still worth adding now Google has dropped FAQ rich results?

Not for Google. Search Engine Journal reported that FAQ rich results stopped appearing on 7 May 2026, and Google removed HowTo rich results in September 2023, as its Search Central blog recorded.

FAQPage is still a valid type and existing markup does no harm, so there is no need to strip it out. This site still carries FAQPage and HowTo markup on its guides; it costs nothing to keep, and we do not expect it to earn anything in Google. The SE Ranking figures above suggest FAQ markup does nothing for ChatGPT citations either. What matters is the visible questions and answers, which give readers and assistants a clean passage to quote.

#Can schema markup get my company a Google knowledge panel?

Not on its own. Google's knowledge panel help page says panels "are automatically generated, and information that appears in a knowledge panel comes from various sources across the web". Organization markup can shape what a panel shows: Google says some properties affect "which logo is shown in Search results and your knowledge panel".

Whether you get one depends on Google recognising your firm as a distinct entity from sources beyond your site, such as registers, press and directories; a consistent sameAs list helps join those up. If a panel appears, claim it so you can suggest corrections.

#What does this look like in practice?

The clearest case we have is ID-Pal, an identity verification firm that had bought NorthRow, a UK business verification platform, and folded it into id-pal.com. "The redirect from northrow.com was technically correct, but nothing on either site told Google or the AI engines that NorthRow and ID-Pal were the same company."

The fix: "About four hours of structured data work declaring NorthRow part of ID-Pal on both domains. Live in week one, with no copy changes and no risk to the redirect." Industry pages were rebuilt alongside it, so the markup was one part of the work. The case study reports: "AI answers still recommending NorthRow as a rival: six in May, one in September 2026." And "Claude names it for UK fund administrators, on a page rebuilt in August, and links NorthRow to ID-Pal in the same answer."

#How do we add structured data without a developer?

Most firms can do a first pass in an afternoon with the tools already in their content system. You need a developer only if your site has no way to add code.

#Step 1: Check what you already have

View your homepage's source (not the inspector, which shows the page after scripts run) and search for "ld+json". Paste the address into the Schema Markup Validator at validator.schema.org, which reads every type, and Google's Rich Results Test, which shows the types that can earn a Google result.

#Step 2: Write the Organization block

Use a schema generator or copy the structure above. Check every field against the visible site, and spell every type the schema.org way.

#Step 3: Add it to the site's HTML

WordPress plugins such as Yoast, Rank Math and Schema Pro need no code; Webflow, HubSpot and most others let you paste code into a page's head. Then view the source again to confirm the block is there. If your system makes this hard, see how your website platform affects what ChatGPT can read.

#Step 4: Add Person, Article and Service markup

Give each author page a Person block, add Article markup to your five most important guides, then one Service block per service page and breadcrumbs below the homepage.

#Step 5: Test it and keep it true

Run each page through the validator and fix every error, since an invalid block is ignored. Then add a markup check to your quarterly content review: if the markup says two hundred clients and the page now says three hundred, the contradiction costs you. Our technical checklist for AI crawlers and search lists the other checks to run at the same time.

#What to do this month

1. View the source of your homepage, search for "ld+json" and paste what you find into the Schema Markup Validator.
2. Write down the ten facts you want every assistant to get right: name, company number, location, founding date, size, services and key people.
3. Correct the Organization block so it states those facts exactly as the visible page does, with every type spelled the schema.org way.
4. Add a sameAs list with your LinkedIn page, Companies House record and any accurate directory, review or professional body profiles.
5. Add Person markup to each author page and Article markup, with author, datePublished and dateModified, to your five most important guides.

#Frequently asked questions

#What schema should a B2B consultancy put on its homepage?

An Organization block with the legal and trading names, logo, description, address, company number, contact point, founders and a sameAs list of your real profiles, plus a WebSite block. Do not use ProfessionalService, which schema.org has deprecated, or LocalBusiness unless clients visit your offices. Services belong on their own pages with a Service block each.

#Which type should we add first?

Organization, on the homepage. Then Person for each author and Article for each guide, then Service for each service line. Add SoftwareApplication or Product only if you sell software or a product. FAQPage comes last, if at all, since it no longer earns a Google display.

#Will it improve our Google rankings?

Not directly; structured data is not a ranking factor in itself. Google says it helps pages qualify for rich results and helps it tell your organisation apart from others. Think of it as making sure you are read correctly, not as a way to be read first.

#Does markup added by a plugin or tag manager count?

Plugin markup that appears in the page source counts. Markup injected by a tag manager after the page loads can be read by Google, which runs JavaScript, but the main AI crawlers do not run it and may never see it. Check the page source to be sure.

#Does every page need structured data?

Every page should reference your Organization, and every guide should carry Article markup. Beyond that, add only the types that describe what the page is. Padding a page with types that do not apply makes the site less coherent to the systems reading it.

Structured data is never urgent and always worth doing, and once it is in place it makes everything else on the site easier to read. Sites built through Website That Sells include full structured data on every page; to find out first what AI assistants say about your firm today, book a free Growth Review.

#Sources